agora inbox for [email protected]
help / color / mirror / Atom feed[PATCH 7/9] Remove the special batch mode, use a larger buffer always
70+ messages / 4 participants
[nested] [flat]
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index d1eafa9700..90d17e5008 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index d1eafa9700..90d17e5008 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 853 ++++++++++++--------
1 file changed, 526 insertions(+), 327 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..fd85c18d83 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -57,11 +57,11 @@
#include "access/brin.h"
#include "access/brin_internal.h"
#include "access/brin_tuple.h"
-#include "access/hash.h" /* XXX strange that it fails because of BRIN_AM_OID without this */
#include "access/reloptions.h"
#include "access/stratnum.h"
#include "access/htup_details.h"
#include "catalog/pg_type.h"
+#include "catalog/pg_am.h"
#include "catalog/pg_amop.h"
#include "utils/array.h"
#include "utils/builtins.h"
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0008-Define-multi-minmax-oclasses-for-types-with-20210215.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0008-Define-multi-minmax-oclasses-for-types-with-20210215.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index d1eafa9700..90d17e5008 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..08d0d55b06 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index d1eafa9700..90d17e5008 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..08d0d55b06 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 853 ++++++++++++--------
1 file changed, 526 insertions(+), 327 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..fd85c18d83 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -57,11 +57,11 @@
#include "access/brin.h"
#include "access/brin_internal.h"
#include "access/brin_tuple.h"
-#include "access/hash.h" /* XXX strange that it fails because of BRIN_AM_OID without this */
#include "access/reloptions.h"
#include "access/stratnum.h"
#include "access/htup_details.h"
#include "catalog/pg_type.h"
+#include "catalog/pg_am.h"
#include "catalog/pg_amop.h"
#include "utils/array.h"
#include "utils/builtins.h"
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0008-Define-multi-minmax-oclasses-for-types-with-20210215.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0008-Define-multi-minmax-oclasses-for-types-with-20210215.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 853 ++++++++++++--------
1 file changed, 526 insertions(+), 327 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..fd85c18d83 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -57,11 +57,11 @@
#include "access/brin.h"
#include "access/brin_internal.h"
#include "access/brin_tuple.h"
-#include "access/hash.h" /* XXX strange that it fails because of BRIN_AM_OID without this */
#include "access/reloptions.h"
#include "access/stratnum.h"
#include "access/htup_details.h"
#include "catalog/pg_type.h"
+#include "catalog/pg_am.h"
#include "catalog/pg_amop.h"
#include "utils/array.h"
#include "utils/builtins.h"
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0008-Define-multi-minmax-oclasses-for-types-with-20210215.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0008-Define-multi-minmax-oclasses-for-types-with-20210215.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..08d0d55b06 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index d1eafa9700..90d17e5008 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 853 ++++++++++++--------
1 file changed, 526 insertions(+), 327 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..fd85c18d83 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -57,11 +57,11 @@
#include "access/brin.h"
#include "access/brin_internal.h"
#include "access/brin_tuple.h"
-#include "access/hash.h" /* XXX strange that it fails because of BRIN_AM_OID without this */
#include "access/reloptions.h"
#include "access/stratnum.h"
#include "access/htup_details.h"
#include "catalog/pg_type.h"
+#include "catalog/pg_am.h"
#include "catalog/pg_amop.h"
#include "utils/array.h"
#include "utils/builtins.h"
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0008-Define-multi-minmax-oclasses-for-types-with-20210215.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0008-Define-multi-minmax-oclasses-for-types-with-20210215.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..08d0d55b06 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 853 ++++++++++++--------
1 file changed, 526 insertions(+), 327 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..fd85c18d83 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -57,11 +57,11 @@
#include "access/brin.h"
#include "access/brin_internal.h"
#include "access/brin_tuple.h"
-#include "access/hash.h" /* XXX strange that it fails because of BRIN_AM_OID without this */
#include "access/reloptions.h"
#include "access/stratnum.h"
#include "access/htup_details.h"
#include "catalog/pg_type.h"
+#include "catalog/pg_am.h"
#include "catalog/pg_amop.h"
#include "utils/array.h"
#include "utils/builtins.h"
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0008-Define-multi-minmax-oclasses-for-types-with-20210215.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0008-Define-multi-minmax-oclasses-for-types-with-20210215.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 853 ++++++++++++--------
1 file changed, 526 insertions(+), 327 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..fd85c18d83 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -57,11 +57,11 @@
#include "access/brin.h"
#include "access/brin_internal.h"
#include "access/brin_tuple.h"
-#include "access/hash.h" /* XXX strange that it fails because of BRIN_AM_OID without this */
#include "access/reloptions.h"
#include "access/stratnum.h"
#include "access/htup_details.h"
#include "catalog/pg_type.h"
+#include "catalog/pg_am.h"
#include "catalog/pg_amop.h"
#include "utils/array.h"
#include "utils/builtins.h"
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0008-Define-multi-minmax-oclasses-for-types-with-20210215.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0008-Define-multi-minmax-oclasses-for-types-with-20210215.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index d1eafa9700..90d17e5008 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..08d0d55b06 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index d1eafa9700..90d17e5008 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..08d0d55b06 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 853 ++++++++++++--------
1 file changed, 526 insertions(+), 327 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..fd85c18d83 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -57,11 +57,11 @@
#include "access/brin.h"
#include "access/brin_internal.h"
#include "access/brin_tuple.h"
-#include "access/hash.h" /* XXX strange that it fails because of BRIN_AM_OID without this */
#include "access/reloptions.h"
#include "access/stratnum.h"
#include "access/htup_details.h"
#include "catalog/pg_type.h"
+#include "catalog/pg_am.h"
#include "catalog/pg_amop.h"
#include "utils/array.h"
#include "utils/builtins.h"
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0008-Define-multi-minmax-oclasses-for-types-with-20210215.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0008-Define-multi-minmax-oclasses-for-types-with-20210215.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 853 ++++++++++++--------
1 file changed, 526 insertions(+), 327 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..fd85c18d83 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -57,11 +57,11 @@
#include "access/brin.h"
#include "access/brin_internal.h"
#include "access/brin_tuple.h"
-#include "access/hash.h" /* XXX strange that it fails because of BRIN_AM_OID without this */
#include "access/reloptions.h"
#include "access/stratnum.h"
#include "access/htup_details.h"
#include "catalog/pg_type.h"
+#include "catalog/pg_am.h"
#include "catalog/pg_amop.h"
#include "utils/array.h"
#include "utils/builtins.h"
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0008-Define-multi-minmax-oclasses-for-types-with-20210215.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0008-Define-multi-minmax-oclasses-for-types-with-20210215.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index d1eafa9700..90d17e5008 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..08d0d55b06 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 853 ++++++++++++--------
1 file changed, 526 insertions(+), 327 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..fd85c18d83 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -57,11 +57,11 @@
#include "access/brin.h"
#include "access/brin_internal.h"
#include "access/brin_tuple.h"
-#include "access/hash.h" /* XXX strange that it fails because of BRIN_AM_OID without this */
#include "access/reloptions.h"
#include "access/stratnum.h"
#include "access/htup_details.h"
#include "catalog/pg_type.h"
+#include "catalog/pg_am.h"
#include "catalog/pg_amop.h"
#include "utils/array.h"
#include "utils/builtins.h"
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0008-Define-multi-minmax-oclasses-for-types-with-20210215.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0008-Define-multi-minmax-oclasses-for-types-with-20210215.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..08d0d55b06 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 853 ++++++++++++--------
1 file changed, 526 insertions(+), 327 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..fd85c18d83 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -57,11 +57,11 @@
#include "access/brin.h"
#include "access/brin_internal.h"
#include "access/brin_tuple.h"
-#include "access/hash.h" /* XXX strange that it fails because of BRIN_AM_OID without this */
#include "access/reloptions.h"
#include "access/stratnum.h"
#include "access/htup_details.h"
#include "catalog/pg_type.h"
+#include "catalog/pg_am.h"
#include "catalog/pg_amop.h"
#include "utils/array.h"
#include "utils/builtins.h"
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0008-Define-multi-minmax-oclasses-for-types-with-20210215.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0008-Define-multi-minmax-oclasses-for-types-with-20210215.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..08d0d55b06 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..08d0d55b06 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index d1eafa9700..90d17e5008 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 853 ++++++++++++--------
1 file changed, 526 insertions(+), 327 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..fd85c18d83 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -57,11 +57,11 @@
#include "access/brin.h"
#include "access/brin_internal.h"
#include "access/brin_tuple.h"
-#include "access/hash.h" /* XXX strange that it fails because of BRIN_AM_OID without this */
#include "access/reloptions.h"
#include "access/stratnum.h"
#include "access/htup_details.h"
#include "catalog/pg_type.h"
+#include "catalog/pg_am.h"
#include "catalog/pg_amop.h"
#include "utils/array.h"
#include "utils/builtins.h"
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0008-Define-multi-minmax-oclasses-for-types-with-20210215.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0008-Define-multi-minmax-oclasses-for-types-with-20210215.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 853 ++++++++++++--------
1 file changed, 526 insertions(+), 327 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..fd85c18d83 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -57,11 +57,11 @@
#include "access/brin.h"
#include "access/brin_internal.h"
#include "access/brin_tuple.h"
-#include "access/hash.h" /* XXX strange that it fails because of BRIN_AM_OID without this */
#include "access/reloptions.h"
#include "access/stratnum.h"
#include "access/htup_details.h"
#include "catalog/pg_type.h"
+#include "catalog/pg_am.h"
#include "catalog/pg_amop.h"
#include "utils/array.h"
#include "utils/builtins.h"
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0008-Define-multi-minmax-oclasses-for-types-with-20210215.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0008-Define-multi-minmax-oclasses-for-types-with-20210215.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index d1eafa9700..90d17e5008 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..08d0d55b06 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index d1eafa9700..90d17e5008 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 853 ++++++++++++--------
1 file changed, 526 insertions(+), 327 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..fd85c18d83 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -57,11 +57,11 @@
#include "access/brin.h"
#include "access/brin_internal.h"
#include "access/brin_tuple.h"
-#include "access/hash.h" /* XXX strange that it fails because of BRIN_AM_OID without this */
#include "access/reloptions.h"
#include "access/stratnum.h"
#include "access/htup_details.h"
#include "catalog/pg_type.h"
+#include "catalog/pg_am.h"
#include "catalog/pg_amop.h"
#include "utils/array.h"
#include "utils/builtins.h"
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0008-Define-multi-minmax-oclasses-for-types-with-20210215.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0008-Define-multi-minmax-oclasses-for-types-with-20210215.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..08d0d55b06 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 853 ++++++++++++--------
1 file changed, 526 insertions(+), 327 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..fd85c18d83 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -57,11 +57,11 @@
#include "access/brin.h"
#include "access/brin_internal.h"
#include "access/brin_tuple.h"
-#include "access/hash.h" /* XXX strange that it fails because of BRIN_AM_OID without this */
#include "access/reloptions.h"
#include "access/stratnum.h"
#include "access/htup_details.h"
#include "catalog/pg_type.h"
+#include "catalog/pg_am.h"
#include "catalog/pg_amop.h"
#include "utils/array.h"
#include "utils/builtins.h"
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0008-Define-multi-minmax-oclasses-for-types-with-20210215.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0008-Define-multi-minmax-oclasses-for-types-with-20210215.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index d1eafa9700..90d17e5008 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index d1eafa9700..90d17e5008 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 853 ++++++++++++--------
1 file changed, 526 insertions(+), 327 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..fd85c18d83 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -57,11 +57,11 @@
#include "access/brin.h"
#include "access/brin_internal.h"
#include "access/brin_tuple.h"
-#include "access/hash.h" /* XXX strange that it fails because of BRIN_AM_OID without this */
#include "access/reloptions.h"
#include "access/stratnum.h"
#include "access/htup_details.h"
#include "catalog/pg_type.h"
+#include "catalog/pg_am.h"
#include "catalog/pg_amop.h"
#include "utils/array.h"
#include "utils/builtins.h"
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0008-Define-multi-minmax-oclasses-for-types-with-20210215.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0008-Define-multi-minmax-oclasses-for-types-with-20210215.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..08d0d55b06 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..08d0d55b06 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..08d0d55b06 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index d1eafa9700..90d17e5008 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index d1eafa9700..90d17e5008 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index d1eafa9700..90d17e5008 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..08d0d55b06 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index d1eafa9700..90d17e5008 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..08d0d55b06 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index d1eafa9700..90d17e5008 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index d1eafa9700..90d17e5008 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 853 ++++++++++++--------
1 file changed, 526 insertions(+), 327 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..fd85c18d83 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -57,11 +57,11 @@
#include "access/brin.h"
#include "access/brin_internal.h"
#include "access/brin_tuple.h"
-#include "access/hash.h" /* XXX strange that it fails because of BRIN_AM_OID without this */
#include "access/reloptions.h"
#include "access/stratnum.h"
#include "access/htup_details.h"
#include "catalog/pg_type.h"
+#include "catalog/pg_am.h"
#include "catalog/pg_amop.h"
#include "utils/array.h"
#include "utils/builtins.h"
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0008-Define-multi-minmax-oclasses-for-types-with-20210215.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0008-Define-multi-minmax-oclasses-for-types-with-20210215.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 853 ++++++++++++--------
1 file changed, 526 insertions(+), 327 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..fd85c18d83 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -57,11 +57,11 @@
#include "access/brin.h"
#include "access/brin_internal.h"
#include "access/brin_tuple.h"
-#include "access/hash.h" /* XXX strange that it fails because of BRIN_AM_OID without this */
#include "access/reloptions.h"
#include "access/stratnum.h"
#include "access/htup_details.h"
#include "catalog/pg_type.h"
+#include "catalog/pg_am.h"
#include "catalog/pg_amop.h"
#include "utils/array.h"
#include "utils/builtins.h"
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0008-Define-multi-minmax-oclasses-for-types-with-20210215.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0008-Define-multi-minmax-oclasses-for-types-with-20210215.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index d1eafa9700..90d17e5008 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..08d0d55b06 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..08d0d55b06 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..08d0d55b06 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index d1eafa9700..90d17e5008 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..08d0d55b06 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 853 ++++++++++++--------
1 file changed, 526 insertions(+), 327 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..fd85c18d83 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -57,11 +57,11 @@
#include "access/brin.h"
#include "access/brin_internal.h"
#include "access/brin_tuple.h"
-#include "access/hash.h" /* XXX strange that it fails because of BRIN_AM_OID without this */
#include "access/reloptions.h"
#include "access/stratnum.h"
#include "access/htup_details.h"
#include "catalog/pg_type.h"
+#include "catalog/pg_am.h"
#include "catalog/pg_amop.h"
#include "utils/array.h"
#include "utils/builtins.h"
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0008-Define-multi-minmax-oclasses-for-types-with-20210215.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0008-Define-multi-minmax-oclasses-for-types-with-20210215.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..08d0d55b06 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 853 ++++++++++++--------
1 file changed, 526 insertions(+), 327 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..fd85c18d83 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -57,11 +57,11 @@
#include "access/brin.h"
#include "access/brin_internal.h"
#include "access/brin_tuple.h"
-#include "access/hash.h" /* XXX strange that it fails because of BRIN_AM_OID without this */
#include "access/reloptions.h"
#include "access/stratnum.h"
#include "access/htup_details.h"
#include "catalog/pg_type.h"
+#include "catalog/pg_am.h"
#include "catalog/pg_amop.h"
#include "utils/array.h"
#include "utils/builtins.h"
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0008-Define-multi-minmax-oclasses-for-types-with-20210215.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0008-Define-multi-minmax-oclasses-for-types-with-20210215.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 853 ++++++++++++--------
1 file changed, 526 insertions(+), 327 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..fd85c18d83 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -57,11 +57,11 @@
#include "access/brin.h"
#include "access/brin_internal.h"
#include "access/brin_tuple.h"
-#include "access/hash.h" /* XXX strange that it fails because of BRIN_AM_OID without this */
#include "access/reloptions.h"
#include "access/stratnum.h"
#include "access/htup_details.h"
#include "catalog/pg_type.h"
+#include "catalog/pg_am.h"
#include "catalog/pg_amop.h"
#include "utils/array.h"
#include "utils/builtins.h"
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0008-Define-multi-minmax-oclasses-for-types-with-20210215.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0008-Define-multi-minmax-oclasses-for-types-with-20210215.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 853 ++++++++++++--------
1 file changed, 526 insertions(+), 327 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..fd85c18d83 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -57,11 +57,11 @@
#include "access/brin.h"
#include "access/brin_internal.h"
#include "access/brin_tuple.h"
-#include "access/hash.h" /* XXX strange that it fails because of BRIN_AM_OID without this */
#include "access/reloptions.h"
#include "access/stratnum.h"
#include "access/htup_details.h"
#include "catalog/pg_type.h"
+#include "catalog/pg_am.h"
#include "catalog/pg_amop.h"
#include "utils/array.h"
#include "utils/builtins.h"
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0008-Define-multi-minmax-oclasses-for-types-with-20210215.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0008-Define-multi-minmax-oclasses-for-types-with-20210215.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 851 ++++++++++++--------
1 file changed, 525 insertions(+), 326 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index d1eafa9700..90d17e5008 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0006-Batch-mode-when-building-new-BRIN-multi-min-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* [PATCH 7/9] Remove the special batch mode, use a larger buffer always
@ 2021-02-02 00:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Tomas Vondra @ 2021-02-02 00:57 UTC (permalink / raw)
Instead of using a batch mode (with a larger input buffer) only for new
ranges, which introduces "special cases" in various places, use it as
the standard approach.
Also, instead of sizing the buffer to cover the whole range, limit it
to some reasonable limit (10x the user-specified size). That should give
us most of the benefits without consuming a lot of memory.
---
src/backend/access/brin/brin_minmax_multi.c | 853 ++++++++++++--------
1 file changed, 526 insertions(+), 327 deletions(-)
diff --git a/src/backend/access/brin/brin_minmax_multi.c b/src/backend/access/brin/brin_minmax_multi.c
index 69a72da337..fd85c18d83 100644
--- a/src/backend/access/brin/brin_minmax_multi.c
+++ b/src/backend/access/brin/brin_minmax_multi.c
@@ -57,11 +57,11 @@
#include "access/brin.h"
#include "access/brin_internal.h"
#include "access/brin_tuple.h"
-#include "access/hash.h" /* XXX strange that it fails because of BRIN_AM_OID without this */
#include "access/reloptions.h"
#include "access/stratnum.h"
#include "access/htup_details.h"
#include "catalog/pg_type.h"
+#include "catalog/pg_am.h"
#include "catalog/pg_amop.h"
#include "utils/array.h"
#include "utils/builtins.h"
@@ -92,7 +92,15 @@
*/
#define PROCNUM_BASE 11
-#define MINMAX_LOAD_FACTOR 0.75
+/*
+ * Sizing the insert buffer - we use 10x the number of values specified
+ * in the reloption, but we cap it to 8192 not to get too large. When
+ * the buffer gets full, we reduce the number of values by half.
+ */
+#define MINMAX_BUFFER_FACTOR 10
+#define MINMAX_BUFFER_MIN 256
+#define MINMAX_BUFFER_MAX 8192
+#define MINMAX_BUFFER_LOAD_FACTOR 0.5
typedef struct MinmaxMultiOpaque
{
@@ -155,23 +163,24 @@ typedef struct Ranges
Oid typid;
Oid colloid;
AttrNumber attno;
+ FmgrInfo *cmp;
/* (2*nranges + nvalues) <= maxvalues */
int nranges; /* number of ranges in the array (stored) */
+ int nsorted; /* number of sorted values (ranges + points) */
int nvalues; /* number of values in the data array (all) */
int maxvalues; /* maximum number of values (reloption) */
/*
- * In batch mode, we simply add the values into a buffer, without any
- * expensive steps (sorting, deduplication, ...). The buffer is sized
- * to be larger than the target number of values per range, which
- * reduces the number of compactions - operating on larger buffers is
- * significantly more efficient, in most cases. We keep the actual
- * target and compact to the requested number of values at the very
- * end, before serializing to on-disk representation.
+ * We simply add the values into a large buffer, without any expensive
+ * steps (sorting, deduplication, ...). The buffer is a multiple of
+ * the target number of values, so the compaction happen less often,
+ * amortizing the costs. We keep the actual target and compact to
+ * the requested number of values at the very end, before serializing
+ * to on-disk representation.
*/
- bool batch_mode;
- int target_maxvalues; /* requested number of values */
+ /* requested number of values */
+ int target_maxvalues;
/* values stored for this range - either raw values, or ranges */
Datum values[FLEXIBLE_ARRAY_MEMBER];
@@ -203,7 +212,7 @@ typedef struct SerializedRanges
static SerializedRanges *range_serialize(Ranges *range);
-static Ranges *range_deserialize(SerializedRanges *range);
+static Ranges *range_deserialize(int maxvalues, SerializedRanges *range);
/* Cache for support and strategy procesures. */
@@ -213,6 +222,14 @@ static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
uint16 attno, Oid subtype, uint16 strategynum);
+typedef struct compare_context
+{
+ FmgrInfo *cmpFn;
+ Oid colloid;
+} compare_context;
+
+static int compare_values(const void *a, const void *b, void *arg);
+
/*
* minmax_multi_init
@@ -240,6 +257,57 @@ minmax_multi_init(int maxvalues)
return ranges;
}
+static void
+AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid);
+
+
+static void
+range_deduplicate_values(Ranges *range)
+{
+ int i, n;
+ int start;
+ compare_context cxt;
+
+ /*
+ * If there are no unsorted values, we're done (this probably can't
+ * happen, as we're adding values to unsorted part).
+ */
+ if (range->nsorted == range->nvalues)
+ return;
+
+ /* sort the values */
+ cxt.colloid = range->colloid;
+ cxt.cmpFn = range->cmp;
+
+ /* how many values to sort? */
+ start = 2 * range->nranges;
+
+ qsort_arg(&range->values[start],
+ range->nvalues, sizeof(Datum),
+ compare_values, (void *) &cxt);
+
+ n = 1;
+ for (i = 1; i < range->nvalues; i++)
+ {
+ /* same as preceding value, so store it */
+ if (compare_values(&range->values[start + i - 1],
+ &range->values[start + i],
+ (void *) &cxt) == 0)
+ continue;
+
+ range->values[start + n] = range->values[start + i];
+
+ n++;
+ }
+
+ /* now all the values are sorted */
+ range->nvalues = n;
+ range->nsorted = n;
+
+ AssertCheckRanges(range, range->cmp, range->colloid);
+}
+
+
/*
* range_serialize
* Serialize the in-memory representation into a compact varlena value.
@@ -262,14 +330,25 @@ range_serialize(Ranges *range)
/* simple sanity checks */
Assert(range->nranges >= 0);
+ Assert(range->nsorted >= 0);
Assert(range->nvalues >= 0);
Assert(range->maxvalues > 0);
+ Assert(range->target_maxvalues > 0);
+
+ /* at this point the range should be compacted to the target size */
+ Assert(2*range->nranges + range->nvalues <= range->target_maxvalues);
+
+ Assert(range->target_maxvalues <= range->maxvalues);
+
+ /* range boundaries are always sorted */
+ Assert(range->nvalues >= range->nsorted);
+
+ /* sort and deduplicate values, if there's unsorted part */
+ range_deduplicate_values(range);
/* see how many Datum values we actually have */
nvalues = 2*range->nranges + range->nvalues;
- Assert(2*range->nranges + range->nvalues <= range->maxvalues);
-
typid = range->typid;
typbyval = get_typbyval(typid);
typlen = get_typlen(typid);
@@ -316,7 +395,7 @@ range_serialize(Ranges *range)
serialized->typid = typid;
serialized->nranges = range->nranges;
serialized->nvalues = range->nvalues;
- serialized->maxvalues = range->maxvalues;
+ serialized->maxvalues = range->target_maxvalues;
/*
* And now copy also the boundary values (like the length calculation
@@ -367,7 +446,7 @@ range_serialize(Ranges *range)
* in the in-memory value array.
*/
static Ranges *
-range_deserialize(SerializedRanges *serialized)
+range_deserialize(int maxvalues, SerializedRanges *serialized)
{
int i,
nvalues;
@@ -384,15 +463,18 @@ range_deserialize(SerializedRanges *serialized)
nvalues = 2*serialized->nranges + serialized->nvalues;
Assert(nvalues <= serialized->maxvalues);
+ Assert(serialized->maxvalues <= maxvalues);
- range = minmax_multi_init(serialized->maxvalues);
+ range = minmax_multi_init(maxvalues);
/* copy the header info */
range->nranges = serialized->nranges;
range->nvalues = serialized->nvalues;
- range->maxvalues = serialized->maxvalues;
+ range->nsorted = serialized->nvalues;
+ range->maxvalues = maxvalues;
+ range->target_maxvalues = serialized->maxvalues;
+
range->typid = serialized->typid;
- range->batch_mode = false;
typbyval = get_typbyval(serialized->typid);
typlen = get_typlen(serialized->typid);
@@ -439,12 +521,6 @@ range_deserialize(SerializedRanges *serialized)
return range;
}
-typedef struct compare_context
-{
- FmgrInfo *cmpFn;
- Oid colloid;
-} compare_context;
-
/*
* Used to represent ranges expanded during merging and combining (to
* reduce number of boundary values to store).
@@ -528,6 +604,115 @@ compare_values(const void *a, const void *b, void *arg)
return 0;
}
+void *bsearch_arg(const void *key, const void *base,
+ size_t nmemb, size_t size,
+ int (*compar) (const void *, const void *, void *),
+ void *arg);
+
+static bool
+has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
+ Datum newval, AttrNumber attno, Oid typid)
+{
+ Datum compar;
+
+ Datum minvalue = ranges->values[0];
+ Datum maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ FmgrInfo *cmpLessFn;
+ FmgrInfo *cmpGreaterFn;
+
+ /* binary search on ranges */
+ int start,
+ end;
+
+ if (ranges->nranges == 0)
+ return false;
+
+ /*
+ * Otherwise, need to compare the new value with boundaries of all
+ * the ranges. First check if it's less than the absolute minimum,
+ * which is the first value in the array.
+ */
+ cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTLessStrategyNumber);
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in the range list */
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * And now compare it to the existing maximum (last value in the
+ * data array). But only if we haven't already ruled out a possible
+ * match in the minvalue check.
+ */
+ cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
+ BTGreaterStrategyNumber);
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ if (DatumGetBool(compar))
+ return false;
+
+ /*
+ * So we know it's in the general min/max, the question is whether it
+ * falls in one of the ranges or gaps. We'll use a binary search on
+ * the ranges.
+ *
+ * it's in the general range, but is it actually covered by any
+ * of the ranges? Repeat the check for each range.
+ *
+ * XXX We simply walk the ranges sequentially, but maybe we could
+ * further leverage the ordering and non-overlap and use bsearch to
+ * speed this up a bit.
+ */
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
+ {
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ return false;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
+
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
+ continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
+
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
+ continue;
+ }
+
+ /* hey, we found a matching range */
+ return true;
+ }
+
+ return false;
+}
+
+
/*
* range_contains_value
* See if the new value is already contained in the range list.
@@ -552,8 +737,6 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
Ranges *ranges, Datum newval)
{
int i;
- FmgrInfo *cmpLessFn;
- FmgrInfo *cmpGreaterFn;
FmgrInfo *cmpEqualFn;
Oid typid = attr->atttypid;
@@ -562,77 +745,8 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
* range, and only when there's still a chance of getting a match we
* inspect the individual ranges.
*/
- if (ranges->nranges > 0)
- {
- Datum compar;
- bool match = true;
-
- Datum minvalue = ranges->values[0];
- Datum maxvalue = ranges->values[2*ranges->nranges - 1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTLessStrategyNumber);
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in the range list */
- if (DatumGetBool(compar))
- match = false;
-
- /*
- * And now compare it to the existing maximum (last value in the
- * data array). But only if we haven't already ruled out a possible
- * match in the minvalue check.
- */
- if (match)
- {
- cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
- BTGreaterStrategyNumber);
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- if (DatumGetBool(compar))
- match = false;
- }
-
- /*
- * So it's in the general range, but is it actually covered by any
- * of the ranges? Repeat the check for each range.
- *
- * XXX We simply walk the ranges sequentially, but maybe we could
- * further leverage the ordering and non-overlap and use bsearch to
- * speed this up a bit.
- */
- for (i = 0; i < ranges->nranges && match; i++)
- {
- /* copy the min/max values from the ranges */
- minvalue = ranges->values[2*i];
- maxvalue = ranges->values[2*i+1];
-
- /*
- * Otherwise, need to compare the new value with boundaries of all
- * the ranges. First check if it's less than the absolute minimum,
- * which is the first value in the array.
- */
- compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
-
- /* smaller than the smallest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
-
- /* larger than the largest value in this range */
- if (DatumGetBool(compar))
- continue;
-
- /* hey, we found a matching row */
- return true;
- }
- }
+ if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
+ return true;
cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
BTEqualStrategyNumber);
@@ -640,92 +754,42 @@ range_contains_value(BrinDesc *bdesc, Oid colloid,
/*
* We're done with the ranges, now let's inspect the exact values.
*
- * XXX Again, we do sequentially search the values - consider leveraging
- * the ordering of values to improve performance.
+ * XXX We do sequential search for small number of values, and bsearch
+ * once we have more than 16 values.
+ *
+ * XXX We only inspect the sorted part - that's OK. For building it may
+ * produce false negatives, but only after we already added some values
+ * to the unsorted part, so we've modified the value. And when querying
+ * the index, there should be no unsorted values.
*/
- for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nvalues; i++)
+ if (ranges->nsorted >= 16)
{
- Datum compar;
+ compare_context cxt;
- compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
- /* found an exact match */
- if (DatumGetBool(compar))
+ if (bsearch_arg(&newval, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) != NULL)
return true;
}
-
- /* the value is not covered by this BRIN tuple */
- return false;
-}
-
-/*
- * insert_value
- * Adds a new value into the single-point part, while maintaining ordering.
- *
- * The function inserts the new value to the right place in the single-point
- * part of the range. It assumes there's enough free space, and then does
- * essentially an insert-sort.
- *
- * XXX Assumes the 'values' array has space for (nvalues+1) entries, and that
- * only the first nvalues are used.
- */
-static void
-insert_value(FmgrInfo *cmp, Oid colloid, Datum *values, int nvalues,
- Datum newvalue)
-{
- int i;
- Datum lt;
-
- /* If there are no values yet, store the new one and we're done. */
- if (!nvalues)
+ else
{
- values[0] = newvalue;
- return;
- }
-
- /*
- * A common case is that the new value is entirely out of the existing
- * range, i.e. it's either smaller or larger than all previous values.
- * So we check and handle this case first - first we check the larger
- * case, because in that case we can just append the value to the end
- * of the array and we're done.
- */
+ for (i = 2*ranges->nranges; i < 2*ranges->nranges + ranges->nsorted; i++)
+ {
+ Datum compar;
- /* Is it greater than all existing values in the array? */
- lt = FunctionCall2Coll(cmp, colloid, values[nvalues-1], newvalue);
- if (DatumGetBool(lt))
- {
- /* just copy it in-place and we're done */
- values[nvalues] = newvalue;
- return;
- }
+ compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
- /*
- * OK, I lied a bit - we won't check the smaller case explicitly, but
- * we'll just compare the value to all existing values in the array.
- * But we happen to start with the smallest value, so we're actually
- * doing the check anyway.
- *
- * XXX We do walk the values sequentially. Perhaps we could/should be
- * smarter and do some sort of bisection, to improve performance?
- */
- for (i = 0; i < nvalues; i++)
- {
- lt = FunctionCall2Coll(cmp, colloid, newvalue, values[i]);
- if (DatumGetBool(lt))
- {
- /*
- * Move values to make space for the new entry, which should go
- * to index 'i'. Entries 0 ... (i-1) should stay where they are.
- */
- memmove(&values[i+1], &values[i], (nvalues-i) * sizeof(Datum));
- values[i] = newvalue;
- return;
+ /* found an exact match */
+ if (DatumGetBool(compar))
+ return true;
}
}
- /* We should never really get here. */
- Assert(false);
+ /* the value is not covered by this BRIN tuple */
+ return false;
}
#ifdef USE_ASSERT_CHECKING
@@ -754,11 +818,12 @@ static void
AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
{
#ifdef USE_ASSERT_CHECKING
- int i, j;
+ int i;
/* some basic sanity checks */
Assert(ranges->nranges >= 0);
- Assert(ranges->nvalues >= 0);
+ Assert(ranges->nsorted >= 0);
+ Assert(ranges->nvalues >= ranges->nsorted);
Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
Assert(ranges->typid != InvalidOid);
@@ -770,32 +835,111 @@ AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
*/
AssertArrayOrder(cmpFn, colloid, ranges->values, 2*ranges->nranges);
- /* finally check that none of the values are not covered by ranges */
+ /* then the single-point ranges (with nvalues boundar values ) */
+ AssertArrayOrder(cmpFn, colloid, &ranges->values[2*ranges->nranges],
+ ranges->nsorted);
+
+ /*
+ * Check that none of the values are not covered by ranges (both
+ * sorted and unsorted)
+ */
for (i = 0; i < ranges->nvalues; i++)
{
+ Datum compar;
+ int start,
+ end;
+ Datum minvalue,
+ maxvalue;
+
Datum value = ranges->values[2 * ranges->nranges + i];
- for (j = 0; j < ranges->nranges; j++)
+ if (ranges->nranges == 0)
+ break;
+
+ minvalue = ranges->values[0];
+ maxvalue = ranges->values[2*ranges->nranges - 1];
+
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
+
+ /* smaller than the smallest value in the first range */
+ if (DatumGetBool(compar))
+ continue;
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
+
+ /* larger than the largest value in the last range */
+ if (DatumGetBool(compar))
+ continue;
+
+ start = 0; /* first range */
+ end = ranges->nranges - 1; /* last range */
+ while (true)
{
- Datum r;
+ int midpoint = (start + end) / 2;
+
+ /* this means we ran out of ranges in the last step */
+ if (start > end)
+ break;
+
+ /* copy the min/max values from the ranges */
+ minvalue = ranges->values[2 * midpoint];
+ maxvalue = ranges->values[2 * midpoint + 1];
- Datum minval = ranges->values[2 * j];
- Datum maxval = ranges->values[2 * j + 1];
+ /*
+ * Is the value smaller than the minval? If yes, we'll recurse
+ * to the left side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
- /* if value is smaller than range minimum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, value, minval);
- if (DatumGetBool(r))
+ /* smaller than the smallest value in this range */
+ if (DatumGetBool(compar))
+ {
+ end = (midpoint - 1);
continue;
+ }
+
+ /*
+ * Is the value greater than the minval? If yes, we'll recurse
+ * to the right side of range array.
+ */
+ compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
- /* if value is greater than range maximum, that's OK */
- r = FunctionCall2Coll(cmpFn, colloid, maxval, value);
- if (DatumGetBool(r))
+ /* larger than the largest value in this range */
+ if (DatumGetBool(compar))
+ {
+ start = (midpoint + 1);
continue;
+ }
- /* value is between [min,max], which is wrong */
+ /* hey, we found a matching range */
Assert(false);
}
}
+
+ /* and values in the unsorted part must not be in sorted part */
+ for (i = ranges->nsorted; i < ranges->nvalues; i++)
+ {
+ compare_context cxt;
+ Datum value = ranges->values[2 * ranges->nranges + i];
+
+ if (ranges->nsorted == 0)
+ break;
+
+ cxt.colloid = ranges->colloid;
+ cxt.cmpFn = ranges->cmp;
+
+ Assert(bsearch_arg(&value, &ranges->values[2*ranges->nranges],
+ ranges->nsorted, sizeof(Datum),
+ compare_values, (void *) &cxt) == NULL);
+ }
#endif
}
@@ -1106,8 +1250,7 @@ build_distances(FmgrInfo *distanceFn, Oid colloid,
*/
static CombineRange *
build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
- bool addvalue, Datum newvalue, int *nranges,
- bool deduplicate)
+ int *nranges)
{
int ncranges;
CombineRange *cranges;
@@ -1115,28 +1258,15 @@ build_combine_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
/* now do the actual merge sort */
ncranges = ranges->nranges + ranges->nvalues;
- /* should we add an extra value? */
- if (addvalue)
- ncranges += 1;
-
cranges = (CombineRange *) palloc0(ncranges * sizeof(CombineRange));
- /* put the new value at the beginning */
- if (addvalue)
- {
- cranges[0].minval = newvalue;
- cranges[0].maxval = newvalue;
- cranges[0].collapsed = true;
-
- /* then the regular and collapsed ranges */
- fill_combine_ranges(&cranges[1], ncranges-1, ranges);
- }
- else
- fill_combine_ranges(cranges, ncranges, ranges);
+ /* fll the combine ranges */
+ fill_combine_ranges(cranges, ncranges, ranges);
/* and sort the ranges */
- ncranges = sort_combine_ranges(cmp, colloid, cranges, ncranges,
- deduplicate);
+ ncranges = sort_combine_ranges(cmp, colloid,
+ cranges, ncranges,
+ true); /* deduplicate */
/* remember how many cranges we built */
*nranges = ncranges;
@@ -1321,19 +1451,28 @@ store_combine_ranges(Ranges *ranges, CombineRange *cranges, int ncranges)
}
}
+ /* all the values are sorted */
+ ranges->nsorted = ranges->nvalues;
+
Assert(count_values(cranges, ncranges) == 2*ranges->nranges + ranges->nvalues);
Assert(2*ranges->nranges + ranges->nvalues <= ranges->maxvalues);
}
+
+
/*
- * range_add_value
- * Add the new value to the multi-minmax range.
+ * Consider freeing space in the ranges.
+ *
+ * Returns true if the value was actually modified.
*/
static bool
-range_add_value(BrinDesc *bdesc, Oid colloid,
- AttrNumber attno, Form_pg_attribute attr,
- Ranges *ranges, Datum newval)
+ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *range)
{
+ MemoryContext ctx;
+ MemoryContext oldctx;
+
FmgrInfo *cmpFn,
*distanceFn;
@@ -1342,109 +1481,44 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
int ncranges;
DistanceValue *distances;
- MemoryContext ctx;
- MemoryContext oldctx;
-
- /* we'll certainly need the comparator, so just look it up now */
- cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
- BTLessStrategyNumber);
-
- /* comprehensive checks of the input ranges */
- AssertCheckRanges(ranges, cmpFn, colloid);
-
- Assert((ranges->nranges >= 0) && (ranges->nvalues >= 0) && (ranges->maxvalues >= 0));
-
/*
- * When batch-building, there should be no ranges. So either the
- * number of ranges is 0 or we're not in batching mode.
+ * If there is free space in the buffer, we're done without having
+ * to modify anything.
*/
- Assert(!ranges->batch_mode || (ranges->nranges == 0));
-
- /* When batch-building, just add it and we're done. */
- if (ranges->batch_mode)
- {
- /* there has to be free space, if we've sized the struct */
- Assert(ranges->nvalues < ranges->maxvalues);
-
- /* Make a copy of the value, if needed. */
- ranges->values[ranges->nvalues++]
- = datumCopy(newval, attr->attbyval, attr->attlen);;
-
- return true;
- }
-
- /*
- * Bail out if the value already is covered by the range.
- *
- * We could also add values until we hit values_per_range, and then
- * do the deduplication in a batch, hoping for better efficiency. But
- * that would mean we actually modify the range every time, which means
- * having to serialize the value, which does palloc, walks the values,
- * copies them, etc. Not exactly cheap.
- *
- * So instead we do the check, which should be fairly cheap - assuming
- * the comparator function is not very expensive.
- *
- * This also implies means the values array can't contain duplicities.
- */
- if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ if (2*range->nranges + range->nvalues < range->maxvalues)
return false;
- /* Make a copy of the value, if needed. */
- newval = datumCopy(newval, attr->attbyval, attr->attlen);
-
- /*
- * If there's space in the values array, copy it in and we're done.
- *
- * We do want to keep the values sorted (to speed up searches), so we
- * do a simple insertion sort. We could do something more elaborate,
- * e.g. by sorting the values only now and then, but for small counts
- * (e.g. when maxvalues is 64) this should be fine.
- */
- if (2*ranges->nranges + ranges->nvalues < ranges->maxvalues)
- {
- Datum *values;
-
- /* beginning of the 'single value' part (for convenience) */
- values = &ranges->values[2*ranges->nranges];
-
- insert_value(cmpFn, colloid, values, ranges->nvalues, newval);
-
- ranges->nvalues++;
-
- /*
- * Check we haven't broken the ordering of boundary values (checks
- * both parts, but that doesn't hurt).
- */
- AssertCheckRanges(ranges, cmpFn, colloid);
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
- /* Also check the range contains the value we just added. */
- // FIXME Assert(ranges, cmpFn, colloid);
+ /* Try deduplicating values in the unsorted part */
+ range_deduplicate_values(range);
- /* yep, we've modified the range */
+ /* did we reduce enough free space by just the deduplication? */
+ if (2*range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
return true;
- }
/*
- * Damn - the new value is not in the range yet, but we don't have space
- * to just insert it. So we need to combine some of the existing ranges,
- * to reduce the number of values we need to store (joining two intervals
- * reduces the number of boundaries to store by 2).
+ * we need to combine some of the existing ranges, to reduce the number
+ * of values we need to store (joining intervals reduces the number of
+ * boundary values).
*
- * To do that we first construct an array of CombineRange items - each
- * combine range tracks if it's a regular range or collapsed range, where
- * "collapsed" means "single point."
+ * We first construct an array of CombineRange items - each combine range
+ * tracks if it's a regular range or a collapsed range, where "collapsed"
+ * means "single point." This makes the processing easier, as it allows
+ * handling ranges and points the same way.
*
- * Existing ranges (we have ranges->nranges of them) map to combine ranges
- * directly, while single points (ranges->nvalues of them) have to be
- * expanded. We neet the combine ranges to be sorted, and we do that by
- * performing a merge sort of ranges, values and new value.
+ * Then we sort the combine ranges - this is necessary, because although
+ * ranges and points were sorted on their own, the new array is not. We
+ * do that by performing a merge sort of the two parts.
*
* The distanceFn calls (which may internally call e.g. numeric_le) may
- * allocate quite a bit of memory, and we must not leak it. Otherwise
- * we'd have problems e.g. when building indexes. So we create a local
- * memory context and make sure we free the memory before leaving this
- * function (not after every call).
+ * allocate quite a bit of memory, and we must not leak it (we might have
+ * to do this repeatedly, even for a single BRIN page range). Otherwise
+ * we'd have problems e.g. when building new indexes. So we use a memory
+ * context and make sure we free the memory at the end (so if we call
+ * the distance function many times, it might be an issue, but meh).
*/
ctx = AllocSetContextCreate(CurrentMemoryContext,
"minmax-multi context",
@@ -1453,9 +1527,7 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
oldctx = MemoryContextSwitchTo(ctx);
/* OK build the combine ranges */
- cranges = build_combine_ranges(cmpFn, colloid, ranges,
- true, newval, &ncranges,
- false);
+ cranges = build_combine_ranges(cmpFn, colloid, range, &ncranges);
/* and we'll also need the 'distance' procedure */
distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
@@ -1469,21 +1541,104 @@ range_add_value(BrinDesc *bdesc, Oid colloid,
* use too low or high value.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- ranges->maxvalues * MINMAX_LOAD_FACTOR,
+ range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
cmpFn, colloid);
- Assert(count_values(cranges, ncranges) <= ranges->maxvalues * MINMAX_LOAD_FACTOR);
+ Assert(count_values(cranges, ncranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
/* decompose the combine ranges into regular ranges and single values */
- store_combine_ranges(ranges, cranges, ncranges);
+ store_combine_ranges(range, cranges, ncranges);
MemoryContextSwitchTo(oldctx);
MemoryContextDelete(ctx);
/* Did we break the ranges somehow? */
+ AssertCheckRanges(range, cmpFn, colloid);
+
+ return true;
+}
+
+/*
+ * range_add_value
+ * Add the new value to the multi-minmax range.
+ */
+static bool
+range_add_value(BrinDesc *bdesc, Oid colloid,
+ AttrNumber attno, Form_pg_attribute attr,
+ Ranges *ranges, Datum newval)
+{
+ FmgrInfo *cmpFn;
+ bool modified = false;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
+
+ /* comprehensive checks of the input ranges */
AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /*
+ * Make sure there's enough free space in the buffer. We only trigger
+ * this when the buffer is full, which means it had to be modified as
+ * we size it to be larger than what is stored on disk.
+ *
+ * XXX This needs to happen before we check if the value is contained
+ * in the range, because the value might be in the unsorted part, and
+ * we don't check that in range_contains_value. The deduplication would
+ * then move it to the sorted part, and we'd add the value too, which
+ * violates the rule that we never have duplicates with the ranges
+ * or sorted values.
+ *
+ * XXX At the moment this only does the deduplication.
+ *
+ * XXX We might also deduplicate and recheck if the value is contained,
+ * but that seems like an overkill. We'd need to deduplicate anyway,
+ * so why not do it now.
+ */
+ modified = ensure_free_space_in_buffer(bdesc, colloid,
+ attno, attr, ranges);
+
+ /*
+ * Bail out if the value already is covered by the range.
+ *
+ * We could also add values until we hit values_per_range, and then
+ * do the deduplication in a batch, hoping for better efficiency. But
+ * that would mean we actually modify the range every time, which means
+ * having to serialize the value, which does palloc, walks the values,
+ * copies them, etc. Not exactly cheap.
+ *
+ * So instead we do the check, which should be fairly cheap - assuming
+ * the comparator function is not very expensive.
+ *
+ * This also implies means the values array can't contain duplicities.
+ */
+ if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval))
+ return modified;
+
+ /* Make a copy of the value, if needed. */
+ newval = datumCopy(newval, attr->attbyval, attr->attlen);
+
+ /*
+ * If there's space in the values array, copy it in and we're done.
+ *
+ * We do want to keep the values sorted (to speed up searches), so we
+ * do a simple insertion sort. We could do something more elaborate,
+ * e.g. by sorting the values only now and then, but for small counts
+ * (e.g. when maxvalues is 64) this should be fine.
+ */
+ ranges->values[2*ranges->nranges + ranges->nvalues] = newval;
+ ranges->nvalues++;
+
+ /*
+ * Check we haven't broken the ordering of boundary values (checks
+ * both parts, but that doesn't hurt).
+ */
+ AssertCheckRanges(ranges, cmpFn, colloid);
+
+ /* Also check the range contains the value we just added. */
// FIXME Assert(ranges, cmpFn, colloid);
+ /* yep, we've modified the range */
return true;
}
@@ -1506,12 +1661,6 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
MemoryContext ctx;
MemoryContext oldctx;
- /*
- * This should only be used in batch mode, and there should be no
- * ranges, just individual values.
- */
- Assert((ranges->batch_mode) && (ranges->nranges == 0));
-
/* we'll certainly need the comparator, so just look it up now */
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
BTLessStrategyNumber);
@@ -1534,8 +1683,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
/* OK build the combine ranges */
cranges = build_combine_ranges(cmpFn, ranges->colloid, ranges,
- false, (Datum) 0, &ncranges,
- true); /* deduplicate */
+ &ncranges); /* deduplicate */
if (ncranges > 1)
{
@@ -1548,7 +1696,7 @@ compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
* don't expect more tuples to be inserted soon.
*/
ncranges = reduce_combine_ranges(cranges, ncranges, distances,
- max_values, cmpFn, ranges->colloid);
+ max_values, cmpFn, ranges->colloid);
Assert(count_values(cranges, ncranges) <= max_values);
}
@@ -2052,8 +2200,7 @@ brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
* In batch mode, we need to compress the accumulated values to the
* actually requested number of values/ranges.
*/
- if (ranges->batch_mode)
- compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
+ compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
s = range_serialize(ranges);
dst[0] = PointerGetDatum(s);
@@ -2114,15 +2261,39 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int target_maxvalues;
+ int maxvalues;
BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+ /* what was specified as a reloption? */
+ target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, target_maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
- ranges = minmax_multi_init(MaxHeapTuplesPerPage * pagesPerRange);
+ ranges = minmax_multi_init(maxvalues);
ranges->attno = attno;
ranges->colloid = colloid;
ranges->typid = attr->atttypid;
- ranges->batch_mode = true;
- ranges->target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
+ ranges->target_maxvalues = target_maxvalues;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
MemoryContextSwitchTo(oldctx);
@@ -2136,10 +2307,38 @@ brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
{
MemoryContext oldctx;
+ int maxvalues;
+ BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
+
oldctx = MemoryContextSwitchTo(column->bv_context);
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+
+ /*
+ * Determine the insert buffer size - we use 10x the target, capped
+ * to the maximum number of values in the heap range. This is more
+ * than enough, considering the actual number of rows per page is
+ * likely much lower, but meh.
+ */
+ maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
+ MaxHeapTuplesPerPage * pagesPerRange);
+
+ /* but always at least the original value */
+ maxvalues = Max(maxvalues, serialized->maxvalues);
+
+ /* always cap by MIN/MAX */
+ maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
+ maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
+
+ ranges = range_deserialize(maxvalues, serialized);
+
+ ranges->attno = attno;
+ ranges->colloid = colloid;
+ ranges->typid = attr->atttypid;
+
+ /* we'll certainly need the comparator, so just look it up now */
+ ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
+ BTLessStrategyNumber);
column->bv_mem_value = PointerGetDatum(ranges);
column->bv_serialize = brin_minmax_multi_serialize;
@@ -2184,7 +2383,7 @@ brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
attno = column->bv_attno;
serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
- ranges = range_deserialize(serialized);
+ ranges = range_deserialize(serialized->maxvalues, serialized);
/* inspect the ranges, and for each one evaluate the scan keys */
for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
@@ -2371,8 +2570,8 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
- ranges_a = range_deserialize(serialized_a);
- ranges_b = range_deserialize(serialized_b);
+ ranges_a = range_deserialize(serialized_a->maxvalues, serialized_a);
+ ranges_b = range_deserialize(serialized_b->maxvalues, serialized_b);
/* make sure neither of the ranges is NULL */
Assert(ranges_a && ranges_b);
@@ -2408,7 +2607,7 @@ brin_minmax_multi_union(PG_FUNCTION_ARGS)
cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
BTLessStrategyNumber);
- /* sort the combine ranges (don't deduplicate) */
+ /* sort the combine ranges (no need to deduplicate) */
sort_combine_ranges(cmpFn, colloid, cranges, ncranges, false);
/*
@@ -2637,7 +2836,7 @@ brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
fmgr_info(outfunc, &fmgrinfo);
/* deserialize the range info easy-to-process pieces */
- ranges_deserialized = range_deserialize(ranges);
+ ranges_deserialized = range_deserialize(ranges->maxvalues, ranges);
appendStringInfo(&str, "nranges: %u nvalues: %u maxvalues: %u",
ranges_deserialized->nranges,
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0008-Define-multi-minmax-oclasses-for-types-with-20210215.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0008-Define-multi-minmax-oclasses-for-types-with-20210215.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 70+ messages in thread
* Reducing output size of nodeToString
@ 2023-12-06 21:08 Matthias van de Meent <[email protected]>
2023-12-07 12:09 ` Re: Reducing output size of nodeToString David Rowley <[email protected]>
2024-01-02 10:30 ` Re: Reducing output size of nodeToString Peter Eisentraut <[email protected]>
0 siblings, 2 replies; 70+ messages in thread
From: Matthias van de Meent @ 2023-12-06 21:08 UTC (permalink / raw)
To: PostgreSQL Hackers <[email protected]>; Peter Eisentraut <[email protected]>; Andres Freund <[email protected]>; +Cc: Michel Pelletier <[email protected]>
Hi,
PFA a patch that reduces the output size of nodeToString by 50%+ in
most cases (measured on pg_rewrite), which on my system reduces the
total size of pg_rewrite by 33% to 472KiB. This does keep the textual
pg_node_tree format alive, but reduces its size signficantly.
The basic techniques used are
- Don't emit scalar fields when they contain a default value, and
make the reading code aware of this.
- Reasonable defaults are set for most datatypes, and overrides can
be added with new pg_node_attr() attributes. No introspection into
non-null Node/Array/etc. is being done though.
- Reset more fields to their default values before storing the values.
- Don't write trailing 0s in outDatum calls for by-ref types. This
saves many bytes for Name fields, but also some other pre-existing
entry points.
Future work will probably have to be on a significantly different
storage format, as the textual format is about to hit its entropy
limits.
See also [0], [1] and [2], where complaints about the verbosity of
nodeToString were vocalized.
Kind regards,
Matthias van de Meent
[0] https://www.postgresql.org/message-id/CAEze2WgGexDM63dOvndLdAWwA6uSmSsc97jmrCuNmrF1JEDK7w%40mail.gma...
[1] https://www.postgresql.org/message-id/flat/CACxu%3DvL_SD%3DWJiFSJyyBuZAp_2v_XBqb1x9JBiqz52a_g9z3jA%4...
[2] https://www.postgresql.org/message-id/4b27fc50-8cd6-46f5-ab20-88dbaadca645%40eisentraut.org
Attachments:
[application/octet-stream] v0-0001-Reduce-the-size-of-serialized-nodes-in-nodeToStri.patch (55.5K, ../../CAEze2WgrCiR3JZmWyB0YTc8HV7ewRdx13j0CqD6mVkYAW+SFGQ@mail.gmail.com/2-v0-0001-Reduce-the-size-of-serialized-nodes-in-nodeToStri.patch)
download | inline diff:
From 6f7d037a03f2020166407b46a6bf6a2b8226f906 Mon Sep 17 00:00:00 2001
From: Matthias van de Meent <[email protected]>
Date: Wed, 6 Dec 2023 21:49:42 +0100
Subject: [PATCH v0] Reduce the size of serialized nodes in nodeToString
Default fields don't need to be written to disk. So, add
infrastructure that handles that correctly.
This updates the rowsecurity tests which had a hard
dependency on the output format of nodeToString, and thus
broke with these changes. It is now more robust against
similar issues, as long as the textual format remains.
---
src/backend/nodes/gen_node_support.pl | 204 ++++++++++++++++----
src/backend/nodes/outfuncs.c | 151 +++++++++------
src/backend/nodes/read.c | 47 +++++
src/backend/nodes/readfuncs.c | 215 +++++++++++++++-------
src/backend/rewrite/rewriteDefine.c | 102 ++++++++++
src/include/nodes/parsenodes.h | 11 +-
src/include/nodes/primnodes.h | 40 ++--
src/include/nodes/readfuncs.h | 1 +
src/test/regress/expected/rowsecurity.out | 5 +-
src/test/regress/sql/rowsecurity.sql | 5 +-
10 files changed, 590 insertions(+), 191 deletions(-)
diff --git a/src/backend/nodes/gen_node_support.pl b/src/backend/nodes/gen_node_support.pl
index 72c7963578..2fe6547be0 100644
--- a/src/backend/nodes/gen_node_support.pl
+++ b/src/backend/nodes/gen_node_support.pl
@@ -361,6 +361,18 @@ foreach my $infile (@ARGV)
elsif ($attr =~ /^nodetag_number\((\d+)\)$/)
{
$manual_nodetag_number{$in_struct} = $1;
+ }
+ elsif ($attr eq 'no_default')
+ {
+
+ }
+ elsif ($attr =~ /^default\(([\w\d."'-]+)\)$/)
+ {
+
+ }
+ elsif ($attr =~ /^default_ref\(([\w\d."'-]+)\)$/)
+ {
+
}
else
{
@@ -436,7 +448,7 @@ foreach my $infile (@ARGV)
}
# normal struct field
elsif ($line =~
- /^\s*(.+)\s*\b(\w+)(\[[\w\s+]+\])?\s*(?:pg_node_attr\(([\w(), ]*)\))?;/
+ /^\s*(.+)\s*\b(\w+)(\[[\w\s+]+\])?\s*(?:pg_node_attr\(([\w\d(), ."'-]*)\))?;/
)
{
if ($is_node_struct)
@@ -468,6 +480,8 @@ foreach my $infile (@ARGV)
if ( $attr !~ /^array_size\(\w+\)$/
&& $attr !~ /^copy_as\(\w+\)$/
&& $attr !~ /^read_as\(\w+\)$/
+ && $attr !~ /^default\([\w\d."'-]+\)$/
+ && $attr !~ /^default_ref\([\w\d."'-]+\)$/
&& !elem $attr,
qw(copy_as_scalar
equal_as_scalar
@@ -478,7 +492,8 @@ foreach my $infile (@ARGV)
read_write_ignore
write_only_relids
write_only_nondefault_pathtarget
- write_only_req_outer))
+ write_only_req_outer
+ no_default))
{
die
"$infile:$lineno: unrecognized attribute \"$attr\"\n";
@@ -494,7 +509,7 @@ foreach my $infile (@ARGV)
}
# function pointer field
elsif ($line =~
- /^\s*([\w\s*]+)\s*\(\*(\w+)\)\s*\((.*)\)\s*(?:pg_node_attr\(([\w(), ]*)\))?;/
+ /^\s*([\w\s*]+)\s*\(\*(\w+)\)\s*\((.*)\)\s*(?:pg_node_attr\(([\w\d(), ."'-]*)\))?;/
)
{
if ($is_node_struct)
@@ -969,6 +984,10 @@ _read${n}(void)
my $array_size_field;
my $read_as_field;
my $read_write_ignore = 0;
+ my $opt = '_OPT';
+ my $default = 0;
+ my $manual_default = 0;
+
foreach my $a (@a)
{
if ($a =~ /^array_size\(([\w.]+)\)$/)
@@ -987,6 +1006,111 @@ _read${n}(void)
{
$read_write_ignore = 1;
}
+ elsif ($a =~ /^default\(([\w\d+."'-]+)\)$/)
+ {
+ $default = $1;
+ $manual_default = 1;
+ die
+ "default() and no_default can't both be set"
+ if $opt eq '';
+ }
+ elsif ($a =~ /^default_ref\(([\w\d+."'-]+)\)$/)
+ {
+ $default = "NODE_FIELD($1)";
+ $manual_default = 1;
+ die
+ "default() and no_default can't both be set"
+ if $opt eq '';
+ }
+ elsif ($a eq 'no_default')
+ {
+ $opt = '';
+ die
+ "default() and no_default can't both be set"
+ if $manual_default eq 1;
+ }
+ }
+
+ my $default_clause = 'invalid';
+
+ if ($opt eq "_OPT")
+ {
+ if ($manual_default eq 1)
+ {
+ # we don't set a typed clause here
+ }
+ elsif ($t eq 'bool')
+ {
+ $default = 'false';
+ }
+ elsif ($t eq 'int' && $f =~ 'location$')
+ {
+ $default = '-1';
+ }
+ elsif ($t eq 'int'
+ || $t eq 'int16'
+ || $t eq 'int32'
+ || $t eq 'uint32'
+ || $t eq 'bits32'
+ || $t eq 'BlockNumber'
+ || $t eq 'Index'
+ || $t eq 'SubTransactionId'
+ || $t eq 'uint64'
+ || $t eq 'AclMode'
+ || $t eq 'Oid'
+ || $t eq 'RelFileNumber'
+ || $t eq 'long'
+ || $t eq 'char')
+ {
+ $default = 0;
+ }
+ elsif ($t eq 'AttrNumber')
+ {
+ $default = 1;
+ }
+ elsif ($t eq 'StrategyNumber')
+ {
+ $default = 3; # BtEqualStrategy
+ }
+ elsif ($t eq 'double'
+ || $t eq 'Cardinality'
+ || $t eq 'Cost'
+ || $t eq 'QualCost'
+ || $t eq 'Selectivity')
+ {
+ $default = '0.0';
+ }
+ elsif ($t eq 'char*')
+ {
+ $default_clause = '';
+ $opt = '_NONNULL';
+ }
+ elsif ($t eq 'Bitmapset*' || $t eq 'Relids')
+ {
+ $default = 'NULL'; # but not included
+ }
+ elsif (elem $t, @enum_types)
+ {
+ $default = 0;
+ }
+ elsif ($t =~ /^(\w+)(\*|\[\w+\])$/ and elem $1, @scalar_types)
+ {
+ $opt = '';
+ }
+
+ if ($opt eq '')
+ {
+ $default_clause = '';
+ }
+ elsif ($default_clause eq 'invalid')
+ {
+ $default_clause = ", $default";
+ }
+ }
+ else
+ {
+ die "invalid optionality" unless $opt eq '';
+ $default_clause = "";
}
if ($read_write_ignore)
@@ -1007,13 +1131,13 @@ _read${n}(void)
# select instructions by field type
if ($t eq 'bool')
{
- print $off "\tWRITE_BOOL_FIELD($f);\n";
- print $rff "\tREAD_BOOL_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_BOOL_FIELD$opt($f$default_clause);\n";
+ print $rff "\tREAD_BOOL_FIELD$opt($f$default_clause);\n" unless $no_read;
}
elsif ($t eq 'int' && $f =~ 'location$')
{
- print $off "\tWRITE_LOCATION_FIELD($f);\n";
- print $rff "\tREAD_LOCATION_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_LOCATION_FIELD_OPT($f);\n";
+ print $rff "\tREAD_LOCATION_FIELD_OPT($f);\n" unless $no_read;
}
elsif ($t eq 'int'
|| $t eq 'int16'
@@ -1021,8 +1145,8 @@ _read${n}(void)
|| $t eq 'AttrNumber'
|| $t eq 'StrategyNumber')
{
- print $off "\tWRITE_INT_FIELD($f);\n";
- print $rff "\tREAD_INT_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_INT_FIELD$opt($f$default_clause);\n";
+ print $rff "\tREAD_INT_FIELD$opt($f$default_clause);\n" unless $no_read;
}
elsif ($t eq 'uint32'
|| $t eq 'bits32'
@@ -1030,71 +1154,71 @@ _read${n}(void)
|| $t eq 'Index'
|| $t eq 'SubTransactionId')
{
- print $off "\tWRITE_UINT_FIELD($f);\n";
- print $rff "\tREAD_UINT_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_UINT_FIELD$opt($f$default_clause);\n";
+ print $rff "\tREAD_UINT_FIELD$opt($f$default_clause);\n" unless $no_read;
}
elsif ($t eq 'uint64'
|| $t eq 'AclMode')
{
- print $off "\tWRITE_UINT64_FIELD($f);\n";
- print $rff "\tREAD_UINT64_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_UINT64_FIELD$opt($f$default_clause);\n";
+ print $rff "\tREAD_UINT64_FIELD$opt($f$default_clause);\n" unless $no_read;
}
elsif ($t eq 'Oid' || $t eq 'RelFileNumber')
{
- print $off "\tWRITE_OID_FIELD($f);\n";
- print $rff "\tREAD_OID_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_OID_FIELD$opt($f$default_clause);\n";
+ print $rff "\tREAD_OID_FIELD$opt($f$default_clause);\n" unless $no_read;
}
elsif ($t eq 'long')
{
- print $off "\tWRITE_LONG_FIELD($f);\n";
- print $rff "\tREAD_LONG_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_LONG_FIELD$opt($f$default_clause);\n";
+ print $rff "\tREAD_LONG_FIELD$opt($f$default_clause);\n" unless $no_read;
}
elsif ($t eq 'char')
{
- print $off "\tWRITE_CHAR_FIELD($f);\n";
- print $rff "\tREAD_CHAR_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_CHAR_FIELD$opt($f$default_clause);\n";
+ print $rff "\tREAD_CHAR_FIELD$opt($f$default_clause);\n" unless $no_read;
}
elsif ($t eq 'double')
{
- print $off "\tWRITE_FLOAT_FIELD($f);\n";
- print $rff "\tREAD_FLOAT_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_FLOAT_FIELD$opt($f$default_clause);\n";
+ print $rff "\tREAD_FLOAT_FIELD$opt($f$default_clause);\n" unless $no_read;
}
elsif ($t eq 'Cardinality')
{
- print $off "\tWRITE_FLOAT_FIELD($f);\n";
- print $rff "\tREAD_FLOAT_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_FLOAT_FIELD$opt($f$default_clause);\n";
+ print $rff "\tREAD_FLOAT_FIELD$opt($f$default_clause);\n" unless $no_read;
}
elsif ($t eq 'Cost')
{
- print $off "\tWRITE_FLOAT_FIELD($f);\n";
- print $rff "\tREAD_FLOAT_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_FLOAT_FIELD$opt($f$default_clause);\n";
+ print $rff "\tREAD_FLOAT_FIELD$opt($f$default_clause);\n" unless $no_read;
}
elsif ($t eq 'QualCost')
{
- print $off "\tWRITE_FLOAT_FIELD($f.startup);\n";
- print $off "\tWRITE_FLOAT_FIELD($f.per_tuple);\n";
- print $rff "\tREAD_FLOAT_FIELD($f.startup);\n" unless $no_read;
- print $rff "\tREAD_FLOAT_FIELD($f.per_tuple);\n" unless $no_read;
+ print $off "\tWRITE_FLOAT_FIELD$opt($f.startup$default_clause);\n";
+ print $off "\tWRITE_FLOAT_FIELD$opt($f.per_tuple$default_clause);\n";
+ print $rff "\tREAD_FLOAT_FIELD$opt($f.startup$default_clause);\n" unless $no_read;
+ print $rff "\tREAD_FLOAT_FIELD$opt($f.per_tuple$default_clause);\n" unless $no_read;
}
elsif ($t eq 'Selectivity')
{
- print $off "\tWRITE_FLOAT_FIELD($f);\n";
- print $rff "\tREAD_FLOAT_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_FLOAT_FIELD$opt($f$default_clause);\n";
+ print $rff "\tREAD_FLOAT_FIELD$opt($f$default_clause);\n" unless $no_read;
}
elsif ($t eq 'char*')
{
- print $off "\tWRITE_STRING_FIELD($f);\n";
- print $rff "\tREAD_STRING_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_STRING_FIELD$opt($f$default_clause);\n";
+ print $rff "\tREAD_STRING_FIELD$opt($f$default_clause);\n" unless $no_read;
}
elsif ($t eq 'Bitmapset*' || $t eq 'Relids')
{
- print $off "\tWRITE_BITMAPSET_FIELD($f);\n";
- print $rff "\tREAD_BITMAPSET_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_BITMAPSET_FIELD$opt($f);\n";
+ print $rff "\tREAD_BITMAPSET_FIELD$opt($f);\n" unless $no_read;
}
elsif (elem $t, @enum_types)
{
- print $off "\tWRITE_ENUM_FIELD($f, $t);\n";
- print $rff "\tREAD_ENUM_FIELD($f, $t);\n" unless $no_read;
+ print $off "\tWRITE_ENUM_FIELD$opt($f, $t$default_clause);\n";
+ print $rff "\tREAD_ENUM_FIELD$opt($f, $t$default_clause);\n" unless $no_read;
}
# arrays of scalar types
elsif ($t =~ /^(\w+)(\*|\[\w+\])$/ and elem $1, @scalar_types)
@@ -1140,7 +1264,7 @@ _read${n}(void)
{
(my $f2 = $f) =~ s/pathtarget/parent/;
print $off "\tif (node->$f != node->$f2->reltarget)\n"
- . "\t\tWRITE_NODE_FIELD($f);\n";
+ . "\t\tWRITE_NODE_FIELD$opt($f);\n";
}
elsif ($t eq 'ParamPathInfo*' && elem 'write_only_req_outer', @a)
{
@@ -1163,8 +1287,8 @@ _read${n}(void)
if (elem $1, @no_read or elem $1, @nodetag_only)
and !$no_read;
- print $off "\tWRITE_NODE_FIELD($f);\n";
- print $rff "\tREAD_NODE_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_NODE_FIELD$opt($f);\n";
+ print $rff "\tREAD_NODE_FIELD$opt($f);\n" unless $no_read;
}
# arrays of node pointers (currently supported for write only)
elsif (($t =~ /^(\w+)\*\*$/ or $t =~ /^struct\s+(\w+)\*\*$/)
diff --git a/src/backend/nodes/outfuncs.c b/src/backend/nodes/outfuncs.c
index e66a99247e..38cb4b03f4 100644
--- a/src/backend/nodes/outfuncs.c
+++ b/src/backend/nodes/outfuncs.c
@@ -43,62 +43,91 @@ static void outDouble(StringInfo str, double d);
/* Write an integer field (anything written as ":fldname %d") */
#define WRITE_INT_FIELD(fldname) \
appendStringInfo(str, " :" CppAsString(fldname) " %d", node->fldname)
+#define WRITE_INT_FIELD_OPT(fldname, default) \
+ ((node->fldname == default) ? (0) : WRITE_INT_FIELD(fldname))
/* Write an unsigned integer field (anything written as ":fldname %u") */
#define WRITE_UINT_FIELD(fldname) \
appendStringInfo(str, " :" CppAsString(fldname) " %u", node->fldname)
+#define WRITE_UINT_FIELD_OPT(fldname, default) \
+ ((node->fldname == default) ? (0) : WRITE_UINT_FIELD(fldname))
/* Write an unsigned integer field (anything written with UINT64_FORMAT) */
#define WRITE_UINT64_FIELD(fldname) \
appendStringInfo(str, " :" CppAsString(fldname) " " UINT64_FORMAT, \
node->fldname)
+#define WRITE_UINT64_FIELD_OPT(fldname, default) \
+ ((node->fldname == default) ? (0) : WRITE_UINT64_FIELD(fldname))
/* Write an OID field (don't hard-wire assumption that OID is same as uint) */
#define WRITE_OID_FIELD(fldname) \
appendStringInfo(str, " :" CppAsString(fldname) " %u", node->fldname)
+#define WRITE_OID_FIELD_OPT(fldname, default) \
+ ((node->fldname == default) ? (0) : WRITE_OID_FIELD(fldname))
/* Write a long-integer field */
#define WRITE_LONG_FIELD(fldname) \
appendStringInfo(str, " :" CppAsString(fldname) " %ld", node->fldname)
+#define WRITE_LONG_FIELD_OPT(fldname, default) \
+ ((node->fldname == default) ? (0) : WRITE_LONG_FIELD(fldname))
/* Write a char field (ie, one ascii character) */
#define WRITE_CHAR_FIELD(fldname) \
(appendStringInfo(str, " :" CppAsString(fldname) " "), \
outChar(str, node->fldname))
+#define WRITE_CHAR_FIELD_OPT(fldname, default) \
+ ((node->fldname == default) ? (0) : WRITE_CHAR_FIELD(fldname))
/* Write an enumerated-type field as an integer code */
#define WRITE_ENUM_FIELD(fldname, enumtype) \
appendStringInfo(str, " :" CppAsString(fldname) " %d", \
(int) node->fldname)
+#define WRITE_ENUM_FIELD_OPT(fldname, enumtype, default) \
+ ((node->fldname == default) ? (0) : WRITE_ENUM_FIELD(fldname, enumtype))
/* Write a float field (actually, they're double) */
#define WRITE_FLOAT_FIELD(fldname) \
(appendStringInfo(str, " :" CppAsString(fldname) " "), \
outDouble(str, node->fldname))
+#define WRITE_FLOAT_FIELD_OPT(fldname, default) \
+ ((node->fldname == default) ? (0) : WRITE_FLOAT_FIELD(fldname))
/* Write a boolean field */
#define WRITE_BOOL_FIELD(fldname) \
appendStringInfo(str, " :" CppAsString(fldname) " %s", \
booltostr(node->fldname))
+#define WRITE_BOOL_FIELD_OPT(fldname, default) \
+ ((node->fldname == default) ? (0) : WRITE_BOOL_FIELD(fldname))
/* Write a character-string (possibly NULL) field */
#define WRITE_STRING_FIELD(fldname) \
(appendStringInfoString(str, " :" CppAsString(fldname) " "), \
outToken(str, node->fldname))
+#define WRITE_STRING_FIELD_OPT(fldname, default) \
+ ((strcmp(node->fldname, default) == 0) ? (0) : \
+ WRITE_STRING_FIELD(fldname))
+#define WRITE_STRING_FIELD_NONNULL(fldname) \
+ ((node->fldname == NULL) ? (0) : WRITE_STRING_FIELD(fldname))
/* Write a parse location field (actually same as INT case) */
#define WRITE_LOCATION_FIELD(fldname) \
appendStringInfo(str, " :" CppAsString(fldname) " %d", node->fldname)
+#define WRITE_LOCATION_FIELD_OPT(fldname) \
+ (node->fldname == -1 ? (0) : WRITE_LOCATION_FIELD(fldname))
/* Write a Node field */
#define WRITE_NODE_FIELD(fldname) \
(appendStringInfoString(str, " :" CppAsString(fldname) " "), \
outNode(str, node->fldname))
+#define WRITE_NODE_FIELD_OPT(fldname) \
+ (node->fldname == NULL ? (0) : WRITE_NODE_FIELD(fldname))
/* Write a bitmapset field */
#define WRITE_BITMAPSET_FIELD(fldname) \
(appendStringInfoString(str, " :" CppAsString(fldname) " "), \
outBitmapset(str, node->fldname))
+#define WRITE_BITMAPSET_FIELD_OPT(fldname) \
+ (node->fldname == NULL ? (0) : WRITE_BITMAPSET_FIELD(fldname))
/* Write a variable-length array (not a List) of Node pointers */
#define WRITE_NODE_ARRAY(fldname, len) \
@@ -132,6 +161,7 @@ static void outDouble(StringInfo str, double d);
#define booltostr(x) ((x) ? "true" : "false")
+#define NODE_FIELD(fldname) (node->fldname)
/*
* outToken
@@ -359,6 +389,11 @@ outDatum(StringInfo str, Datum value, int typlen, bool typbyval)
else
{
appendStringInfo(str, "%u [ ", (unsigned int) length);
+
+ /* truncate zero postfix of the data */
+ while (length != 0 && s[length - 1] == 0)
+ length -= 1;
+
for (i = 0; i < length; i++)
appendStringInfo(str, "%d ", (int) (s[i]));
appendStringInfoChar(str, ']');
@@ -381,18 +416,18 @@ _outConst(StringInfo str, const Const *node)
WRITE_NODE_TYPE("CONST");
WRITE_OID_FIELD(consttype);
- WRITE_INT_FIELD(consttypmod);
- WRITE_OID_FIELD(constcollid);
+ WRITE_INT_FIELD_OPT(consttypmod, -1);
+ WRITE_OID_FIELD_OPT(constcollid, InvalidOid);
WRITE_INT_FIELD(constlen);
- WRITE_BOOL_FIELD(constbyval);
- WRITE_BOOL_FIELD(constisnull);
- WRITE_LOCATION_FIELD(location);
+ WRITE_BOOL_FIELD_OPT(constbyval, true);
+ WRITE_BOOL_FIELD_OPT(constisnull, false);
+ WRITE_LOCATION_FIELD_OPT(location);
- appendStringInfoString(str, " :constvalue ");
- if (node->constisnull)
- appendStringInfoString(str, "<>");
- else
+ if (!node->constisnull)
+ {
+ appendStringInfoString(str, " :constvalue ");
outDatum(str, node->constvalue, node->constlen, node->constbyval);
+ }
}
static void
@@ -419,7 +454,7 @@ _outBoolExpr(StringInfo str, const BoolExpr *node)
outToken(str, opstr);
WRITE_NODE_FIELD(args);
- WRITE_LOCATION_FIELD(location);
+ WRITE_LOCATION_FIELD_OPT(location);
}
static void
@@ -495,39 +530,39 @@ _outRangeTblEntry(StringInfo str, const RangeTblEntry *node)
WRITE_NODE_TYPE("RANGETBLENTRY");
/* put alias + eref first to make dump more legible */
- WRITE_NODE_FIELD(alias);
+ WRITE_NODE_FIELD_OPT(alias);
WRITE_NODE_FIELD(eref);
- WRITE_ENUM_FIELD(rtekind, RTEKind);
+ WRITE_ENUM_FIELD_OPT(rtekind, RTEKind, 0);
switch (node->rtekind)
{
case RTE_RELATION:
WRITE_OID_FIELD(relid);
- WRITE_CHAR_FIELD(relkind);
- WRITE_INT_FIELD(rellockmode);
- WRITE_NODE_FIELD(tablesample);
- WRITE_UINT_FIELD(perminfoindex);
+ WRITE_CHAR_FIELD_OPT(relkind, 'r');
+ WRITE_INT_FIELD_OPT(rellockmode, 1);
+ WRITE_NODE_FIELD_OPT(tablesample);
+ WRITE_UINT_FIELD_OPT(perminfoindex, 1);
break;
case RTE_SUBQUERY:
WRITE_NODE_FIELD(subquery);
- WRITE_BOOL_FIELD(security_barrier);
+ WRITE_BOOL_FIELD_OPT(security_barrier, false);
/* we re-use these RELATION fields, too: */
WRITE_OID_FIELD(relid);
- WRITE_CHAR_FIELD(relkind);
- WRITE_INT_FIELD(rellockmode);
- WRITE_UINT_FIELD(perminfoindex);
+ WRITE_CHAR_FIELD_OPT(relkind, 0);
+ WRITE_INT_FIELD_OPT(rellockmode, 0);
+ WRITE_UINT_FIELD_OPT(perminfoindex, 1);
break;
case RTE_JOIN:
- WRITE_ENUM_FIELD(jointype, JoinType);
- WRITE_INT_FIELD(joinmergedcols);
+ WRITE_ENUM_FIELD_OPT(jointype, JoinType, 0);
+ WRITE_INT_FIELD_OPT(joinmergedcols, 0);
WRITE_NODE_FIELD(joinaliasvars);
WRITE_NODE_FIELD(joinleftcols);
WRITE_NODE_FIELD(joinrightcols);
- WRITE_NODE_FIELD(join_using_alias);
+ WRITE_NODE_FIELD_OPT(join_using_alias);
break;
case RTE_FUNCTION:
WRITE_NODE_FIELD(functions);
- WRITE_BOOL_FIELD(funcordinality);
+ WRITE_BOOL_FIELD_OPT(funcordinality, false);
break;
case RTE_TABLEFUNC:
WRITE_NODE_FIELD(tablefunc);
@@ -541,7 +576,7 @@ _outRangeTblEntry(StringInfo str, const RangeTblEntry *node)
case RTE_CTE:
WRITE_STRING_FIELD(ctename);
WRITE_UINT_FIELD(ctelevelsup);
- WRITE_BOOL_FIELD(self_reference);
+ WRITE_BOOL_FIELD_OPT(self_reference, false);
WRITE_NODE_FIELD(coltypes);
WRITE_NODE_FIELD(coltypmods);
WRITE_NODE_FIELD(colcollations);
@@ -563,10 +598,10 @@ _outRangeTblEntry(StringInfo str, const RangeTblEntry *node)
break;
}
- WRITE_BOOL_FIELD(lateral);
- WRITE_BOOL_FIELD(inh);
- WRITE_BOOL_FIELD(inFromCl);
- WRITE_NODE_FIELD(securityQuals);
+ WRITE_BOOL_FIELD_OPT(lateral, false);
+ WRITE_BOOL_FIELD_OPT(inh, true);
+ WRITE_BOOL_FIELD_OPT(inFromCl, true);
+ WRITE_NODE_FIELD_OPT(securityQuals);
}
static void
@@ -638,7 +673,7 @@ _outA_Expr(StringInfo str, const A_Expr *node)
WRITE_NODE_FIELD(lexpr);
WRITE_NODE_FIELD(rexpr);
- WRITE_LOCATION_FIELD(location);
+ WRITE_LOCATION_FIELD_OPT(location);
}
static void
@@ -696,7 +731,7 @@ _outA_Const(StringInfo str, const A_Const *node)
appendStringInfoString(str, " :val ");
outNode(str, &node->val);
}
- WRITE_LOCATION_FIELD(location);
+ WRITE_LOCATION_FIELD_OPT(location);
}
static void
@@ -705,9 +740,9 @@ _outConstraint(StringInfo str, const Constraint *node)
WRITE_NODE_TYPE("CONSTRAINT");
WRITE_STRING_FIELD(conname);
- WRITE_BOOL_FIELD(deferrable);
- WRITE_BOOL_FIELD(initdeferred);
- WRITE_LOCATION_FIELD(location);
+ WRITE_BOOL_FIELD_OPT(deferrable, false);
+ WRITE_BOOL_FIELD_OPT(initdeferred, false);
+ WRITE_LOCATION_FIELD_OPT(location);
appendStringInfoString(str, " :contype ");
switch (node->contype)
@@ -719,10 +754,10 @@ _outConstraint(StringInfo str, const Constraint *node)
case CONSTR_NOTNULL:
appendStringInfoString(str, "NOT_NULL");
WRITE_NODE_FIELD(keys);
- WRITE_INT_FIELD(inhcount);
- WRITE_BOOL_FIELD(is_no_inherit);
- WRITE_BOOL_FIELD(skip_validation);
- WRITE_BOOL_FIELD(initially_valid);
+ WRITE_INT_FIELD_OPT(inhcount, 0);
+ WRITE_BOOL_FIELD_OPT(is_no_inherit, false);
+ WRITE_BOOL_FIELD_OPT(skip_validation, false);
+ WRITE_BOOL_FIELD_OPT(initially_valid, true);
break;
case CONSTR_DEFAULT:
@@ -733,7 +768,7 @@ _outConstraint(StringInfo str, const Constraint *node)
case CONSTR_IDENTITY:
appendStringInfoString(str, "IDENTITY");
- WRITE_NODE_FIELD(options);
+ WRITE_NODE_FIELD_OPT(options);
WRITE_CHAR_FIELD(generated_when);
break;
@@ -746,46 +781,46 @@ _outConstraint(StringInfo str, const Constraint *node)
case CONSTR_CHECK:
appendStringInfoString(str, "CHECK");
- WRITE_BOOL_FIELD(is_no_inherit);
+ WRITE_BOOL_FIELD_OPT(is_no_inherit, false);
WRITE_NODE_FIELD(raw_expr);
WRITE_STRING_FIELD(cooked_expr);
- WRITE_BOOL_FIELD(skip_validation);
- WRITE_BOOL_FIELD(initially_valid);
+ WRITE_BOOL_FIELD_OPT(skip_validation, false);
+ WRITE_BOOL_FIELD_OPT(initially_valid, true);
break;
case CONSTR_PRIMARY:
appendStringInfoString(str, "PRIMARY_KEY");
WRITE_NODE_FIELD(keys);
- WRITE_NODE_FIELD(including);
- WRITE_NODE_FIELD(options);
+ WRITE_NODE_FIELD_OPT(including);
+ WRITE_NODE_FIELD_OPT(options);
WRITE_STRING_FIELD(indexname);
- WRITE_STRING_FIELD(indexspace);
- WRITE_BOOL_FIELD(reset_default_tblspc);
+ WRITE_STRING_FIELD_NONNULL(indexspace);
+ WRITE_BOOL_FIELD_OPT(reset_default_tblspc, false);
/* access_method and where_clause not currently used */
break;
case CONSTR_UNIQUE:
appendStringInfoString(str, "UNIQUE");
- WRITE_BOOL_FIELD(nulls_not_distinct);
+ WRITE_BOOL_FIELD_OPT(nulls_not_distinct, true);
WRITE_NODE_FIELD(keys);
- WRITE_NODE_FIELD(including);
- WRITE_NODE_FIELD(options);
+ WRITE_NODE_FIELD_OPT(including);
+ WRITE_NODE_FIELD_OPT(options);
WRITE_STRING_FIELD(indexname);
- WRITE_STRING_FIELD(indexspace);
- WRITE_BOOL_FIELD(reset_default_tblspc);
+ WRITE_STRING_FIELD_NONNULL(indexspace);
+ WRITE_BOOL_FIELD_OPT(reset_default_tblspc, false);
/* access_method and where_clause not currently used */
break;
case CONSTR_EXCLUSION:
appendStringInfoString(str, "EXCLUSION");
WRITE_NODE_FIELD(exclusions);
- WRITE_NODE_FIELD(including);
- WRITE_NODE_FIELD(options);
+ WRITE_NODE_FIELD_OPT(including);
+ WRITE_NODE_FIELD_OPT(options);
WRITE_STRING_FIELD(indexname);
- WRITE_STRING_FIELD(indexspace);
- WRITE_BOOL_FIELD(reset_default_tblspc);
+ WRITE_STRING_FIELD_NONNULL(indexspace);
+ WRITE_BOOL_FIELD_OPT(reset_default_tblspc, false);
WRITE_STRING_FIELD(access_method);
- WRITE_NODE_FIELD(where_clause);
+ WRITE_NODE_FIELD_OPT(where_clause);
break;
case CONSTR_FOREIGN:
@@ -799,8 +834,8 @@ _outConstraint(StringInfo str, const Constraint *node)
WRITE_NODE_FIELD(fk_del_set_cols);
WRITE_NODE_FIELD(old_conpfeqop);
WRITE_OID_FIELD(old_pktable_oid);
- WRITE_BOOL_FIELD(skip_validation);
- WRITE_BOOL_FIELD(initially_valid);
+ WRITE_BOOL_FIELD_OPT(skip_validation, false);
+ WRITE_BOOL_FIELD_OPT(initially_valid, true);
break;
case CONSTR_ATTR_DEFERRABLE:
diff --git a/src/backend/nodes/read.c b/src/backend/nodes/read.c
index 813eda3e73..52b71902e8 100644
--- a/src/backend/nodes/read.c
+++ b/src/backend/nodes/read.c
@@ -205,6 +205,50 @@ pg_strtok(int *length)
return ret_str;
}
+/*
+ * Check if the next token is 'expect_token'.
+ *
+ * It handles similar to pg_strtok, except that this does not consume the
+ * next token, and has special casing for some less common tokens.
+ */
+bool
+pg_strtoken_next(const char *expect_token)
+{
+ const char *local_str; /* working pointer to string */
+ Size expect_len = strlen(expect_token);
+ char next_char;
+
+ local_str = pg_strtok_ptr;
+
+ while (*local_str == ' ' || *local_str == '\n' || *local_str == '\t')
+ local_str++;
+
+ if (*local_str == '\0')
+ return false; /* no more tokens */
+
+ Assert(expect_len > 0);
+
+ next_char = local_str[expect_len];
+
+ /* check if the next few bytes match the token */
+ if (strncmp(local_str, expect_token, expect_len) != 0)
+ return false;
+
+ /*
+ * Check that the token was actually terminated at the end of the
+ * expected token. Otherwise, we'd get positive matches for mathing the
+ * token of "is" against a local_str of "isn't", which is clearly wrong.
+ */
+ return (next_char == '\0' ||
+ next_char == ' ' ||
+ next_char == '\n' ||
+ next_char == '\t' ||
+ next_char == '(' ||
+ next_char == ')' ||
+ next_char == '{' ||
+ next_char == '}');
+}
+
/*
* debackslash -
* create a palloc'd string holding the given token.
@@ -338,7 +382,10 @@ nodeRead(const char *token, int tok_len)
result = parseNodeString();
token = pg_strtok(&tok_len);
if (token == NULL || token[0] != '}')
+ {
+ Assert(false);
elog(ERROR, "did not find '}' at end of input node");
+ }
break;
case LEFT_PAREN:
{
diff --git a/src/backend/nodes/readfuncs.c b/src/backend/nodes/readfuncs.c
index cc2021c1f7..88699bff54 100644
--- a/src/backend/nodes/readfuncs.c
+++ b/src/backend/nodes/readfuncs.c
@@ -56,113 +56,173 @@
READ_LOCALS_NO_FIELDS(nodeTypeName); \
READ_TEMP_LOCALS()
+/* */
+#define READ_OPT_SCAFFOLD(fldname, read_field_code, default_value) \
+ if (pg_strtoken_next(":" CppAsString(fldname))) \
+ { \
+ read_field_code; \
+ } \
+ else \
+ local_node->fldname = default_value
+
+#define MY_NODE local_node
+
/* Read an integer field (anything written as ":fldname %d") */
#define READ_INT_FIELD(fldname) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
token = pg_strtok(&length); /* get field value */ \
local_node->fldname = atoi(token)
+#define READ_INT_FIELD_OPT(fldname, default_value) \
+ READ_OPT_SCAFFOLD(fldname, READ_INT_FIELD(fldname), default_value)
/* Read an unsigned integer field (anything written as ":fldname %u") */
#define READ_UINT_FIELD(fldname) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
token = pg_strtok(&length); /* get field value */ \
local_node->fldname = atoui(token)
+#define READ_UINT_FIELD_OPT(fldname, default_value) \
+ READ_OPT_SCAFFOLD(fldname, READ_UINT_FIELD(fldname), default_value)
/* Read an unsigned integer field (anything written using UINT64_FORMAT) */
#define READ_UINT64_FIELD(fldname) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
token = pg_strtok(&length); /* get field value */ \
local_node->fldname = strtou64(token, NULL, 10)
+#define READ_UINT64_FIELD_OPT(fldname, default_value) \
+ READ_OPT_SCAFFOLD(fldname, READ_UINT64_FIELD(fldname), default_value)
/* Read a long integer field (anything written as ":fldname %ld") */
#define READ_LONG_FIELD(fldname) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
token = pg_strtok(&length); /* get field value */ \
local_node->fldname = atol(token)
+#define READ_LONG_FIELD_OPT(fldname, default_value) \
+ READ_OPT_SCAFFOLD(fldname, READ_LONG_FIELD(fldname), default_value)
/* Read an OID field (don't hard-wire assumption that OID is same as uint) */
#define READ_OID_FIELD(fldname) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
token = pg_strtok(&length); /* get field value */ \
local_node->fldname = atooid(token)
+#define READ_OID_FIELD_OPT(fldname, default_value) \
+ READ_OPT_SCAFFOLD(fldname, READ_OID_FIELD(fldname), default_value)
/* Read a char field (ie, one ascii character) */
#define READ_CHAR_FIELD(fldname) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
token = pg_strtok(&length); /* get field value */ \
/* avoid overhead of calling debackslash() for one char */ \
local_node->fldname = (length == 0) ? '\0' : (token[0] == '\\' ? token[1] : token[0])
+#define READ_CHAR_FIELD_OPT(fldname, default_value) \
+ READ_OPT_SCAFFOLD(fldname, READ_CHAR_FIELD(fldname), default_value)
/* Read an enumerated-type field that was written as an integer code */
#define READ_ENUM_FIELD(fldname, enumtype) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
token = pg_strtok(&length); /* get field value */ \
local_node->fldname = (enumtype) atoi(token)
+#define READ_ENUM_FIELD_OPT(fldname, enumtype, default_value) \
+ READ_OPT_SCAFFOLD(fldname, READ_ENUM_FIELD(fldname, enumtype), default_value)
/* Read a float field */
#define READ_FLOAT_FIELD(fldname) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
token = pg_strtok(&length); /* get field value */ \
local_node->fldname = atof(token)
+#define READ_FLOAT_FIELD_OPT(fldname, default_value) \
+ READ_OPT_SCAFFOLD(fldname, READ_FLOAT_FIELD(fldname), default_value)
/* Read a boolean field */
#define READ_BOOL_FIELD(fldname) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
token = pg_strtok(&length); /* get field value */ \
local_node->fldname = strtobool(token)
+#define READ_BOOL_FIELD_OPT(fldname, default_value) \
+ READ_OPT_SCAFFOLD(fldname, READ_BOOL_FIELD(fldname), default_value)
/* Read a character-string field */
#define READ_STRING_FIELD(fldname) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
token = pg_strtok(&length); /* get field value */ \
local_node->fldname = nullable_string(token, length)
+#define READ_STRING_FIELD_OPT(fldname, default_value) \
+ READ_OPT_SCAFFOLD(fldname, READ_STRING_FIELD(fldname), default_value)
+#define READ_STRING_FIELD_NONNULL(fldname) \
+ READ_OPT_SCAFFOLD(fldname, READ_STRING_FIELD(fldname), NULL)
/* Read a parse location field (and possibly throw away the value) */
#ifdef WRITE_READ_PARSE_PLAN_TREES
#define READ_LOCATION_FIELD(fldname) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
token = pg_strtok(&length); /* get field value */ \
local_node->fldname = restore_location_fields ? atoi(token) : -1
#else
#define READ_LOCATION_FIELD(fldname) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
token = pg_strtok(&length); /* get field value */ \
(void) token; /* in case not used elsewhere */ \
local_node->fldname = -1 /* set field to "unknown" */
#endif
+#define READ_LOCATION_FIELD_OPT(fldname) \
+ READ_OPT_SCAFFOLD(fldname, READ_LOCATION_FIELD(fldname), -1)
/* Read a Node field */
#define READ_NODE_FIELD(fldname) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
(void) token; /* in case not used elsewhere */ \
local_node->fldname = nodeRead(NULL, 0)
+#define READ_NODE_FIELD_OPT(fldname) \
+ READ_OPT_SCAFFOLD(fldname, READ_NODE_FIELD(fldname), NULL)
/* Read a bitmapset field */
#define READ_BITMAPSET_FIELD(fldname) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
(void) token; /* in case not used elsewhere */ \
local_node->fldname = _readBitmapset()
+/* Read a bitmapset field */
+#define READ_BITMAPSET_FIELD_OPT(fldname) \
+ READ_OPT_SCAFFOLD(fldname, READ_BITMAPSET_FIELD(fldname), NULL)
/* Read an attribute number array */
#define READ_ATTRNUMBER_ARRAY(fldname, len) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
local_node->fldname = readAttrNumberCols(len)
/* Read an oid array */
#define READ_OID_ARRAY(fldname, len) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
local_node->fldname = readOidCols(len)
/* Read an int array */
#define READ_INT_ARRAY(fldname, len) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
local_node->fldname = readIntCols(len)
/* Read a bool array */
#define READ_BOOL_ARRAY(fldname, len) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
local_node->fldname = readBoolCols(len)
+#define NODE_FIELD(fldname) (local_node->fldname)
+
/* Routine exit */
#define READ_DONE() \
return local_node
@@ -261,18 +321,19 @@ _readConst(void)
READ_LOCALS(Const);
READ_OID_FIELD(consttype);
- READ_INT_FIELD(consttypmod);
- READ_OID_FIELD(constcollid);
+ READ_INT_FIELD_OPT(consttypmod, -1);
+ READ_OID_FIELD_OPT(constcollid, InvalidOid);
READ_INT_FIELD(constlen);
- READ_BOOL_FIELD(constbyval);
- READ_BOOL_FIELD(constisnull);
- READ_LOCATION_FIELD(location);
+ READ_BOOL_FIELD_OPT(constbyval, true);
+ READ_BOOL_FIELD_OPT(constisnull, false);
+ READ_LOCATION_FIELD_OPT(location);
- token = pg_strtok(&length); /* skip :constvalue */
- if (local_node->constisnull)
- token = pg_strtok(&length); /* skip "<>" */
- else
+ if (!local_node->constisnull)
+ {
+ token = pg_strtok(&length);
+ Assert(strncmp(token, ":constvalue", strlen(":constvalue")) == 0);
local_node->constvalue = readDatum(local_node->constbyval);
+ }
READ_DONE();
}
@@ -283,19 +344,26 @@ _readBoolExpr(void)
READ_LOCALS(BoolExpr);
/* do-it-yourself enum representation */
- token = pg_strtok(&length); /* skip :boolop */
- token = pg_strtok(&length); /* get field value */
- if (length == 3 && strncmp(token, "and", 3) == 0)
- local_node->boolop = AND_EXPR;
- else if (length == 2 && strncmp(token, "or", 2) == 0)
- local_node->boolop = OR_EXPR;
- else if (length == 3 && strncmp(token, "not", 3) == 0)
- local_node->boolop = NOT_EXPR;
+ if (pg_strtoken_next(":boolop"))
+ {
+ token = pg_strtok(&length); /* skip :boolop */
+ token = pg_strtok(&length); /* get field value */
+ if (length == 3 && strncmp(token, "and", 3) == 0)
+ local_node->boolop = AND_EXPR;
+ else if (length == 2 && strncmp(token, "or", 2) == 0)
+ local_node->boolop = OR_EXPR;
+ else if (length == 3 && strncmp(token, "not", 3) == 0)
+ local_node->boolop = NOT_EXPR;
+ else
+ elog(ERROR, "unrecognized boolop \"%.*s\"", length, token);
+ }
else
- elog(ERROR, "unrecognized boolop \"%.*s\"", length, token);
+ {
+ local_node->boolop = AND_EXPR;
+ }
READ_NODE_FIELD(args);
- READ_LOCATION_FIELD(location);
+ READ_LOCATION_FIELD_OPT(location);
READ_DONE();
}
@@ -338,7 +406,7 @@ _readA_Const(void)
}
}
- READ_LOCATION_FIELD(location);
+ READ_LOCATION_FIELD_OPT(location);
READ_DONE();
}
@@ -352,9 +420,9 @@ _readConstraint(void)
READ_LOCALS(Constraint);
READ_STRING_FIELD(conname);
- READ_BOOL_FIELD(deferrable);
- READ_BOOL_FIELD(initdeferred);
- READ_LOCATION_FIELD(location);
+ READ_BOOL_FIELD_OPT(deferrable, false);
+ READ_BOOL_FIELD_OPT(initdeferred, false);
+ READ_LOCATION_FIELD_OPT(location);
token = pg_strtok(&length); /* skip :contype */
token = pg_strtok(&length); /* get field value */
@@ -395,10 +463,10 @@ _readConstraint(void)
case CONSTR_NOTNULL:
READ_NODE_FIELD(keys);
- READ_INT_FIELD(inhcount);
- READ_BOOL_FIELD(is_no_inherit);
- READ_BOOL_FIELD(skip_validation);
- READ_BOOL_FIELD(initially_valid);
+ READ_INT_FIELD_OPT(inhcount, 0);
+ READ_BOOL_FIELD_OPT(is_no_inherit, false);
+ READ_BOOL_FIELD_OPT(skip_validation, false);
+ READ_BOOL_FIELD_OPT(initially_valid, true);
break;
case CONSTR_DEFAULT:
@@ -407,7 +475,7 @@ _readConstraint(void)
break;
case CONSTR_IDENTITY:
- READ_NODE_FIELD(options);
+ READ_NODE_FIELD_OPT(options);
READ_CHAR_FIELD(generated_when);
break;
@@ -418,43 +486,43 @@ _readConstraint(void)
break;
case CONSTR_CHECK:
- READ_BOOL_FIELD(is_no_inherit);
+ READ_BOOL_FIELD_OPT(is_no_inherit, false);
READ_NODE_FIELD(raw_expr);
READ_STRING_FIELD(cooked_expr);
- READ_BOOL_FIELD(skip_validation);
- READ_BOOL_FIELD(initially_valid);
+ READ_BOOL_FIELD_OPT(skip_validation, false);
+ READ_BOOL_FIELD_OPT(initially_valid, true);
break;
case CONSTR_PRIMARY:
READ_NODE_FIELD(keys);
- READ_NODE_FIELD(including);
- READ_NODE_FIELD(options);
+ READ_NODE_FIELD_OPT(including);
+ READ_NODE_FIELD_OPT(options);
READ_STRING_FIELD(indexname);
- READ_STRING_FIELD(indexspace);
- READ_BOOL_FIELD(reset_default_tblspc);
+ READ_STRING_FIELD_NONNULL(indexspace);
+ READ_BOOL_FIELD_OPT(reset_default_tblspc, false);
/* access_method and where_clause not currently used */
break;
case CONSTR_UNIQUE:
- READ_BOOL_FIELD(nulls_not_distinct);
+ READ_BOOL_FIELD_OPT(nulls_not_distinct, true);
READ_NODE_FIELD(keys);
- READ_NODE_FIELD(including);
- READ_NODE_FIELD(options);
+ READ_NODE_FIELD_OPT(including);
+ READ_NODE_FIELD_OPT(options);
READ_STRING_FIELD(indexname);
- READ_STRING_FIELD(indexspace);
- READ_BOOL_FIELD(reset_default_tblspc);
+ READ_STRING_FIELD_NONNULL(indexspace);
+ READ_BOOL_FIELD_OPT(reset_default_tblspc, false);
/* access_method and where_clause not currently used */
break;
case CONSTR_EXCLUSION:
READ_NODE_FIELD(exclusions);
- READ_NODE_FIELD(including);
- READ_NODE_FIELD(options);
+ READ_NODE_FIELD_OPT(including);
+ READ_NODE_FIELD_OPT(options);
READ_STRING_FIELD(indexname);
- READ_STRING_FIELD(indexspace);
- READ_BOOL_FIELD(reset_default_tblspc);
+ READ_STRING_FIELD_NONNULL(indexspace);
+ READ_BOOL_FIELD_OPT(reset_default_tblspc, false);
READ_STRING_FIELD(access_method);
- READ_NODE_FIELD(where_clause);
+ READ_NODE_FIELD_OPT(where_clause);
break;
case CONSTR_FOREIGN:
@@ -467,8 +535,8 @@ _readConstraint(void)
READ_NODE_FIELD(fk_del_set_cols);
READ_NODE_FIELD(old_conpfeqop);
READ_OID_FIELD(old_pktable_oid);
- READ_BOOL_FIELD(skip_validation);
- READ_BOOL_FIELD(initially_valid);
+ READ_BOOL_FIELD_OPT(skip_validation, false);
+ READ_BOOL_FIELD_OPT(initially_valid, true);
break;
case CONSTR_ATTR_DEFERRABLE:
@@ -492,39 +560,39 @@ _readRangeTblEntry(void)
READ_LOCALS(RangeTblEntry);
/* put alias + eref first to make dump more legible */
- READ_NODE_FIELD(alias);
+ READ_NODE_FIELD_OPT(alias);
READ_NODE_FIELD(eref);
- READ_ENUM_FIELD(rtekind, RTEKind);
+ READ_ENUM_FIELD_OPT(rtekind, RTEKind, 0);
switch (local_node->rtekind)
{
case RTE_RELATION:
READ_OID_FIELD(relid);
- READ_CHAR_FIELD(relkind);
- READ_INT_FIELD(rellockmode);
- READ_NODE_FIELD(tablesample);
- READ_UINT_FIELD(perminfoindex);
+ READ_CHAR_FIELD_OPT(relkind, 'r');
+ READ_INT_FIELD_OPT(rellockmode, 1);
+ READ_NODE_FIELD_OPT(tablesample);
+ READ_UINT_FIELD_OPT(perminfoindex, 1);
break;
case RTE_SUBQUERY:
READ_NODE_FIELD(subquery);
- READ_BOOL_FIELD(security_barrier);
+ READ_BOOL_FIELD_OPT(security_barrier, false);
/* we re-use these RELATION fields, too: */
READ_OID_FIELD(relid);
- READ_CHAR_FIELD(relkind);
- READ_INT_FIELD(rellockmode);
- READ_UINT_FIELD(perminfoindex);
+ READ_CHAR_FIELD_OPT(relkind, 0);
+ READ_INT_FIELD_OPT(rellockmode, 0);
+ READ_UINT_FIELD_OPT(perminfoindex, 1);
break;
case RTE_JOIN:
- READ_ENUM_FIELD(jointype, JoinType);
- READ_INT_FIELD(joinmergedcols);
+ READ_ENUM_FIELD_OPT(jointype, JoinType, 0);
+ READ_INT_FIELD_OPT(joinmergedcols, 0);
READ_NODE_FIELD(joinaliasvars);
READ_NODE_FIELD(joinleftcols);
READ_NODE_FIELD(joinrightcols);
- READ_NODE_FIELD(join_using_alias);
+ READ_NODE_FIELD_OPT(join_using_alias);
break;
case RTE_FUNCTION:
READ_NODE_FIELD(functions);
- READ_BOOL_FIELD(funcordinality);
+ READ_BOOL_FIELD_OPT(funcordinality, false);
break;
case RTE_TABLEFUNC:
READ_NODE_FIELD(tablefunc);
@@ -547,7 +615,7 @@ _readRangeTblEntry(void)
case RTE_CTE:
READ_STRING_FIELD(ctename);
READ_UINT_FIELD(ctelevelsup);
- READ_BOOL_FIELD(self_reference);
+ READ_BOOL_FIELD_OPT(self_reference, false);
READ_NODE_FIELD(coltypes);
READ_NODE_FIELD(coltypmods);
READ_NODE_FIELD(colcollations);
@@ -570,10 +638,10 @@ _readRangeTblEntry(void)
break;
}
- READ_BOOL_FIELD(lateral);
- READ_BOOL_FIELD(inh);
- READ_BOOL_FIELD(inFromCl);
- READ_NODE_FIELD(securityQuals);
+ READ_BOOL_FIELD_OPT(lateral, false);
+ READ_BOOL_FIELD_OPT(inh, true);
+ READ_BOOL_FIELD_OPT(inFromCl, true);
+ READ_NODE_FIELD_OPT(securityQuals);
READ_DONE();
}
@@ -660,7 +728,7 @@ _readA_Expr(void)
READ_NODE_FIELD(lexpr);
READ_NODE_FIELD(rexpr);
- READ_LOCATION_FIELD(location);
+ READ_LOCATION_FIELD_OPT(location);
READ_DONE();
}
@@ -768,16 +836,29 @@ readDatum(bool typbyval)
s = (char *) palloc(length);
for (i = 0; i < length; i++)
{
+ if (pg_strtoken_next("]"))
+ break;
+
token = pg_strtok(&tokenLength);
+
+ Assert(token[0] != ']');
+
s[i] = (char) atoi(token);
}
+
+ for (; i < length; i++)
+ s[i] = 0;
+
res = PointerGetDatum(s);
}
token = pg_strtok(&tokenLength); /* read the ']' */
if (token == NULL || token[0] != ']')
+ {
+ Assert(false);
elog(ERROR, "expected \"]\" to end datum, but got \"%s\"; length = %zu",
token ? token : "[NULL]", length);
+ }
return res;
}
diff --git a/src/backend/rewrite/rewriteDefine.c b/src/backend/rewrite/rewriteDefine.c
index e36fc72e1e..5ab0ec089e 100644
--- a/src/backend/rewrite/rewriteDefine.c
+++ b/src/backend/rewrite/rewriteDefine.c
@@ -219,6 +219,98 @@ DefineRule(RuleStmt *stmt, const char *queryString)
actions);
}
+/*
+ * Walker that clears 'location'-related fields to -1.
+ *
+ * This is used for reducing the size of pg_rewrite entries exclusively.
+ */
+static bool
+strip_location_walker(Node *node, void *context)
+{
+ const int query_recurse_flags = 0;
+ ListCell *temp;
+
+#define RESET_FOR(_type_) \
+ case T_##_type_: \
+ { \
+ castNode(_type_, node)->location = -1; \
+ break; \
+ }
+
+ if (node == NULL)
+ return false;
+
+ if (IsA(node, Query))
+ {
+ Query *query = castNode(Query, node);
+ query->stmt_location = -1;
+ query->stmt_len = 0;
+
+ return query_tree_walker(query, strip_location_walker, context, query_recurse_flags);
+ }
+
+ switch (nodeTag(node))
+ {
+ RESET_FOR(RangeVar);
+ RESET_FOR(TableFunc);
+ RESET_FOR(Var);
+ RESET_FOR(Const);
+ RESET_FOR(Param);
+ RESET_FOR(Aggref);
+ RESET_FOR(GroupingFunc);
+ RESET_FOR(WindowFunc);
+ RESET_FOR(FuncExpr);
+ RESET_FOR(NamedArgExpr);
+ RESET_FOR(OpExpr);
+ RESET_FOR(DistinctExpr);
+ RESET_FOR(NullIfExpr);
+ RESET_FOR(ScalarArrayOpExpr);
+ RESET_FOR(BoolExpr);
+ RESET_FOR(SubLink);
+ RESET_FOR(RelabelType);
+ RESET_FOR(CoerceViaIO);
+ RESET_FOR(ArrayCoerceExpr);
+ RESET_FOR(ConvertRowtypeExpr);
+ RESET_FOR(CollateExpr);
+ RESET_FOR(CaseWhen);
+ RESET_FOR(ArrayExpr);
+ RESET_FOR(RowExpr);
+ RESET_FOR(CoalesceExpr);
+ RESET_FOR(MinMaxExpr);
+ RESET_FOR(SQLValueFunction);
+ RESET_FOR(XmlExpr);
+ RESET_FOR(JsonFormat);
+ RESET_FOR(JsonConstructorExpr);
+ RESET_FOR(JsonIsPredicate);
+ RESET_FOR(NullTest);
+ RESET_FOR(BooleanTest);
+ RESET_FOR(CoerceToDomain);
+ RESET_FOR(CoerceToDomainValue);
+ RESET_FOR(SetToDefault);
+ case T_CaseExpr:
+ {
+ /*
+ * The expression_tree_walker does not call us for CaseWhen nodes,
+ * but instead directly wires us through to its inner Nodes.
+ * To correctly reset the location fields, we manually iterate
+ * to get those fields reset.
+ */
+ CaseExpr *caseExpr = castNode(CaseExpr, node);
+ caseExpr->location = -1;
+ foreach(temp, caseExpr->args)
+ {
+ CaseWhen *when = lfirst_node(CaseWhen, temp);
+ when->location = -1;
+ }
+ break;
+ }
+ default:
+ break;
+ }
+
+ return expression_tree_walker(node, strip_location_walker, (void *) context);
+#undef RESET_FOR
+}
/*
* DefineQueryRewrite
@@ -474,6 +566,16 @@ DefineQueryRewrite(const char *rulename,
/* discard rule if it's null action and not INSTEAD; it's a no-op */
if (action != NIL || is_instead)
{
+ /*
+ * Clear location info from the statement. Because we don't store
+ * the original query, maintaining this information is meaningless.
+ */
+ foreach(l, action)
+ {
+ query = lfirst_node(Query, l);
+ strip_location_walker((Node *) query, NULL);
+ }
+
ruleId = InsertRule(rulename,
event_type,
event_relid,
diff --git a/src/include/nodes/parsenodes.h b/src/include/nodes/parsenodes.h
index e494309da8..b853c78552 100644
--- a/src/include/nodes/parsenodes.h
+++ b/src/include/nodes/parsenodes.h
@@ -117,7 +117,8 @@ typedef struct Query
{
NodeTag type;
- CmdType commandType; /* select|insert|update|delete|merge|utility */
+ /* select|insert|update|delete|merge|utility */
+ CmdType commandType pg_node_attr(default(1));
/* where did I come from? */
QuerySource querySource pg_node_attr(query_jumble_ignore);
@@ -130,7 +131,7 @@ typedef struct Query
uint64 queryId pg_node_attr(equal_ignore, query_jumble_ignore, read_write_ignore, read_as(0));
/* do I set the command result tag? */
- bool canSetTag pg_node_attr(query_jumble_ignore);
+ bool canSetTag pg_node_attr(query_jumble_ignore, default(true));
Node *utilityStmt; /* non-null if commandType == CMD_UTILITY */
@@ -1234,8 +1235,10 @@ typedef struct RTEPermissionInfo
NodeTag type;
Oid relid; /* relation OID */
- bool inh; /* separately check inheritance children? */
- AclMode requiredPerms; /* bitmask of required access permissions */
+ /* separately check inheritance children? */
+ bool inh pg_node_attr(default(true));
+ /* bitmask of required access permissions */
+ AclMode requiredPerms pg_node_attr(default(2));
Oid checkAsUser; /* if valid, check access as this role */
Bitmapset *selectedCols; /* columns needing SELECT permission */
Bitmapset *insertedCols; /* columns needing INSERT permission */
diff --git a/src/include/nodes/primnodes.h b/src/include/nodes/primnodes.h
index bb930afb52..e2604edb03 100644
--- a/src/include/nodes/primnodes.h
+++ b/src/include/nodes/primnodes.h
@@ -82,10 +82,10 @@ typedef struct RangeVar
char *relname;
/* expand rel by inheritance? recursively act on children? */
- bool inh;
+ bool inh pg_node_attr(default(true));
/* see RELPERSISTENCE_* in pg_class.h */
- char relpersistence;
+ char relpersistence pg_node_attr(default('p'));
/* table alias & optional column aliases */
Alias *alias;
@@ -126,7 +126,7 @@ typedef struct TableFunc
/* nullability flag for each output column */
Bitmapset *notnulls pg_node_attr(query_jumble_ignore);
/* counts from 0; -1 if none specified */
- int ordinalitycol pg_node_attr(query_jumble_ignore);
+ int ordinalitycol pg_node_attr(query_jumble_ignore, default(-1));
/* token location, or -1 if unknown */
int location;
} TableFunc;
@@ -238,7 +238,7 @@ typedef struct Var
* index of this var's relation in the range table, or
* INNER_VAR/OUTER_VAR/etc
*/
- int varno;
+ int varno pg_node_attr(default(1));
/*
* attribute number of this var, or zero for all attrs ("whole-row Var")
@@ -248,7 +248,7 @@ typedef struct Var
/* pg_type OID for the type of this var */
Oid vartype pg_node_attr(query_jumble_ignore);
/* pg_attribute typmod value */
- int32 vartypmod pg_node_attr(query_jumble_ignore);
+ int32 vartypmod pg_node_attr(query_jumble_ignore, default(-1));
/* OID of collation, or InvalidOid if none */
Oid varcollid pg_node_attr(query_jumble_ignore);
@@ -271,9 +271,9 @@ typedef struct Var
* their varno/varattno match.
*/
/* syntactic relation index (0 if unknown) */
- Index varnosyn pg_node_attr(equal_ignore, query_jumble_ignore);
+ Index varnosyn pg_node_attr(equal_ignore, query_jumble_ignore, default_ref(varno));
/* syntactic attribute number */
- AttrNumber varattnosyn pg_node_attr(equal_ignore, query_jumble_ignore);
+ AttrNumber varattnosyn pg_node_attr(equal_ignore, query_jumble_ignore, default_ref(varattno));
/* token location, or -1 if unknown */
int location;
@@ -297,7 +297,7 @@ typedef struct Const
/* pg_type OID of the constant's datatype */
Oid consttype;
/* typmod value, if any */
- int32 consttypmod pg_node_attr(query_jumble_ignore);
+ int32 consttypmod pg_node_attr(query_jumble_ignore, default(-1));
/* OID of collation, or InvalidOid if none */
Oid constcollid pg_node_attr(query_jumble_ignore);
/* typlen of the constant's datatype */
@@ -363,7 +363,7 @@ typedef struct Param
int paramid; /* numeric ID for parameter */
Oid paramtype; /* pg_type OID of parameter's datatype */
/* typmod value, if known */
- int32 paramtypmod pg_node_attr(query_jumble_ignore);
+ int32 paramtypmod pg_node_attr(query_jumble_ignore, default(-1));
/* OID of collation, or InvalidOid if none */
Oid paramcollid pg_node_attr(query_jumble_ignore);
/* token location, or -1 if unknown */
@@ -621,7 +621,7 @@ typedef struct SubscriptingRef
/* type of the SubscriptingRef's result */
Oid refrestype pg_node_attr(query_jumble_ignore);
/* typmod of the result */
- int32 reftypmod pg_node_attr(query_jumble_ignore);
+ int32 reftypmod pg_node_attr(query_jumble_ignore, default(-1));
/* collation of result, or InvalidOid if none */
Oid refcollid pg_node_attr(query_jumble_ignore);
/* expressions that evaluate to upper container indexes */
@@ -1065,7 +1065,7 @@ typedef struct FieldSelect
/* type of the field (result type of this node) */
Oid resulttype pg_node_attr(query_jumble_ignore);
/* output typmod (usually -1) */
- int32 resulttypmod pg_node_attr(query_jumble_ignore);
+ int32 resulttypmod pg_node_attr(query_jumble_ignore, default(-1));
/* OID of collation of the field */
Oid resultcollid pg_node_attr(query_jumble_ignore);
} FieldSelect;
@@ -1119,7 +1119,7 @@ typedef struct RelabelType
Expr *arg; /* input expression */
Oid resulttype; /* output type of coercion expression */
/* output typmod (usually -1) */
- int32 resulttypmod pg_node_attr(query_jumble_ignore);
+ int32 resulttypmod pg_node_attr(query_jumble_ignore, default(-1));
/* OID of collation, or InvalidOid if none */
Oid resultcollid pg_node_attr(query_jumble_ignore);
/* how to display this node */
@@ -1169,7 +1169,7 @@ typedef struct ArrayCoerceExpr
Expr *elemexpr; /* expression representing per-element work */
Oid resulttype; /* output type of coercion (an array type) */
/* output typmod (also element typmod) */
- int32 resulttypmod pg_node_attr(query_jumble_ignore);
+ int32 resulttypmod pg_node_attr(query_jumble_ignore, default(-1));
/* OID of collation, or InvalidOid if none */
Oid resultcollid pg_node_attr(query_jumble_ignore);
/* how to display this node */
@@ -1495,7 +1495,7 @@ typedef struct SQLValueFunction
* include this Oid in the query jumbling.
*/
Oid type pg_node_attr(query_jumble_ignore);
- int32 typmod;
+ int32 typmod pg_node_attr(default(-1));
int location; /* token location, or -1 if unknown */
} SQLValueFunction;
@@ -1547,7 +1547,7 @@ typedef struct XmlExpr
bool indent;
/* target type/typmod for XMLSERIALIZE */
Oid type pg_node_attr(query_jumble_ignore);
- int32 typmod pg_node_attr(query_jumble_ignore);
+ int32 typmod pg_node_attr(query_jumble_ignore, default(-1));
/* token location, or -1 if unknown */
int location;
} XmlExpr;
@@ -1597,7 +1597,7 @@ typedef struct JsonReturning
NodeTag type;
JsonFormat *format; /* output JSON format */
Oid typid; /* target type Oid */
- int32 typmod; /* target type modifier */
+ int32 typmod pg_node_attr(default(-1)); /* target type modifier */
} JsonReturning;
/*
@@ -1760,11 +1760,11 @@ typedef struct CoerceToDomain
Expr *arg; /* input expression */
Oid resulttype; /* domain type ID (result type) */
/* output typmod (currently always -1) */
- int32 resulttypmod pg_node_attr(query_jumble_ignore);
+ int32 resulttypmod pg_node_attr(query_jumble_ignore, default(-1));
/* OID of collation, or InvalidOid if none */
Oid resultcollid pg_node_attr(query_jumble_ignore);
/* how to display this node */
- CoercionForm coercionformat pg_node_attr(query_jumble_ignore);
+ CoercionForm coercionformat pg_node_attr(query_jumble_ignore, default(1));
int location; /* token location, or -1 if unknown */
} CoerceToDomain;
@@ -1929,7 +1929,7 @@ typedef struct TargetEntry
/* OID of column's source table */
Oid resorigtbl pg_node_attr(query_jumble_ignore);
/* column's number in source table */
- AttrNumber resorigcol pg_node_attr(query_jumble_ignore);
+ AttrNumber resorigcol pg_node_attr(query_jumble_ignore, default(0));
/* set to true to eliminate the attribute from final target list */
bool resjunk pg_node_attr(query_jumble_ignore);
} TargetEntry;
@@ -1975,7 +1975,7 @@ typedef struct TargetEntry
typedef struct RangeTblRef
{
NodeTag type;
- int rtindex;
+ int rtindex pg_node_attr(default(1));
} RangeTblRef;
/*----------
diff --git a/src/include/nodes/readfuncs.h b/src/include/nodes/readfuncs.h
index cba6f0be75..e668b24ad0 100644
--- a/src/include/nodes/readfuncs.h
+++ b/src/include/nodes/readfuncs.h
@@ -27,6 +27,7 @@ extern PGDLLIMPORT bool restore_location_fields;
* prototypes for functions in read.c (the lisp token parser)
*/
extern const char *pg_strtok(int *length);
+extern bool pg_strtoken_next(const char *expect_token);
extern char *debackslash(const char *token, int length);
extern void *nodeRead(const char *token, int tok_len);
diff --git a/src/test/regress/expected/rowsecurity.out b/src/test/regress/expected/rowsecurity.out
index 6988128aa4..942bce62a9 100644
--- a/src/test/regress/expected/rowsecurity.out
+++ b/src/test/regress/expected/rowsecurity.out
@@ -3937,7 +3937,10 @@ CREATE TABLE coll_t (c) AS VALUES ('bar'::text);
CREATE POLICY coll_p ON coll_t USING (c < ('foo'::text COLLATE "C"));
ALTER TABLE coll_t ENABLE ROW LEVEL SECURITY;
GRANT SELECT ON coll_t TO regress_rls_alice;
-SELECT (string_to_array(polqual, ':'))[7] AS inputcollid FROM pg_policy WHERE polrelid = 'coll_t'::regclass;
+SELECT split inputcollid
+FROM pg_policy,
+ lateral unnest(string_to_array(polqual, ':')) as split
+WHERE polrelid = 'coll_t'::regclass and split LIKE '%inputcollid%';
inputcollid
------------------
inputcollid 950
diff --git a/src/test/regress/sql/rowsecurity.sql b/src/test/regress/sql/rowsecurity.sql
index dec7340538..9c6a6396e2 100644
--- a/src/test/regress/sql/rowsecurity.sql
+++ b/src/test/regress/sql/rowsecurity.sql
@@ -1732,7 +1732,10 @@ CREATE TABLE coll_t (c) AS VALUES ('bar'::text);
CREATE POLICY coll_p ON coll_t USING (c < ('foo'::text COLLATE "C"));
ALTER TABLE coll_t ENABLE ROW LEVEL SECURITY;
GRANT SELECT ON coll_t TO regress_rls_alice;
-SELECT (string_to_array(polqual, ':'))[7] AS inputcollid FROM pg_policy WHERE polrelid = 'coll_t'::regclass;
+SELECT split inputcollid
+FROM pg_policy,
+ lateral unnest(string_to_array(polqual, ':')) as split
+WHERE polrelid = 'coll_t'::regclass and split LIKE '%inputcollid%';
SET SESSION AUTHORIZATION regress_rls_alice;
SELECT * FROM coll_t;
ROLLBACK;
--
2.40.1
^ permalink raw reply [nested|flat] 70+ messages in thread
* Re: Reducing output size of nodeToString
2023-12-06 21:08 Reducing output size of nodeToString Matthias van de Meent <[email protected]>
@ 2023-12-07 12:09 ` David Rowley <[email protected]>
1 sibling, 0 replies; 70+ messages in thread
From: David Rowley @ 2023-12-07 12:09 UTC (permalink / raw)
To: Matthias van de Meent <[email protected]>; +Cc: PostgreSQL Hackers <[email protected]>; Peter Eisentraut <[email protected]>; Andres Freund <[email protected]>; Michel Pelletier <[email protected]>
On Thu, 7 Dec 2023 at 10:09, Matthias van de Meent
<[email protected]> wrote:
> PFA a patch that reduces the output size of nodeToString by 50%+ in
> most cases (measured on pg_rewrite), which on my system reduces the
> total size of pg_rewrite by 33% to 472KiB. This does keep the textual
> pg_node_tree format alive, but reduces its size significantly.
It would be very cool to have the technology proposed by Andres back
in 2019 [1]. With that, we could easily write various output
functions. One could be compact and easily machine-readable and
another designed to be better for humans for debugging purposes.
We could also easily serialize plans to binary format for copying to
parallel workers rather than converting them to a text-based
serialized format. It would also allow us to do things like serialize
PREPAREd plans into a nicely compact single allocation that we could
just pfree in a single pfree call on DEALLOCATE.
Likely we could just use the existing Perl scripts to form the
metadata arrays rather than the clang parsing stuff Andres used in his
patch.
Anyway, just wanted to ensure you knew about this idea.
David
[1] https://postgr.es/m/flat/20190828234136.fk2ndqtld3onfrrp%40alap3.anarazel.de
^ permalink raw reply [nested|flat] 70+ messages in thread
* Re: Reducing output size of nodeToString
2023-12-06 21:08 Reducing output size of nodeToString Matthias van de Meent <[email protected]>
@ 2024-01-02 10:30 ` Peter Eisentraut <[email protected]>
2024-01-03 23:23 ` Re: Reducing output size of nodeToString Matthias van de Meent <[email protected]>
1 sibling, 1 reply; 70+ messages in thread
From: Peter Eisentraut @ 2024-01-02 10:30 UTC (permalink / raw)
To: Matthias van de Meent <[email protected]>; PostgreSQL Hackers <[email protected]>; Andres Freund <[email protected]>; +Cc: Michel Pelletier <[email protected]>
On 06.12.23 22:08, Matthias van de Meent wrote:
> PFA a patch that reduces the output size of nodeToString by 50%+ in
> most cases (measured on pg_rewrite), which on my system reduces the
> total size of pg_rewrite by 33% to 472KiB. This does keep the textual
> pg_node_tree format alive, but reduces its size signficantly.
>
> The basic techniques used are
> - Don't emit scalar fields when they contain a default value, and
> make the reading code aware of this.
> - Reasonable defaults are set for most datatypes, and overrides can
> be added with new pg_node_attr() attributes. No introspection into
> non-null Node/Array/etc. is being done though.
> - Reset more fields to their default values before storing the values.
> - Don't write trailing 0s in outDatum calls for by-ref types. This
> saves many bytes for Name fields, but also some other pre-existing
> entry points.
Based on our discussions, my understanding is that you wanted to produce
an updated patch set that is split up a bit.
My suggestion is to make incremental patches along these lines:
- Omit from output all fields that have value zero.
- Omit location fields that have value -1.
- Omit trailing zeroes for scalar values.
- Recent location fields before storing in pg_rewrite (or possibly
catalogs in general?)
- And then whatever is left, including the "default" value system that
you have proposed.
The last one I have some doubts about, as previously expressed, but the
first few seem sensible to me. By splitting it up we can consider these
incrementally.
^ permalink raw reply [nested|flat] 70+ messages in thread
* Re: Reducing output size of nodeToString
2023-12-06 21:08 Reducing output size of nodeToString Matthias van de Meent <[email protected]>
2024-01-02 10:30 ` Re: Reducing output size of nodeToString Peter Eisentraut <[email protected]>
@ 2024-01-03 23:23 ` Matthias van de Meent <[email protected]>
0 siblings, 0 replies; 70+ messages in thread
From: Matthias van de Meent @ 2024-01-03 23:23 UTC (permalink / raw)
To: Peter Eisentraut <[email protected]>; +Cc: PostgreSQL Hackers <[email protected]>; Andres Freund <[email protected]>; Michel Pelletier <[email protected]>
On Tue, 2 Jan 2024 at 11:30, Peter Eisentraut <[email protected]> wrote:
>
> On 06.12.23 22:08, Matthias van de Meent wrote:
> > PFA a patch that reduces the output size of nodeToString by 50%+ in
> > most cases (measured on pg_rewrite), which on my system reduces the
> > total size of pg_rewrite by 33% to 472KiB. This does keep the textual
> > pg_node_tree format alive, but reduces its size signficantly.
> >
> > The basic techniques used are
> > - Don't emit scalar fields when they contain a default value, and
> > make the reading code aware of this.
> > - Reasonable defaults are set for most datatypes, and overrides can
> > be added with new pg_node_attr() attributes. No introspection into
> > non-null Node/Array/etc. is being done though.
> > - Reset more fields to their default values before storing the values.
> > - Don't write trailing 0s in outDatum calls for by-ref types. This
> > saves many bytes for Name fields, but also some other pre-existing
> > entry points.
>
> Based on our discussions, my understanding is that you wanted to produce
> an updated patch set that is split up a bit.
I mentioned that I've been working on implementing (but have not yet
completed) a binary serialization format, with an implementation based
on Andres' generated metadata idea. However, that requires more
elaborate infrastructure than is currently available, so while I said
I'd expected it to be complete before the Christmas weekend, it'll
take some more time - I'm not sure it'll be ready for PG17.
In the meantime here's an updated version of the v0 patch, formally
keeping the textual format alive, while reducing the size
significantly (nearing 2/3 reduction), taking your comments into
account. I think the gains are worth the consideration without taking
into account the as-of-yet unimplemented binary format.
> My suggestion is to make incremental patches along these lines:
> [...]
Something like the attached? It splits out into the following
0001: basic 'omit default values'
0002: reset location and other querystring-related node fields for all
catalogs of type pg_node_tree.
0003: add default marking on typmod fields.
0004 & 0006: various node fields marked with default() based on
observed common or initial values of those fields
0005: truncate trailing 0s from outDatum
0007 (new): do run-length + gap coding for bitmapset and the various
integer list types. This saves a surprising amount of bytes.
> The last one I have some doubts about, as previously expressed, but the
> first few seem sensible to me. By splitting it up we can consider these
> incrementally.
That makes a lot of sense. The numbers for the full patchset do seem
quite positive though: The metrics of the query below show a 40%
decrease in size of a fresh pg_rewrite (standard toast compression)
and a 5% decrease in size of the template0 database. The uncompressed
data of pg_rewrite.ev_action is also 60% smaller.
select pg_database_size('template0') as "template0"
, pg_total_relation_size('pg_rewrite') as "pg_rewrite"
, sum(pg_column_size(ev_action)) as "compressed"
, sum(octet_length(ev_action)) as "raw"
from pg_rewrite;
version | template0 | pg_rewrite | compressed | raw
---------|-----------+------------+------------+---------
master | 7545359 | 761856 | 573307 | 2998712
0001 | 7365135 | 622592 | 438224 | 1943772
0002 | 7258639 | 573440 | 401660 | 1835803
0003 | 7258639 | 565248 | 386211 | 1672539
0004 | 7176719 | 483328 | 317099 | 1316552
0005 | 7176719 | 483328 | 315556 | 1300420
0006 | 7160335 | 466944 | 302806 | 1208621
0007 | 7143951 | 450560 | 287659 | 1187237
While looking through the data, I noticed the larger views now consist
for a significant portion out of range table entries, specifically the
Alias and Var nodes (which are mostly repeated and/or repetative
values, but split across Nodes). I think column-major storage would be
more efficient to write, but I'm not sure it's worth the effort in
planner code.
Kind regards,
Matthias van de Meent
Neon (https://neon.tech)
Attachments:
[application/octet-stream] v1-0001-pg_node_tree-Don-t-serialize-fields-with-type-def.patch (22.8K, ../../CAEze2Wigkd1+J4s=7wUqW8Y4g9mDWSC28119ukbKkf799WBpzg@mail.gmail.com/2-v1-0001-pg_node_tree-Don-t-serialize-fields-with-type-def.patch)
download | inline diff:
From b597cbb73e4326e2707114ea0ffa2d5a0bc68465 Mon Sep 17 00:00:00 2001
From: Matthias van de Meent <[email protected]>
Date: Tue, 2 Jan 2024 23:55:02 +0100
Subject: [PATCH v1 1/7] pg_node_tree: Don't serialize fields with type-default
values.
Often, the values in nodes are their default values. By not serializing
those fields and inserting the defaults during deserialization, we
reduce the size of pg_node_tree attributes seen in e.g. pg_rewrite by a
significant factor.
In passing, we fix a test that had a strict dependency on the
serialization of pg_node_tree; we now do the checks in a more generic
manner, making it more stable and ensuring its stability in future work.
---
src/backend/nodes/outfuncs.c | 101 +++++++++++++----
src/backend/nodes/read.c | 45 ++++++++
src/backend/nodes/readfuncs.c | 129 +++++++++++++++++++---
src/include/nodes/readfuncs.h | 1 +
src/test/regress/expected/rowsecurity.out | 5 +-
src/test/regress/sql/rowsecurity.sql | 5 +-
6 files changed, 247 insertions(+), 39 deletions(-)
diff --git a/src/backend/nodes/outfuncs.c b/src/backend/nodes/outfuncs.c
index e66a99247e..29f4e43581 100644
--- a/src/backend/nodes/outfuncs.c
+++ b/src/backend/nodes/outfuncs.c
@@ -41,94 +41,157 @@ static void outDouble(StringInfo str, double d);
appendStringInfoString(str, nodelabel)
/* Write an integer field (anything written as ":fldname %d") */
-#define WRITE_INT_FIELD(fldname) \
+#define WRITE_INT_FIELD_DIRECT(fldname) \
appendStringInfo(str, " :" CppAsString(fldname) " %d", node->fldname)
+#define WRITE_INT_FIELD_DEFAULT(fldname, default) \
+ ((node->fldname == default) ? (0) : WRITE_INT_FIELD_DIRECT(fldname))
+#define WRITE_INT_FIELD(fldname) \
+ WRITE_INT_FIELD_DEFAULT(fldname, 0)
/* Write an unsigned integer field (anything written as ":fldname %u") */
-#define WRITE_UINT_FIELD(fldname) \
+#define WRITE_UINT_FIELD_DIRECT(fldname) \
appendStringInfo(str, " :" CppAsString(fldname) " %u", node->fldname)
+#define WRITE_UINT_FIELD_DEFAULT(fldname, default) \
+ ((node->fldname == default) ? (0) : WRITE_UINT_FIELD_DIRECT(fldname))
+#define WRITE_UINT_FIELD(fldname) \
+ WRITE_UINT_FIELD_DEFAULT(fldname, 0)
/* Write an unsigned integer field (anything written with UINT64_FORMAT) */
-#define WRITE_UINT64_FIELD(fldname) \
+#define WRITE_UINT64_FIELD_DIRECT(fldname) \
appendStringInfo(str, " :" CppAsString(fldname) " " UINT64_FORMAT, \
node->fldname)
+#define WRITE_UINT64_FIELD_DEFAULT(fldname, default) \
+ ((node->fldname == default) ? (0) : WRITE_UINT64_FIELD_DIRECT(fldname))
+#define WRITE_UINT64_FIELD(fldname) \
+ WRITE_UINT64_FIELD_DEFAULT(fldname, 0)
/* Write an OID field (don't hard-wire assumption that OID is same as uint) */
-#define WRITE_OID_FIELD(fldname) \
+#define WRITE_OID_FIELD_DIRECT(fldname) \
appendStringInfo(str, " :" CppAsString(fldname) " %u", node->fldname)
+#define WRITE_OID_FIELD_DEFAULT(fldname, default) \
+ ((node->fldname == default) ? (0) : WRITE_OID_FIELD_DIRECT(fldname))
+#define WRITE_OID_FIELD(fldname) \
+ WRITE_OID_FIELD_DEFAULT(fldname, 0)
/* Write a long-integer field */
-#define WRITE_LONG_FIELD(fldname) \
+#define WRITE_LONG_FIELD_DIRECT(fldname) \
appendStringInfo(str, " :" CppAsString(fldname) " %ld", node->fldname)
+#define WRITE_LONG_FIELD_DEFAULT(fldname, default) \
+ ((node->fldname == default) ? (0) : WRITE_LONG_FIELD_DIRECT(fldname))
+#define WRITE_LONG_FIELD(fldname) \
+ WRITE_LONG_FIELD_DEFAULT(fldname, 0)
+
/* Write a char field (ie, one ascii character) */
-#define WRITE_CHAR_FIELD(fldname) \
+#define WRITE_CHAR_FIELD_DIRECT(fldname) \
(appendStringInfo(str, " :" CppAsString(fldname) " "), \
outChar(str, node->fldname))
+#define WRITE_CHAR_FIELD_DEFAULT(fldname, default) \
+ ((node->fldname == default) ? (0) : WRITE_CHAR_FIELD_DIRECT(fldname))
+#define WRITE_CHAR_FIELD(fldname) \
+ WRITE_CHAR_FIELD_DEFAULT(fldname, '\0')
/* Write an enumerated-type field as an integer code */
-#define WRITE_ENUM_FIELD(fldname, enumtype) \
+#define WRITE_ENUM_FIELD_DIRECT(fldname, enumtype) \
appendStringInfo(str, " :" CppAsString(fldname) " %d", \
(int) node->fldname)
+#define WRITE_ENUM_FIELD_DEFAULT(fldname, enumtype, default) \
+ ((node->fldname == default) ? (0) : WRITE_ENUM_FIELD_DIRECT(fldname, enumtype))
+#define WRITE_ENUM_FIELD(fldname, enumtype) \
+ WRITE_ENUM_FIELD_DEFAULT(fldname, enumtype, 0)
/* Write a float field (actually, they're double) */
-#define WRITE_FLOAT_FIELD(fldname) \
+#define WRITE_FLOAT_FIELD_DIRECT(fldname) \
(appendStringInfo(str, " :" CppAsString(fldname) " "), \
outDouble(str, node->fldname))
+#define WRITE_FLOAT_FIELD_DEFAULT(fldname, default) \
+ ((node->fldname == default) ? (0) : WRITE_FLOAT_FIELD_DIRECT(fldname))
+#define WRITE_FLOAT_FIELD(fldname) \
+ WRITE_FLOAT_FIELD_DEFAULT(fldname, 0.0)
/* Write a boolean field */
-#define WRITE_BOOL_FIELD(fldname) \
+#define WRITE_BOOL_FIELD_DIRECT(fldname) \
appendStringInfo(str, " :" CppAsString(fldname) " %s", \
booltostr(node->fldname))
+#define WRITE_BOOL_FIELD_DEFAULT(fldname, default) \
+ ((node->fldname == default) ? (0) : WRITE_BOOL_FIELD_DIRECT(fldname))
+#define WRITE_BOOL_FIELD(fldname) \
+ WRITE_BOOL_FIELD_DEFAULT(fldname, false)
+
+/*
+ * Non-null defaults of by-ref types are exceedingly rare (if not generally
+ * nonexistent), so we don't (yet) have a specialized macro for non-NULL
+ * defaults omission.
+ */
/* Write a character-string (possibly NULL) field */
-#define WRITE_STRING_FIELD(fldname) \
+#define WRITE_STRING_FIELD_DIRECT(fldname) \
(appendStringInfoString(str, " :" CppAsString(fldname) " "), \
outToken(str, node->fldname))
+#define WRITE_STRING_FIELD(fldname) \
+ ((node->fldname == NULL) ? (0) : WRITE_STRING_FIELD_DIRECT(fldname))
/* Write a parse location field (actually same as INT case) */
-#define WRITE_LOCATION_FIELD(fldname) \
+#define WRITE_LOCATION_FIELD_DIRECT(fldname) \
appendStringInfo(str, " :" CppAsString(fldname) " %d", node->fldname)
+#define WRITE_LOCATION_FIELD(fldname) \
+ (node->fldname == -1 ? (0) : WRITE_LOCATION_FIELD_DIRECT(fldname))
/* Write a Node field */
-#define WRITE_NODE_FIELD(fldname) \
+#define WRITE_NODE_FIELD_DIRECT(fldname) \
(appendStringInfoString(str, " :" CppAsString(fldname) " "), \
outNode(str, node->fldname))
+#define WRITE_NODE_FIELD(fldname) \
+ (node->fldname == NULL ? (0) : WRITE_NODE_FIELD_DIRECT(fldname))
/* Write a bitmapset field */
-#define WRITE_BITMAPSET_FIELD(fldname) \
+#define WRITE_BITMAPSET_FIELD_DIRECT(fldname) \
(appendStringInfoString(str, " :" CppAsString(fldname) " "), \
outBitmapset(str, node->fldname))
+#define WRITE_BITMAPSET_FIELD(fldname) \
+ (node->fldname == NULL ? (0) : WRITE_BITMAPSET_FIELD_DIRECT(fldname))
/* Write a variable-length array (not a List) of Node pointers */
-#define WRITE_NODE_ARRAY(fldname, len) \
+#define WRITE_NODE_ARRAY_DIRECT(fldname, len) \
(appendStringInfoString(str, " :" CppAsString(fldname) " "), \
writeNodeArray(str, (const Node * const *) node->fldname, len))
+#define WRITE_NODE_ARRAY(fldname, len) \
+ (node->fldname == NULL ? (0) : WRITE_NODE_ARRAY_DIRECT(fldname, len))
/* Write a variable-length array of AttrNumber */
-#define WRITE_ATTRNUMBER_ARRAY(fldname, len) \
+#define WRITE_ATTRNUMBER_ARRAY_DIRECT(fldname, len) \
(appendStringInfoString(str, " :" CppAsString(fldname) " "), \
writeAttrNumberCols(str, node->fldname, len))
+#define WRITE_ATTRNUMBER_ARRAY(fldname, len) \
+ (node->fldname == NULL ? (0) : WRITE_ATTRNUMBER_ARRAY_DIRECT(fldname, len))
/* Write a variable-length array of Oid */
-#define WRITE_OID_ARRAY(fldname, len) \
+#define WRITE_OID_ARRAY_DIRECT(fldname, len) \
(appendStringInfoString(str, " :" CppAsString(fldname) " "), \
writeOidCols(str, node->fldname, len))
+#define WRITE_OID_ARRAY(fldname, len) \
+ (node->fldname == NULL ? (0) : WRITE_OID_ARRAY_DIRECT(fldname, len))
/* Write a variable-length array of Index */
-#define WRITE_INDEX_ARRAY(fldname, len) \
+#define WRITE_INDEX_ARRAY_DIRECT(fldname, len) \
(appendStringInfoString(str, " :" CppAsString(fldname) " "), \
writeIndexCols(str, node->fldname, len))
+#define WRITE_INDEX_ARRAY(fldname, len) \
+ (node->fldname == NULL ? (0) : WRITE_INDEX_ARRAY_DIRECT(fldname, len))
/* Write a variable-length array of int */
-#define WRITE_INT_ARRAY(fldname, len) \
+#define WRITE_INT_ARRAY_DIRECT(fldname, len) \
(appendStringInfoString(str, " :" CppAsString(fldname) " "), \
writeIntCols(str, node->fldname, len))
+#define WRITE_INT_ARRAY(fldname, len) \
+ (node->fldname == NULL ? (0) : WRITE_INT_ARRAY_DIRECT(fldname, len))
/* Write a variable-length array of bool */
-#define WRITE_BOOL_ARRAY(fldname, len) \
+#define WRITE_BOOL_ARRAY_DIRECT(fldname, len) \
(appendStringInfoString(str, " :" CppAsString(fldname) " "), \
writeBoolCols(str, node->fldname, len))
+#define WRITE_BOOL_ARRAY(fldname, len) \
+ (node->fldname == NULL ? (0) : WRITE_BOOL_ARRAY_DIRECT(fldname, len))
#define booltostr(x) ((x) ? "true" : "false")
diff --git a/src/backend/nodes/read.c b/src/backend/nodes/read.c
index 813eda3e73..2815b268be 100644
--- a/src/backend/nodes/read.c
+++ b/src/backend/nodes/read.c
@@ -205,6 +205,51 @@ pg_strtok(int *length)
return ret_str;
}
+/*
+ * Check if the next token is 'expect_token'.
+ *
+ * It handles similar to pg_strtok, except that this does not consume the
+ * next token, and has special casing for some less common tokens.
+ */
+bool
+pg_strtoken_next(const char *expect_token)
+{
+ const char *local_str; /* working pointer to string */
+ Size expect_len = strlen(expect_token);
+ char next_char;
+
+ local_str = pg_strtok_ptr;
+
+ while (*local_str == ' ' || *local_str == '\n' || *local_str == '\t')
+ local_str++;
+
+ if (*local_str == '\0')
+ return false; /* no more tokens */
+
+ Assert(expect_len > 0);
+
+ next_char = local_str[expect_len];
+
+ /* check if the next few bytes match the token */
+ if (strncmp(local_str, expect_token, expect_len) != 0)
+ return false;
+
+ /*
+ * Check that the token was actually terminated at the end of the
+ * expected token with a character that is a separate token.
+ * Otherwise, we'd get positive matches for mathing the token of "is"
+ * against a local_str of "isn't", which is clearly wrong.
+ */
+ return (next_char == '\0' ||
+ next_char == ' ' ||
+ next_char == '\n' ||
+ next_char == '\t' ||
+ next_char == '(' ||
+ next_char == ')' ||
+ next_char == '{' ||
+ next_char == '}');
+}
+
/*
* debackslash -
* create a palloc'd string holding the given token.
diff --git a/src/backend/nodes/readfuncs.c b/src/backend/nodes/readfuncs.c
index cc2021c1f7..74a2204389 100644
--- a/src/backend/nodes/readfuncs.c
+++ b/src/backend/nodes/readfuncs.c
@@ -56,112 +56,205 @@
READ_LOCALS_NO_FIELDS(nodeTypeName); \
READ_TEMP_LOCALS()
+/* a scaffold function to read an optionally-omitted field */
+#define READ_OPT_SCAFFOLD(fldname, read_field_code, default_value) \
+ if (pg_strtoken_next(":" CppAsString(fldname))) \
+ { \
+ read_field_code; \
+ } \
+ else \
+ local_node->fldname = default_value
+
/* Read an integer field (anything written as ":fldname %d") */
-#define READ_INT_FIELD(fldname) \
+#define READ_INT_FIELD_DIRECT(fldname) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
token = pg_strtok(&length); /* get field value */ \
local_node->fldname = atoi(token)
+#define READ_INT_FIELD_DEFAULT(fldname, default_value) \
+ READ_OPT_SCAFFOLD(fldname, READ_INT_FIELD_DIRECT(fldname), default_value)
+#define READ_INT_FIELD(fldname) \
+ READ_INT_FIELD_DEFAULT(fldname, 0)
/* Read an unsigned integer field (anything written as ":fldname %u") */
-#define READ_UINT_FIELD(fldname) \
+#define READ_UINT_FIELD_DIRECT(fldname) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
token = pg_strtok(&length); /* get field value */ \
local_node->fldname = atoui(token)
+#define READ_UINT_FIELD_DEFAULT(fldname, default_value) \
+ READ_OPT_SCAFFOLD(fldname, READ_UINT_FIELD_DIRECT(fldname), default_value)
+#define READ_UINT_FIELD(fldname) \
+ READ_UINT_FIELD_DEFAULT(fldname, 0)
/* Read an unsigned integer field (anything written using UINT64_FORMAT) */
-#define READ_UINT64_FIELD(fldname) \
+#define READ_UINT64_FIELD_DIRECT(fldname) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
token = pg_strtok(&length); /* get field value */ \
local_node->fldname = strtou64(token, NULL, 10)
+#define READ_UINT64_FIELD_DEFAULT(fldname, default_value) \
+ READ_OPT_SCAFFOLD(fldname, READ_UINT64_FIELD_DIRECT(fldname), default_value)
+#define READ_UINT64_FIELD(fldname) \
+ READ_UINT64_FIELD_DEFAULT(fldname, 0)
/* Read a long integer field (anything written as ":fldname %ld") */
-#define READ_LONG_FIELD(fldname) \
+#define READ_LONG_FIELD_DIRECT(fldname) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
token = pg_strtok(&length); /* get field value */ \
local_node->fldname = atol(token)
+#define READ_LONG_FIELD_DEFAULT(fldname, default_value) \
+ READ_OPT_SCAFFOLD(fldname, READ_LONG_FIELD_DIRECT(fldname), default_value)
+#define READ_LONG_FIELD(fldname) \
+ READ_LONG_FIELD_DEFAULT(fldname, 0)
/* Read an OID field (don't hard-wire assumption that OID is same as uint) */
-#define READ_OID_FIELD(fldname) \
+#define READ_OID_FIELD_DIRECT(fldname) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
token = pg_strtok(&length); /* get field value */ \
local_node->fldname = atooid(token)
+#define READ_OID_FIELD_DEFAULT(fldname, default_value) \
+ READ_OPT_SCAFFOLD(fldname, READ_OID_FIELD_DIRECT(fldname), default_value)
+#define READ_OID_FIELD(fldname) \
+ READ_OID_FIELD_DEFAULT(fldname, 0)
/* Read a char field (ie, one ascii character) */
-#define READ_CHAR_FIELD(fldname) \
+#define READ_CHAR_FIELD_DIRECT(fldname) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
token = pg_strtok(&length); /* get field value */ \
/* avoid overhead of calling debackslash() for one char */ \
local_node->fldname = (length == 0) ? '\0' : (token[0] == '\\' ? token[1] : token[0])
+#define READ_CHAR_FIELD_DEFAULT(fldname, default_value) \
+ READ_OPT_SCAFFOLD(fldname, READ_CHAR_FIELD_DIRECT(fldname), default_value)
+#define READ_CHAR_FIELD(fldname) \
+ READ_CHAR_FIELD_DEFAULT(fldname, '\0')
/* Read an enumerated-type field that was written as an integer code */
-#define READ_ENUM_FIELD(fldname, enumtype) \
+#define READ_ENUM_FIELD_DIRECT(fldname, enumtype) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
token = pg_strtok(&length); /* get field value */ \
local_node->fldname = (enumtype) atoi(token)
+#define READ_ENUM_FIELD_DEFAULT(fldname, enumtype, default_value) \
+ READ_OPT_SCAFFOLD(fldname, READ_ENUM_FIELD_DIRECT(fldname, enumtype), default_value)
+#define READ_ENUM_FIELD(fldname, enumtype) \
+ READ_ENUM_FIELD_DEFAULT(fldname, enumtype, 0)
/* Read a float field */
-#define READ_FLOAT_FIELD(fldname) \
+#define READ_FLOAT_FIELD_DIRECT(fldname) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
token = pg_strtok(&length); /* get field value */ \
local_node->fldname = atof(token)
+#define READ_FLOAT_FIELD_DEFAULT(fldname, default_value) \
+ READ_OPT_SCAFFOLD(fldname, READ_FLOAT_FIELD_DIRECT(fldname), default_value)
+#define READ_FLOAT_FIELD(fldname) \
+ READ_FLOAT_FIELD_DEFAULT(fldname, 0.0)
/* Read a boolean field */
-#define READ_BOOL_FIELD(fldname) \
+#define READ_BOOL_FIELD_DIRECT(fldname) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
token = pg_strtok(&length); /* get field value */ \
local_node->fldname = strtobool(token)
+#define READ_BOOL_FIELD_DEFAULT(fldname, default_value) \
+ READ_OPT_SCAFFOLD(fldname, READ_BOOL_FIELD_DIRECT(fldname), default_value)
+#define READ_BOOL_FIELD(fldname) \
+ READ_BOOL_FIELD_DEFAULT(fldname, false)
/* Read a character-string field */
-#define READ_STRING_FIELD(fldname) \
+#define READ_STRING_FIELD_DIRECT(fldname) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
token = pg_strtok(&length); /* get field value */ \
local_node->fldname = nullable_string(token, length)
+/* see WRITE_STRING_FIELD in outfuncs.c */
+#define READ_STRING_FIELD(fldname) \
+ READ_OPT_SCAFFOLD(fldname, READ_STRING_FIELD_DIRECT(fldname), NULL)
/* Read a parse location field (and possibly throw away the value) */
#ifdef WRITE_READ_PARSE_PLAN_TREES
-#define READ_LOCATION_FIELD(fldname) \
+#define READ_LOCATION_FIELD_DIRECT(fldname) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
token = pg_strtok(&length); /* get field value */ \
local_node->fldname = restore_location_fields ? atoi(token) : -1
#else
-#define READ_LOCATION_FIELD(fldname) \
+#define READ_LOCATION_FIELD_DIRECT(fldname) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
token = pg_strtok(&length); /* get field value */ \
(void) token; /* in case not used elsewhere */ \
local_node->fldname = -1 /* set field to "unknown" */
#endif
+/* The default Location field value is -1 ('unknown') */
+#define READ_LOCATION_FIELD(fldname) \
+ READ_OPT_SCAFFOLD(fldname, READ_LOCATION_FIELD_DIRECT(fldname), -1)
/* Read a Node field */
-#define READ_NODE_FIELD(fldname) \
+#define READ_NODE_FIELD_DIRECT(fldname) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
(void) token; /* in case not used elsewhere */ \
local_node->fldname = nodeRead(NULL, 0)
+#define READ_NODE_FIELD_DEFAULT(fldname, default_value) \
+ READ_OPT_SCAFFOLD(fldname, READ_NODE_FIELD_DIRECT(fldname), default_value)
+#define READ_NODE_FIELD(fldname) \
+ READ_NODE_FIELD_DEFAULT(fldname, NULL)
/* Read a bitmapset field */
-#define READ_BITMAPSET_FIELD(fldname) \
+#define READ_BITMAPSET_FIELD_DIRECT(fldname) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
(void) token; /* in case not used elsewhere */ \
local_node->fldname = _readBitmapset()
+#define READ_BITMAPSET_FIELD_DEFAULT(fldname, default_value) \
+ READ_OPT_SCAFFOLD(fldname, READ_BITMAPSET_FIELD_DIRECT(fldname), default_value)
+#define READ_BITMAPSET_FIELD(fldname) \
+ READ_BITMAPSET_FIELD_DEFAULT(fldname, NULL)
/* Read an attribute number array */
-#define READ_ATTRNUMBER_ARRAY(fldname, len) \
+#define READ_ATTRNUMBER_ARRAY_DIRECT(fldname, len) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
local_node->fldname = readAttrNumberCols(len)
+#define READ_ATTRNUMBER_ARRAY_DEFAULT(fldname, len, default_value) \
+ READ_OPT_SCAFFOLD(fldname, READ_ATTRNUMBER_ARRAY_DIRECT(fldname, len), default_value)
+#define READ_ATTRNUMBER_ARRAY(fldname, len) \
+ READ_ATTRNUMBER_ARRAY_DEFAULT(fldname, len, NULL)
/* Read an oid array */
-#define READ_OID_ARRAY(fldname, len) \
+#define READ_OID_ARRAY_DIRECT(fldname, len) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
local_node->fldname = readOidCols(len)
+#define READ_OID_ARRAY_DEFAULT(fldname, len, default_value) \
+ READ_OPT_SCAFFOLD(fldname, READ_OID_ARRAY_DIRECT(fldname, len), default_value)
+#define READ_OID_ARRAY(fldname, len) \
+ READ_OID_ARRAY_DEFAULT(fldname, len, NULL)
/* Read an int array */
-#define READ_INT_ARRAY(fldname, len) \
+#define READ_INT_ARRAY_DIRECT(fldname, len) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
local_node->fldname = readIntCols(len)
+#define READ_INT_ARRAY_DEFAULT(fldname, len, default_value) \
+ READ_OPT_SCAFFOLD(fldname, READ_INT_ARRAY_DIRECT(fldname, len), default_value)
+#define READ_INT_ARRAY(fldname, len) \
+ READ_INT_ARRAY_DEFAULT(fldname, len, NULL)
/* Read a bool array */
-#define READ_BOOL_ARRAY(fldname, len) \
+#define READ_BOOL_ARRAY_DIRECT(fldname, len) \
token = pg_strtok(&length); /* skip :fldname */ \
+ Assert(strncmp(token, ":" CppAsString(fldname), length) == 0); \
local_node->fldname = readBoolCols(len)
+#define READ_BOOL_ARRAY_DEFAULT(fldname, len, default_value) \
+ READ_OPT_SCAFFOLD(fldname, READ_BOOL_ARRAY_DIRECT(fldname, len), default_value)
+#define READ_BOOL_ARRAY(fldname, len) \
+ READ_BOOL_ARRAY_DEFAULT(fldname, len, NULL)
/* Routine exit */
#define READ_DONE() \
diff --git a/src/include/nodes/readfuncs.h b/src/include/nodes/readfuncs.h
index cba6f0be75..e668b24ad0 100644
--- a/src/include/nodes/readfuncs.h
+++ b/src/include/nodes/readfuncs.h
@@ -27,6 +27,7 @@ extern PGDLLIMPORT bool restore_location_fields;
* prototypes for functions in read.c (the lisp token parser)
*/
extern const char *pg_strtok(int *length);
+extern bool pg_strtoken_next(const char *expect_token);
extern char *debackslash(const char *token, int length);
extern void *nodeRead(const char *token, int tok_len);
diff --git a/src/test/regress/expected/rowsecurity.out b/src/test/regress/expected/rowsecurity.out
index 6988128aa4..a69aa40f82 100644
--- a/src/test/regress/expected/rowsecurity.out
+++ b/src/test/regress/expected/rowsecurity.out
@@ -3937,7 +3937,10 @@ CREATE TABLE coll_t (c) AS VALUES ('bar'::text);
CREATE POLICY coll_p ON coll_t USING (c < ('foo'::text COLLATE "C"));
ALTER TABLE coll_t ENABLE ROW LEVEL SECURITY;
GRANT SELECT ON coll_t TO regress_rls_alice;
-SELECT (string_to_array(polqual, ':'))[7] AS inputcollid FROM pg_policy WHERE polrelid = 'coll_t'::regclass;
+SELECT split AS inputcollid
+FROM pg_policy,
+ lateral unnest(string_to_array(polqual, ':')) as split
+WHERE polrelid = 'coll_t'::regclass and split LIKE '%inputcollid%';
inputcollid
------------------
inputcollid 950
diff --git a/src/test/regress/sql/rowsecurity.sql b/src/test/regress/sql/rowsecurity.sql
index dec7340538..cb7a4c776b 100644
--- a/src/test/regress/sql/rowsecurity.sql
+++ b/src/test/regress/sql/rowsecurity.sql
@@ -1732,7 +1732,10 @@ CREATE TABLE coll_t (c) AS VALUES ('bar'::text);
CREATE POLICY coll_p ON coll_t USING (c < ('foo'::text COLLATE "C"));
ALTER TABLE coll_t ENABLE ROW LEVEL SECURITY;
GRANT SELECT ON coll_t TO regress_rls_alice;
-SELECT (string_to_array(polqual, ':'))[7] AS inputcollid FROM pg_policy WHERE polrelid = 'coll_t'::regclass;
+SELECT split AS inputcollid
+FROM pg_policy,
+ lateral unnest(string_to_array(polqual, ':')) as split
+WHERE polrelid = 'coll_t'::regclass and split LIKE '%inputcollid%';
SET SESSION AUTHORIZATION regress_rls_alice;
SELECT * FROM coll_t;
ROLLBACK;
--
2.40.1
[application/octet-stream] v1-0002-pg_node_tree-reset-node-location-before-catalog-s.patch (12.9K, ../../CAEze2Wigkd1+J4s=7wUqW8Y4g9mDWSC28119ukbKkf799WBpzg@mail.gmail.com/3-v1-0002-pg_node_tree-reset-node-location-before-catalog-s.patch)
download | inline diff:
From 7466ff936bd0829a622125eb351c6df57c4c4f3d Mon Sep 17 00:00:00 2001
From: Matthias van de Meent <[email protected]>
Date: Wed, 3 Jan 2024 01:39:42 +0100
Subject: [PATCH v1 2/7] pg_node_tree: reset node->location before catalog's
serialization
We don't store original query texts, so any lingering "location" value
can only be useful in forensic debugging. In normal operation, however,
a non-default value will show up as measurable overhead in
serialization, so we reset it to its default (-1), saving several
10s of kB.
---
src/backend/catalog/heap.c | 11 +++-
src/backend/catalog/index.c | 2 +
src/backend/catalog/pg_attrdef.c | 5 +-
src/backend/catalog/pg_proc.c | 6 ++
src/backend/catalog/pg_publication.c | 4 ++
src/backend/commands/policy.c | 7 ++
src/backend/commands/statscmds.c | 1 +
src/backend/commands/trigger.c | 2 +
src/backend/commands/typecmds.c | 6 +-
src/backend/nodes/nodeFuncs.c | 95 ++++++++++++++++++++++++++++
src/backend/rewrite/rewriteDefine.c | 8 +++
src/include/nodes/nodeFuncs.h | 2 +
12 files changed, 146 insertions(+), 3 deletions(-)
diff --git a/src/backend/catalog/heap.c b/src/backend/catalog/heap.c
index b93894889d..419d4ab30e 100644
--- a/src/backend/catalog/heap.c
+++ b/src/backend/catalog/heap.c
@@ -2062,8 +2062,10 @@ StoreRelCheck(Relation rel, const char *ccname, Node *expr,
Oid constrOid;
/*
- * Flatten expression to string form for storage.
+ * Remove references into the (to be dropped) query string, and
+ * flatten expression to string form for storage.
*/
+ reset_querytext_references(expr, NULL);
ccbin = nodeToString(expr);
/*
@@ -3676,6 +3678,7 @@ StorePartitionKey(Relation rel,
{
char *exprString;
+ reset_querytext_references((Node *) partexprs, NULL);
exprString = nodeToString(partexprs);
partexprDatum = CStringGetTextDatum(exprString);
pfree(exprString);
@@ -3834,6 +3837,12 @@ StorePartitionBound(Relation rel, Relation parent, PartitionBoundSpec *bound)
memset(new_val, 0, sizeof(new_val));
memset(new_null, false, sizeof(new_null));
memset(new_repl, false, sizeof(new_repl));
+
+ /*
+ * We don't keep the original query text around, so remove any
+ * references to it to reduce its stored size.
+ */
+ reset_querytext_references((Node *) bound, NULL);
new_val[Anum_pg_class_relpartbound - 1] = CStringGetTextDatum(nodeToString(bound));
new_null[Anum_pg_class_relpartbound - 1] = false;
new_repl[Anum_pg_class_relpartbound - 1] = true;
diff --git a/src/backend/catalog/index.c b/src/backend/catalog/index.c
index 7b186c0220..e01ff9cf7f 100644
--- a/src/backend/catalog/index.c
+++ b/src/backend/catalog/index.c
@@ -588,6 +588,7 @@ UpdateIndexRelation(Oid indexoid,
{
char *exprsString;
+ reset_querytext_references((Node *) indexInfo->ii_Expressions, NULL);
exprsString = nodeToString(indexInfo->ii_Expressions);
exprsDatum = CStringGetTextDatum(exprsString);
pfree(exprsString);
@@ -603,6 +604,7 @@ UpdateIndexRelation(Oid indexoid,
{
char *predString;
+ reset_querytext_references((Node *) indexInfo->ii_Predicate, NULL);
predString = nodeToString(make_ands_explicit(indexInfo->ii_Predicate));
predDatum = CStringGetTextDatum(predString);
pfree(predString);
diff --git a/src/backend/catalog/pg_attrdef.c b/src/backend/catalog/pg_attrdef.c
index ade0b6d8e6..49cb869e3c 100644
--- a/src/backend/catalog/pg_attrdef.c
+++ b/src/backend/catalog/pg_attrdef.c
@@ -23,6 +23,7 @@
#include "catalog/objectaccess.h"
#include "catalog/pg_attrdef.h"
#include "executor/executor.h"
+#include "nodes/nodeFuncs.h"
#include "optimizer/optimizer.h"
#include "utils/array.h"
#include "utils/builtins.h"
@@ -62,8 +63,10 @@ StoreAttrDefault(Relation rel, AttrNumber attnum,
adrel = table_open(AttrDefaultRelationId, RowExclusiveLock);
/*
- * Flatten expression to string form for storage.
+ * Reset any references to the original query text (which isn't stored),
+ * and flatten the expression to string form for storage.
*/
+ reset_querytext_references(expr, NULL);
adbin = nodeToString(expr);
/*
diff --git a/src/backend/catalog/pg_proc.c b/src/backend/catalog/pg_proc.c
index b5fd364003..ca0d1f6174 100644
--- a/src/backend/catalog/pg_proc.c
+++ b/src/backend/catalog/pg_proc.c
@@ -331,7 +331,10 @@ ProcedureCreate(const char *procedureName,
else
nulls[Anum_pg_proc_proargnames - 1] = true;
if (parameterDefaults != NIL)
+ {
+ reset_querytext_references((Node *) parameterDefaults, NULL);
values[Anum_pg_proc_proargdefaults - 1] = CStringGetTextDatum(nodeToString(parameterDefaults));
+ }
else
nulls[Anum_pg_proc_proargdefaults - 1] = true;
if (trftypes != PointerGetDatum(NULL))
@@ -344,7 +347,10 @@ ProcedureCreate(const char *procedureName,
else
nulls[Anum_pg_proc_probin - 1] = true;
if (prosqlbody)
+ {
+ reset_querytext_references(prosqlbody, NULL);
values[Anum_pg_proc_prosqlbody - 1] = CStringGetTextDatum(nodeToString(prosqlbody));
+ }
else
nulls[Anum_pg_proc_prosqlbody - 1] = true;
if (proconfig != PointerGetDatum(NULL))
diff --git a/src/backend/catalog/pg_publication.c b/src/backend/catalog/pg_publication.c
index c488b6370b..d4b772e543 100644
--- a/src/backend/catalog/pg_publication.c
+++ b/src/backend/catalog/pg_publication.c
@@ -36,6 +36,7 @@
#include "commands/publicationcmds.h"
#include "funcapi.h"
#include "miscadmin.h"
+#include "nodes/nodeFuncs.h"
#include "utils/array.h"
#include "utils/builtins.h"
#include "utils/catcache.h"
@@ -422,7 +423,10 @@ publication_add_relation(Oid pubid, PublicationRelInfo *pri,
/* Add qualifications, if available */
if (pri->whereClause != NULL)
+ {
+ reset_querytext_references(pri->whereClause, NULL);
values[Anum_pg_publication_rel_prqual - 1] = CStringGetTextDatum(nodeToString(pri->whereClause));
+ }
else
nulls[Anum_pg_publication_rel_prqual - 1] = true;
diff --git a/src/backend/commands/policy.c b/src/backend/commands/policy.c
index 76a45e56bf..ce01336ad3 100644
--- a/src/backend/commands/policy.c
+++ b/src/backend/commands/policy.c
@@ -30,6 +30,7 @@
#include "commands/policy.h"
#include "miscadmin.h"
#include "nodes/makefuncs.h"
+#include "nodes/nodeFuncs.h"
#include "nodes/pg_list.h"
#include "parser/parse_clause.h"
#include "parser/parse_collate.h"
@@ -701,13 +702,19 @@ CreatePolicy(CreatePolicyStmt *stmt)
/* Add qual if present. */
if (qual)
+ {
+ reset_querytext_references(qual, NULL);
values[Anum_pg_policy_polqual - 1] = CStringGetTextDatum(nodeToString(qual));
+ }
else
isnull[Anum_pg_policy_polqual - 1] = true;
/* Add WITH CHECK qual if present */
if (with_check_qual)
+ {
+ reset_querytext_references(with_check_qual, NULL);
values[Anum_pg_policy_polwithcheck - 1] = CStringGetTextDatum(nodeToString(with_check_qual));
+ }
else
isnull[Anum_pg_policy_polwithcheck - 1] = true;
diff --git a/src/backend/commands/statscmds.c b/src/backend/commands/statscmds.c
index 20be564df2..c9e85f0af8 100644
--- a/src/backend/commands/statscmds.c
+++ b/src/backend/commands/statscmds.c
@@ -477,6 +477,7 @@ CreateStatistics(CreateStatsStmt *stmt)
{
char *exprsString;
+ reset_querytext_references((Node *) stxexprs, NULL);
exprsString = nodeToString(stxexprs);
exprsDatum = CStringGetTextDatum(exprsString);
pfree(exprsString);
diff --git a/src/backend/commands/trigger.c b/src/backend/commands/trigger.c
index 52177759ab..b1ffa8b17f 100644
--- a/src/backend/commands/trigger.c
+++ b/src/backend/commands/trigger.c
@@ -39,6 +39,7 @@
#include "miscadmin.h"
#include "nodes/bitmapset.h"
#include "nodes/makefuncs.h"
+#include "nodes/nodeFuncs.h"
#include "optimizer/optimizer.h"
#include "parser/parse_clause.h"
#include "parser/parse_collate.h"
@@ -674,6 +675,7 @@ CreateTriggerFiringOn(CreateTrigStmt *stmt, const char *queryString,
/* we'll need the rtable for recordDependencyOnExpr */
whenRtable = pstate->p_rtable;
+ reset_querytext_references(whenClause, NULL);
qual = nodeToString(whenClause);
free_parsestate(pstate);
diff --git a/src/backend/commands/typecmds.c b/src/backend/commands/typecmds.c
index aaf9728697..b76f1d1c9f 100644
--- a/src/backend/commands/typecmds.c
+++ b/src/backend/commands/typecmds.c
@@ -58,6 +58,7 @@
#include "executor/executor.h"
#include "miscadmin.h"
#include "nodes/makefuncs.h"
+#include "nodes/nodeFuncs.h"
#include "optimizer/optimizer.h"
#include "parser/parse_coerce.h"
#include "parser/parse_collate.h"
@@ -924,6 +925,7 @@ DefineDomain(CreateDomainStmt *stmt)
defaultValue =
deparse_expression(defaultExpr,
NIL, false, false);
+ reset_querytext_references(defaultExpr, NULL);
defaultValueBin = nodeToString(defaultExpr);
}
}
@@ -3506,8 +3508,10 @@ domainAddConstraint(Oid domainOid, Oid domainNamespace, Oid baseTypeOid,
errmsg("cannot use table references in domain check constraint")));
/*
- * Convert to string form for storage.
+ * Remove location references into the original query string (which won't
+ * be stored) and convert to string form for storage.
*/
+ reset_querytext_references(expr, NULL);
ccbin = nodeToString(expr);
/*
diff --git a/src/backend/nodes/nodeFuncs.c b/src/backend/nodes/nodeFuncs.c
index c03f4f23e2..44a922464b 100644
--- a/src/backend/nodes/nodeFuncs.c
+++ b/src/backend/nodes/nodeFuncs.c
@@ -1903,6 +1903,101 @@ check_functions_in_node(Node *node, check_function_callback checker,
return false;
}
+/*
+ * Reset all query text related fields to their original value.
+ *
+ * This is used for reducing the size of nodeToText, like in pg_rewrite.
+ */
+bool
+reset_querytext_references(Node *node, void *context)
+{
+ const int query_recurse_flags = 0;
+ ListCell *temp;
+
+#define RESET_FOR(_type_) \
+ case T_##_type_: \
+ { \
+ castNode(_type_, node)->location = -1; \
+ break; \
+ }
+
+ if (node == NULL)
+ return false;
+
+ if (IsA(node, Query))
+ {
+ Query *query = castNode(Query, node);
+ query->stmt_location = -1;
+ query->stmt_len = 0;
+
+ return query_tree_walker(query, reset_querytext_references, context,
+ query_recurse_flags);
+ }
+
+ switch (nodeTag(node))
+ {
+ RESET_FOR(RangeVar);
+ RESET_FOR(TableFunc);
+ RESET_FOR(Var);
+ RESET_FOR(Const);
+ RESET_FOR(Param);
+ RESET_FOR(Aggref);
+ RESET_FOR(GroupingFunc);
+ RESET_FOR(WindowFunc);
+ RESET_FOR(FuncExpr);
+ RESET_FOR(NamedArgExpr);
+ RESET_FOR(OpExpr);
+ RESET_FOR(DistinctExpr);
+ RESET_FOR(NullIfExpr);
+ RESET_FOR(ScalarArrayOpExpr);
+ RESET_FOR(BoolExpr);
+ RESET_FOR(SubLink);
+ RESET_FOR(RelabelType);
+ RESET_FOR(CoerceViaIO);
+ RESET_FOR(ArrayCoerceExpr);
+ RESET_FOR(ConvertRowtypeExpr);
+ RESET_FOR(CollateExpr);
+ RESET_FOR(CaseWhen);
+ RESET_FOR(ArrayExpr);
+ RESET_FOR(RowExpr);
+ RESET_FOR(CoalesceExpr);
+ RESET_FOR(MinMaxExpr);
+ RESET_FOR(SQLValueFunction);
+ RESET_FOR(XmlExpr);
+ RESET_FOR(JsonFormat);
+ RESET_FOR(JsonConstructorExpr);
+ RESET_FOR(JsonIsPredicate);
+ RESET_FOR(NullTest);
+ RESET_FOR(BooleanTest);
+ RESET_FOR(CoerceToDomain);
+ RESET_FOR(CoerceToDomainValue);
+ RESET_FOR(SetToDefault);
+ case T_CaseExpr:
+ {
+ /*
+ * The expression_tree_walker does not call us for CaseWhen nodes,
+ * but instead directly wires us through to its inner Nodes.
+ * To correctly reset the location fields, we manually iterate
+ * to get those fields reset.
+ */
+ CaseExpr *caseExpr = castNode(CaseExpr, node);
+ caseExpr->location = -1;
+ foreach(temp, caseExpr->args)
+ {
+ CaseWhen *when = lfirst_node(CaseWhen, temp);
+ when->location = -1;
+ }
+ break;
+ }
+ default:
+ break;
+ }
+
+ return expression_tree_walker(node, reset_querytext_references,
+ (void *) context);
+#undef RESET_FOR
+}
+
/*
* Standard expression-tree walking support
diff --git a/src/backend/rewrite/rewriteDefine.c b/src/backend/rewrite/rewriteDefine.c
index e36fc72e1e..57650935ac 100644
--- a/src/backend/rewrite/rewriteDefine.c
+++ b/src/backend/rewrite/rewriteDefine.c
@@ -474,6 +474,14 @@ DefineQueryRewrite(const char *rulename,
/* discard rule if it's null action and not INSTEAD; it's a no-op */
if (action != NIL || is_instead)
{
+ /*
+ * Clear location info from the action and event_qual data. We don't
+ * store the original query, so keeping track of this information is
+ * meaningless and hinders serialization/compression efforts.
+ */
+ reset_querytext_references(event_qual, NULL);
+ reset_querytext_references((Node *) action, NULL);
+
ruleId = InsertRule(rulename,
event_type,
event_relid,
diff --git a/src/include/nodes/nodeFuncs.h b/src/include/nodes/nodeFuncs.h
index 20921b45b9..f513b2ba37 100644
--- a/src/include/nodes/nodeFuncs.h
+++ b/src/include/nodes/nodeFuncs.h
@@ -139,6 +139,8 @@ get_notclausearg(const void *notclause)
extern bool check_functions_in_node(Node *node, check_function_callback checker,
void *context);
+extern bool reset_querytext_references(Node *node, void *context);
+
/*
* The following functions are usually passed walker or mutator callbacks
* that are declared like "bool walker(Node *node, my_struct *context)"
--
2.40.1
[application/octet-stream] v1-0005-NodeSupport-Don-t-emit-trailing-0s-in-outDatum.patch (2.4K, ../../CAEze2Wigkd1+J4s=7wUqW8Y4g9mDWSC28119ukbKkf799WBpzg@mail.gmail.com/4-v1-0005-NodeSupport-Don-t-emit-trailing-0s-in-outDatum.patch)
download | inline diff:
From 5a2434438bc3922e9b4fcc1a64e57b97e0ef8183 Mon Sep 17 00:00:00 2001
From: Matthias van de Meent <[email protected]>
Date: Wed, 3 Jan 2024 02:54:51 +0100
Subject: [PATCH v1 5/7] NodeSupport: Don't emit trailing 0s in outDatum
This reduces raw serialized size of the initdb pg_rewrite dataset by
a limited 1%, and compressed dataset by 0.5%.
---
src/backend/nodes/outfuncs.c | 13 ++++++++++++-
src/backend/nodes/readfuncs.c | 13 +++++++++++++
2 files changed, 25 insertions(+), 1 deletion(-)
diff --git a/src/backend/nodes/outfuncs.c b/src/backend/nodes/outfuncs.c
index 73f1298dee..f6cfcc8a4c 100644
--- a/src/backend/nodes/outfuncs.c
+++ b/src/backend/nodes/outfuncs.c
@@ -410,9 +410,15 @@ outDatum(StringInfo str, Datum value, int typlen, bool typbyval)
if (typbyval)
{
+ int write_length = sizeof(Datum);
s = (char *) (&value);
appendStringInfo(str, "%u [ ", (unsigned int) length);
- for (i = 0; i < (Size) sizeof(Datum); i++)
+
+ /* truncate postfix zeroes */
+ while (write_length != 0 && s[write_length - 1] == 0)
+ write_length -= 1;
+
+ for (i = 0; i < write_length; i++)
appendStringInfo(str, "%d ", (int) (s[i]));
appendStringInfoChar(str, ']');
}
@@ -424,6 +430,11 @@ outDatum(StringInfo str, Datum value, int typlen, bool typbyval)
else
{
appendStringInfo(str, "%u [ ", (unsigned int) length);
+
+ /* truncate postfix zeroes */
+ while (length != 0 && s[length - 1] == 0)
+ length -= 1;
+
for (i = 0; i < length; i++)
appendStringInfo(str, "%d ", (int) (s[i]));
appendStringInfoChar(str, ']');
diff --git a/src/backend/nodes/readfuncs.c b/src/backend/nodes/readfuncs.c
index dd1e505bd7..4287a7ec6e 100644
--- a/src/backend/nodes/readfuncs.c
+++ b/src/backend/nodes/readfuncs.c
@@ -852,9 +852,14 @@ readDatum(bool typbyval)
s = (char *) (&res);
for (i = 0; i < (Size) sizeof(Datum); i++)
{
+ if (pg_strtoken_next("]"))
+ break;
+
token = pg_strtok(&tokenLength);
s[i] = (char) atoi(token);
}
+ for (; i < (Size) sizeof(Datum); i++)
+ s[i] = 0;
}
else if (length <= 0)
res = (Datum) NULL;
@@ -863,9 +868,17 @@ readDatum(bool typbyval)
s = (char *) palloc(length);
for (i = 0; i < length; i++)
{
+ if (pg_strtoken_next("]"))
+ break;
+
token = pg_strtok(&tokenLength);
+
s[i] = (char) atoi(token);
}
+
+ for (; i < length; i++)
+ s[i] = 0;
+
res = PointerGetDatum(s);
}
--
2.40.1
[application/octet-stream] v1-0004-NodeSupport-add-some-more-default-markers-for-var.patch (4.6K, ../../CAEze2Wigkd1+J4s=7wUqW8Y4g9mDWSC28119ukbKkf799WBpzg@mail.gmail.com/5-v1-0004-NodeSupport-add-some-more-default-markers-for-var.patch)
download | inline diff:
From 63bd9155b631b16e82781f9a6405ac0ee24cde15 Mon Sep 17 00:00:00 2001
From: Matthias van de Meent <[email protected]>
Date: Wed, 3 Jan 2024 02:41:23 +0100
Subject: [PATCH v1 4/7] NodeSupport: add some more default markers for various
fields
This reduces the size of pg_rewrite by a further 14%.
---
src/include/nodes/parsenodes.h | 8 +++++---
src/include/nodes/primnodes.h | 22 ++++++++++++++--------
2 files changed, 19 insertions(+), 11 deletions(-)
diff --git a/src/include/nodes/parsenodes.h b/src/include/nodes/parsenodes.h
index b9f122b46c..e11b044b89 100644
--- a/src/include/nodes/parsenodes.h
+++ b/src/include/nodes/parsenodes.h
@@ -117,7 +117,8 @@ typedef struct Query
{
NodeTag type;
- CmdType commandType; /* select|insert|update|delete|merge|utility */
+ /* select|insert|update|delete|merge|utility */
+ CmdType commandType pg_node_attr(default(CMD_SELECT));
/* where did I come from? */
QuerySource querySource pg_node_attr(query_jumble_ignore);
@@ -130,7 +131,7 @@ typedef struct Query
uint64 queryId pg_node_attr(equal_ignore, query_jumble_ignore, read_write_ignore, read_as(0));
/* do I set the command result tag? */
- bool canSetTag pg_node_attr(query_jumble_ignore);
+ bool canSetTag pg_node_attr(query_jumble_ignore, default(true));
Node *utilityStmt; /* non-null if commandType == CMD_UTILITY */
@@ -1235,7 +1236,8 @@ typedef struct RTEPermissionInfo
Oid relid; /* relation OID */
bool inh; /* separately check inheritance children? */
- AclMode requiredPerms; /* bitmask of required access permissions */
+ /* bitmask of required access permissions, in views usually ACL_SELECT */
+ AclMode requiredPerms pg_node_attr(default(ACL_SELECT));
Oid checkAsUser; /* if valid, check access as this role */
Bitmapset *selectedCols; /* columns needing SELECT permission */
Bitmapset *insertedCols; /* columns needing INSERT permission */
diff --git a/src/include/nodes/primnodes.h b/src/include/nodes/primnodes.h
index c9b30b1b0d..acb275900f 100644
--- a/src/include/nodes/primnodes.h
+++ b/src/include/nodes/primnodes.h
@@ -82,7 +82,7 @@ typedef struct RangeVar
char *relname;
/* expand rel by inheritance? recursively act on children? */
- bool inh;
+ bool inh pg_node_attr(default(true));
/* see RELPERSISTENCE_* in pg_class.h */
char relpersistence;
@@ -126,7 +126,7 @@ typedef struct TableFunc
/* nullability flag for each output column */
Bitmapset *notnulls pg_node_attr(query_jumble_ignore);
/* counts from 0; -1 if none specified */
- int ordinalitycol pg_node_attr(query_jumble_ignore);
+ int ordinalitycol pg_node_attr(query_jumble_ignore, default(-1));
/* token location, or -1 if unknown */
int location;
} TableFunc;
@@ -237,8 +237,10 @@ typedef struct Var
/*
* index of this var's relation in the range table, or
* INNER_VAR/OUTER_VAR/etc
+ *
+ * As we start at 1, using that as a default reduces our serialized size.
*/
- int varno;
+ int varno pg_node_attr(default(1));
/*
* attribute number of this var, or zero for all attrs ("whole-row Var")
@@ -269,11 +271,13 @@ typedef struct Var
* varnosyn/varattnosyn are ignored for equality, because Vars with
* different syntactic identifiers are semantically the same as long as
* their varno/varattno match.
+ * As they're generally the same value as varno/varattno, we refer to
+ * those fields for initial/default values.
*/
/* syntactic relation index (0 if unknown) */
- Index varnosyn pg_node_attr(equal_ignore, query_jumble_ignore);
+ Index varnosyn pg_node_attr(equal_ignore, query_jumble_ignore, default_ref(varno));
/* syntactic attribute number */
- AttrNumber varattnosyn pg_node_attr(equal_ignore, query_jumble_ignore);
+ AttrNumber varattnosyn pg_node_attr(equal_ignore, query_jumble_ignore, default_ref(varattno));
/* token location, or -1 if unknown */
int location;
@@ -1763,8 +1767,10 @@ typedef struct CoerceToDomain
int32 resulttypmod pg_node_attr(query_jumble_ignore, default(-1));
/* OID of collation, or InvalidOid if none */
Oid resultcollid pg_node_attr(query_jumble_ignore);
- /* how to display this node */
- CoercionForm coercionformat pg_node_attr(query_jumble_ignore);
+ /* how to display this node.
+ * This node usually only remains after explicit cast, so that's chosen as
+ * the default. */
+ CoercionForm coercionformat pg_node_attr(query_jumble_ignore, default(COERCE_EXPLICIT_CAST));
int location; /* token location, or -1 if unknown */
} CoerceToDomain;
@@ -1975,7 +1981,7 @@ typedef struct TargetEntry
typedef struct RangeTblRef
{
NodeTag type;
- int rtindex;
+ int rtindex pg_node_attr(default(1));
} RangeTblRef;
/*----------
--
2.40.1
[application/octet-stream] v1-0003-Nodesupport-add-support-for-custom-default-values.patch (13.2K, ../../CAEze2Wigkd1+J4s=7wUqW8Y4g9mDWSC28119ukbKkf799WBpzg@mail.gmail.com/6-v1-0003-Nodesupport-add-support-for-custom-default-values.patch)
download | inline diff:
From b53fa22d12dbf7b104284866802923838f97ad2e Mon Sep 17 00:00:00 2001
From: Matthias van de Meent <[email protected]>
Date: Wed, 3 Jan 2024 02:17:19 +0100
Subject: [PATCH v1 3/7] Nodesupport: add support for custom default values
Initial testing on typmod values shows that pg_rewrite's ev_action
column data shrinks by about 4% with only these changes.
---
src/backend/nodes/gen_node_support.pl | 86 +++++++++++++++++----------
src/backend/nodes/outfuncs.c | 2 +
src/backend/nodes/readfuncs.c | 2 +
src/include/nodes/parsenodes.h | 2 +-
src/include/nodes/pathnodes.h | 2 +-
src/include/nodes/primnodes.h | 22 +++----
6 files changed, 73 insertions(+), 43 deletions(-)
diff --git a/src/backend/nodes/gen_node_support.pl b/src/backend/nodes/gen_node_support.pl
index c7091d6bf2..5586368475 100644
--- a/src/backend/nodes/gen_node_support.pl
+++ b/src/backend/nodes/gen_node_support.pl
@@ -362,6 +362,12 @@ foreach my $infile (@ARGV)
{
$manual_nodetag_number{$in_struct} = $1;
}
+ elsif ($attr =~ /^default\(([\w\d."'-]+)\)$/)
+ {
+ }
+ elsif ($attr =~ /^default_ref\(([\w\d."'-]+)\)$/)
+ {
+ }
else
{
die
@@ -436,7 +442,7 @@ foreach my $infile (@ARGV)
}
# normal struct field
elsif ($line =~
- /^\s*(.+)\s*\b(\w+)(\[[\w\s+]+\])?\s*(?:pg_node_attr\(([\w(), ]*)\))?;/
+ /^\s*(.+)\s*\b(\w+)(\[[\w\s+]+\])?\s*(?:pg_node_attr\(([\w\d(), ."'-]*)\))?;/
)
{
if ($is_node_struct)
@@ -468,6 +474,8 @@ foreach my $infile (@ARGV)
if ( $attr !~ /^array_size\(\w+\)$/
&& $attr !~ /^copy_as\(\w+\)$/
&& $attr !~ /^read_as\(\w+\)$/
+ && $attr !~ /^default\([\w\d."'-]+\)$/
+ && $attr !~ /^default_ref\([\w\d."'-]+\)$/
&& !elem $attr,
qw(copy_as_scalar
equal_as_scalar
@@ -494,7 +502,7 @@ foreach my $infile (@ARGV)
}
# function pointer field
elsif ($line =~
- /^\s*([\w\s*]+)\s*\(\*(\w+)\)\s*\((.*)\)\s*(?:pg_node_attr\(([\w(), ]*)\))?;/
+ /^\s*([\w\s*]+)\s*\(\*(\w+)\)\s*\((.*)\)\s*(?:pg_node_attr\(([\w\d(), ."'-]*)\))?;/
)
{
if ($is_node_struct)
@@ -969,6 +977,11 @@ _read${n}(void)
my $array_size_field;
my $read_as_field;
my $read_write_ignore = 0;
+ # macro suffix ("_DEFAULT" for manual default overrides) and default
+ # value argument.
+ my $s = "";
+ my $d = "";
+
foreach my $a (@a)
{
if ($a =~ /^array_size\(([\w.]+)\)$/)
@@ -987,6 +1000,19 @@ _read${n}(void)
{
$read_write_ignore = 1;
}
+ elsif ($a =~ /^default\(([\w\d+."'-]+)\)$/)
+ {
+ $d = ", $1";
+ }
+ elsif ($a =~ /^default_ref\(([\w\d+."'-]+)\)$/)
+ {
+ $d = ", NODE_FIELD($1)";
+ }
+ }
+
+ if (!($d eq ""))
+ {
+ $s = "_DEFAULT";
}
if ($read_write_ignore)
@@ -1007,13 +1033,13 @@ _read${n}(void)
# select instructions by field type
if ($t eq 'bool')
{
- print $off "\tWRITE_BOOL_FIELD($f);\n";
- print $rff "\tREAD_BOOL_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_BOOL_FIELD$s($f$d);\n";
+ print $rff "\tREAD_BOOL_FIELD$s($f$d);\n" unless $no_read;
}
elsif ($t eq 'int' && $f =~ 'location$')
{
- print $off "\tWRITE_LOCATION_FIELD($f);\n";
- print $rff "\tREAD_LOCATION_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_LOCATION_FIELD$s($f$d);\n";
+ print $rff "\tREAD_LOCATION_FIELD$s($f$d);\n" unless $no_read;
}
elsif ($t eq 'int'
|| $t eq 'int16'
@@ -1021,8 +1047,8 @@ _read${n}(void)
|| $t eq 'AttrNumber'
|| $t eq 'StrategyNumber')
{
- print $off "\tWRITE_INT_FIELD($f);\n";
- print $rff "\tREAD_INT_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_INT_FIELD$s($f$d);\n";
+ print $rff "\tREAD_INT_FIELD$s($f$d);\n" unless $no_read;
}
elsif ($t eq 'uint32'
|| $t eq 'bits32'
@@ -1030,44 +1056,44 @@ _read${n}(void)
|| $t eq 'Index'
|| $t eq 'SubTransactionId')
{
- print $off "\tWRITE_UINT_FIELD($f);\n";
- print $rff "\tREAD_UINT_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_UINT_FIELD$s($f$d);\n";
+ print $rff "\tREAD_UINT_FIELD$s($f$d);\n" unless $no_read;
}
elsif ($t eq 'uint64'
|| $t eq 'AclMode')
{
- print $off "\tWRITE_UINT64_FIELD($f);\n";
- print $rff "\tREAD_UINT64_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_UINT64_FIELD$s($f$d);\n";
+ print $rff "\tREAD_UINT64_FIELD$s($f$d);\n" unless $no_read;
}
elsif ($t eq 'Oid' || $t eq 'RelFileNumber')
{
- print $off "\tWRITE_OID_FIELD($f);\n";
- print $rff "\tREAD_OID_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_OID_FIELD$s($f$d);\n";
+ print $rff "\tREAD_OID_FIELD$s($f$d);\n" unless $no_read;
}
elsif ($t eq 'long')
{
- print $off "\tWRITE_LONG_FIELD($f);\n";
- print $rff "\tREAD_LONG_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_LONG_FIELD$s($f$d);\n";
+ print $rff "\tREAD_LONG_FIELD$s($f$d);\n" unless $no_read;
}
elsif ($t eq 'char')
{
- print $off "\tWRITE_CHAR_FIELD($f);\n";
- print $rff "\tREAD_CHAR_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_CHAR_FIELD$s($f$d);\n";
+ print $rff "\tREAD_CHAR_FIELD$s($f$d);\n" unless $no_read;
}
elsif ($t eq 'double')
{
- print $off "\tWRITE_FLOAT_FIELD($f);\n";
- print $rff "\tREAD_FLOAT_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_FLOAT_FIELD$s($f$d);\n";
+ print $rff "\tREAD_FLOAT_FIELD$s($f$d);\n" unless $no_read;
}
elsif ($t eq 'Cardinality')
{
- print $off "\tWRITE_FLOAT_FIELD($f);\n";
- print $rff "\tREAD_FLOAT_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_FLOAT_FIELD$s($f$d);\n";
+ print $rff "\tREAD_FLOAT_FIELD$s($f$d);\n" unless $no_read;
}
elsif ($t eq 'Cost')
{
- print $off "\tWRITE_FLOAT_FIELD($f);\n";
- print $rff "\tREAD_FLOAT_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_FLOAT_FIELD$s($f$d);\n";
+ print $rff "\tREAD_FLOAT_FIELD$s($f$d);\n" unless $no_read;
}
elsif ($t eq 'QualCost')
{
@@ -1078,13 +1104,13 @@ _read${n}(void)
}
elsif ($t eq 'Selectivity')
{
- print $off "\tWRITE_FLOAT_FIELD($f);\n";
- print $rff "\tREAD_FLOAT_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_FLOAT_FIELD$s($f$d);\n";
+ print $rff "\tREAD_FLOAT_FIELD$s($f$d);\n" unless $no_read;
}
elsif ($t eq 'char*')
{
- print $off "\tWRITE_STRING_FIELD($f);\n";
- print $rff "\tREAD_STRING_FIELD($f);\n" unless $no_read;
+ print $off "\tWRITE_STRING_FIELD($f$d);\n";
+ print $rff "\tREAD_STRING_FIELD($f$d);\n" unless $no_read;
}
elsif ($t eq 'Bitmapset*' || $t eq 'Relids')
{
@@ -1093,8 +1119,8 @@ _read${n}(void)
}
elsif (elem $t, @enum_types)
{
- print $off "\tWRITE_ENUM_FIELD($f, $t);\n";
- print $rff "\tREAD_ENUM_FIELD($f, $t);\n" unless $no_read;
+ print $off "\tWRITE_ENUM_FIELD$s($f, $t$d);\n";
+ print $rff "\tREAD_ENUM_FIELD$s($f, $t$d);\n" unless $no_read;
}
# arrays of scalar types
elsif ($t =~ /^(\w+)(\*|\[\w+\])$/ and elem $1, @scalar_types)
diff --git a/src/backend/nodes/outfuncs.c b/src/backend/nodes/outfuncs.c
index 29f4e43581..73f1298dee 100644
--- a/src/backend/nodes/outfuncs.c
+++ b/src/backend/nodes/outfuncs.c
@@ -193,6 +193,8 @@ static void outDouble(StringInfo str, double d);
#define WRITE_BOOL_ARRAY(fldname, len) \
(node->fldname == NULL ? (0) : WRITE_BOOL_ARRAY_DIRECT(fldname, len))
+#define NODE_FIELD(fldname) (node->fldname)
+
#define booltostr(x) ((x) ? "true" : "false")
diff --git a/src/backend/nodes/readfuncs.c b/src/backend/nodes/readfuncs.c
index 74a2204389..dd1e505bd7 100644
--- a/src/backend/nodes/readfuncs.c
+++ b/src/backend/nodes/readfuncs.c
@@ -256,6 +256,8 @@
#define READ_BOOL_ARRAY(fldname, len) \
READ_BOOL_ARRAY_DEFAULT(fldname, len, NULL)
+#define NODE_FIELD(fldname) (local_node->fldname)
+
/* Routine exit */
#define READ_DONE() \
return local_node
diff --git a/src/include/nodes/parsenodes.h b/src/include/nodes/parsenodes.h
index e494309da8..b9f122b46c 100644
--- a/src/include/nodes/parsenodes.h
+++ b/src/include/nodes/parsenodes.h
@@ -1608,7 +1608,7 @@ typedef struct CTECycleClause
int location;
/* These fields are set during parse analysis: */
Oid cycle_mark_type; /* common type of _value and _default */
- int cycle_mark_typmod;
+ int cycle_mark_typmod pg_node_attr(default(-1));
Oid cycle_mark_collation;
Oid cycle_mark_neop; /* <> operator for type */
} CTECycleClause;
diff --git a/src/include/nodes/pathnodes.h b/src/include/nodes/pathnodes.h
index ed85dc7414..e688fc2b79 100644
--- a/src/include/nodes/pathnodes.h
+++ b/src/include/nodes/pathnodes.h
@@ -3379,7 +3379,7 @@ typedef struct AggTransInfo
Oid aggtranstype;
/* Additional data about transtype */
- int32 aggtranstypmod;
+ int32 aggtranstypmod pg_node_attr(default(-1));
int transtypeLen;
bool transtypeByVal;
diff --git a/src/include/nodes/primnodes.h b/src/include/nodes/primnodes.h
index bb930afb52..c9b30b1b0d 100644
--- a/src/include/nodes/primnodes.h
+++ b/src/include/nodes/primnodes.h
@@ -248,7 +248,7 @@ typedef struct Var
/* pg_type OID for the type of this var */
Oid vartype pg_node_attr(query_jumble_ignore);
/* pg_attribute typmod value */
- int32 vartypmod pg_node_attr(query_jumble_ignore);
+ int32 vartypmod pg_node_attr(query_jumble_ignore, default(-1));
/* OID of collation, or InvalidOid if none */
Oid varcollid pg_node_attr(query_jumble_ignore);
@@ -297,7 +297,7 @@ typedef struct Const
/* pg_type OID of the constant's datatype */
Oid consttype;
/* typmod value, if any */
- int32 consttypmod pg_node_attr(query_jumble_ignore);
+ int32 consttypmod pg_node_attr(query_jumble_ignore, default(-1));
/* OID of collation, or InvalidOid if none */
Oid constcollid pg_node_attr(query_jumble_ignore);
/* typlen of the constant's datatype */
@@ -363,7 +363,7 @@ typedef struct Param
int paramid; /* numeric ID for parameter */
Oid paramtype; /* pg_type OID of parameter's datatype */
/* typmod value, if known */
- int32 paramtypmod pg_node_attr(query_jumble_ignore);
+ int32 paramtypmod pg_node_attr(query_jumble_ignore, default(-1));
/* OID of collation, or InvalidOid if none */
Oid paramcollid pg_node_attr(query_jumble_ignore);
/* token location, or -1 if unknown */
@@ -621,7 +621,7 @@ typedef struct SubscriptingRef
/* type of the SubscriptingRef's result */
Oid refrestype pg_node_attr(query_jumble_ignore);
/* typmod of the result */
- int32 reftypmod pg_node_attr(query_jumble_ignore);
+ int32 reftypmod pg_node_attr(query_jumble_ignore, default(-1));
/* collation of result, or InvalidOid if none */
Oid refcollid pg_node_attr(query_jumble_ignore);
/* expressions that evaluate to upper container indexes */
@@ -1065,7 +1065,7 @@ typedef struct FieldSelect
/* type of the field (result type of this node) */
Oid resulttype pg_node_attr(query_jumble_ignore);
/* output typmod (usually -1) */
- int32 resulttypmod pg_node_attr(query_jumble_ignore);
+ int32 resulttypmod pg_node_attr(query_jumble_ignore, default(-1));
/* OID of collation of the field */
Oid resultcollid pg_node_attr(query_jumble_ignore);
} FieldSelect;
@@ -1119,7 +1119,7 @@ typedef struct RelabelType
Expr *arg; /* input expression */
Oid resulttype; /* output type of coercion expression */
/* output typmod (usually -1) */
- int32 resulttypmod pg_node_attr(query_jumble_ignore);
+ int32 resulttypmod pg_node_attr(query_jumble_ignore, default(-1));
/* OID of collation, or InvalidOid if none */
Oid resultcollid pg_node_attr(query_jumble_ignore);
/* how to display this node */
@@ -1169,7 +1169,7 @@ typedef struct ArrayCoerceExpr
Expr *elemexpr; /* expression representing per-element work */
Oid resulttype; /* output type of coercion (an array type) */
/* output typmod (also element typmod) */
- int32 resulttypmod pg_node_attr(query_jumble_ignore);
+ int32 resulttypmod pg_node_attr(query_jumble_ignore, default(-1));
/* OID of collation, or InvalidOid if none */
Oid resultcollid pg_node_attr(query_jumble_ignore);
/* how to display this node */
@@ -1495,7 +1495,7 @@ typedef struct SQLValueFunction
* include this Oid in the query jumbling.
*/
Oid type pg_node_attr(query_jumble_ignore);
- int32 typmod;
+ int32 typmod pg_node_attr(default(-1));
int location; /* token location, or -1 if unknown */
} SQLValueFunction;
@@ -1547,7 +1547,7 @@ typedef struct XmlExpr
bool indent;
/* target type/typmod for XMLSERIALIZE */
Oid type pg_node_attr(query_jumble_ignore);
- int32 typmod pg_node_attr(query_jumble_ignore);
+ int32 typmod pg_node_attr(query_jumble_ignore, default(-1));
/* token location, or -1 if unknown */
int location;
} XmlExpr;
@@ -1597,7 +1597,7 @@ typedef struct JsonReturning
NodeTag type;
JsonFormat *format; /* output JSON format */
Oid typid; /* target type Oid */
- int32 typmod; /* target type modifier */
+ int32 typmod pg_node_attr(default(-1)); /* target type modifier */
} JsonReturning;
/*
@@ -1760,7 +1760,7 @@ typedef struct CoerceToDomain
Expr *arg; /* input expression */
Oid resulttype; /* domain type ID (result type) */
/* output typmod (currently always -1) */
- int32 resulttypmod pg_node_attr(query_jumble_ignore);
+ int32 resulttypmod pg_node_attr(query_jumble_ignore, default(-1));
/* OID of collation, or InvalidOid if none */
Oid resultcollid pg_node_attr(query_jumble_ignore);
/* how to display this node */
--
2.40.1
[application/octet-stream] v1-0007-NodeSupport-Apply-RLE-and-differential-encoding-o.patch (6.5K, ../../CAEze2Wigkd1+J4s=7wUqW8Y4g9mDWSC28119ukbKkf799WBpzg@mail.gmail.com/7-v1-0007-NodeSupport-Apply-RLE-and-differential-encoding-o.patch)
download | inline diff:
From 3a129c7765a6066ec164d504bcdc2942ace00959 Mon Sep 17 00:00:00 2001
From: Matthias van de Meent <[email protected]>
Date: Wed, 3 Jan 2024 23:41:44 +0100
Subject: [PATCH v1 7/7] NodeSupport: Apply RLE and differential encoding on
lists and bitmaps
This reduces raw serialized size of the initdb pg_rewrite.ev_action
dataset by ~1.6%, it's pglz-compressed dataset by 5%/15kB (!), and the
size of the template0 database by another 2 blocks, for a total
improvement of 400kB/49 blocks.
---
src/backend/nodes/outfuncs.c | 89 ++++++++++++++++++++++++++++++++---
src/backend/nodes/read.c | 38 +++++++++++++--
src/backend/nodes/readfuncs.c | 15 +++++-
3 files changed, 132 insertions(+), 10 deletions(-)
diff --git a/src/backend/nodes/outfuncs.c b/src/backend/nodes/outfuncs.c
index 83f7e43242..d44e1db3d5 100644
--- a/src/backend/nodes/outfuncs.c
+++ b/src/backend/nodes/outfuncs.c
@@ -337,18 +337,49 @@ static void
_outList(StringInfo str, const List *node)
{
const ListCell *lc;
+ union LCSurrogate {
+ int32 i;
+ Oid o;
+ TransactionId x;
+ } previous = { .i = 0};
+ int run = 0;
+ int run_delta;
+ const char *fmt;
appendStringInfoChar(str, '(');
if (IsA(node, IntList))
+ {
appendStringInfoChar(str, 'i');
+ fmt = " %d";
+ run_delta = 1;
+ }
else if (IsA(node, OidList))
+ {
appendStringInfoChar(str, 'o');
+ fmt = " %u";
+ run_delta = 0;
+ }
else if (IsA(node, XidList))
+ {
appendStringInfoChar(str, 'x');
+ fmt = " %u";
+ run_delta = 1;
+ }
+ else if (IsA(node, List))
+ {
+ /* silence the compiler about uninitialized variables */
+ fmt = "";
+ run_delta = 0;
+ }
+ else
+ elog(ERROR, "unrecognized list node type: %d",
+ (int) node->type);
foreach(lc, node)
{
+ union LCSurrogate val = {.i = 0};
+
/*
* For the sake of backward compatibility, we emit a slightly
* different whitespace format for lists of nodes vs. other types of
@@ -359,18 +390,30 @@ _outList(StringInfo str, const List *node)
outNode(str, lfirst(lc));
if (lnext(node, lc))
appendStringInfoChar(str, ' ');
+ continue;
}
else if (IsA(node, IntList))
- appendStringInfo(str, " %d", lfirst_int(lc));
+ val.i = lfirst_int(lc);
else if (IsA(node, OidList))
- appendStringInfo(str, " %u", lfirst_oid(lc));
+ val.o = lfirst_oid(lc);
else if (IsA(node, XidList))
- appendStringInfo(str, " %u", lfirst_xid(lc));
+ val.x = lfirst_xid(lc);
+
+ if (val.i == previous.i + run_delta)
+ run += 1;
else
- elog(ERROR, "unrecognized list node type: %d",
- (int) node->type);
+ {
+ if (run > 0)
+ appendStringInfo(str, " +%d", run);
+ run = 0;
+ appendStringInfo(str, fmt, val.i);
+ }
+ previous = val;
}
+ if (run > 0)
+ appendStringInfo(str, " +%d", run);
+
appendStringInfoChar(str, ')');
}
@@ -387,12 +430,46 @@ void
outBitmapset(StringInfo str, const Bitmapset *bms)
{
int x;
+ int prev = 0;
+ int run = 0;
appendStringInfoChar(str, '(');
appendStringInfoChar(str, 'b');
x = -1;
while ((x = bms_next_member(bms, x)) >= 0)
- appendStringInfo(str, " %d", x);
+ {
+ int increment = x - prev;
+ if (increment == 1)
+ {
+ run++;
+ }
+ else if (run != 0)
+ {
+ /*
+ * If the run is only one value, the '-' sign of run notation is
+ * larger than just the plain differential value, so only
+ * write out runs longer than 1.
+ */
+ if (run == 1)
+ appendStringInfo(str, " %d", 1);
+ else
+ appendStringInfo(str, " +%d", run);
+
+ appendStringInfo(str, " %d", increment);
+ run = 0;
+ }
+ else
+ {
+ appendStringInfo(str, " %d", increment);
+ }
+ prev = x;
+ }
+
+ if (run == 1)
+ appendStringInfo(str, " %d", run);
+ else if (run > 1)
+ appendStringInfo(str, " +%d", run);
+
appendStringInfoChar(str, ')');
}
diff --git a/src/backend/nodes/read.c b/src/backend/nodes/read.c
index 2815b268be..362672c24e 100644
--- a/src/backend/nodes/read.c
+++ b/src/backend/nodes/read.c
@@ -402,6 +402,8 @@ nodeRead(const char *token, int tok_len)
elog(ERROR, "unterminated List structure");
if (tok_len == 1 && token[0] == 'i')
{
+ int prev = 0;
+
/* List of integers */
for (;;)
{
@@ -417,12 +419,23 @@ nodeRead(const char *token, int tok_len)
if (endptr != token + tok_len)
elog(ERROR, "unrecognized integer: \"%.*s\"",
tok_len, token);
- l = lappend_int(l, val);
+
+ if (val > 0 && token[0] == '+')
+ {
+ for (int i = 0; i < val; i++)
+ l = lappend_int(l, ++prev);
+ }
+ else
+ {
+ prev = val;
+ l = lappend_int(l, val);
+ }
}
result = (Node *) l;
}
else if (tok_len == 1 && token[0] == 'o')
{
+ Oid prev = 0;
/* List of OIDs */
for (;;)
{
@@ -438,12 +451,22 @@ nodeRead(const char *token, int tok_len)
if (endptr != token + tok_len)
elog(ERROR, "unrecognized OID: \"%.*s\"",
tok_len, token);
- l = lappend_oid(l, val);
+ if (token[0] == '+')
+ {
+ for (int i = 0; i < val; i++)
+ l = lappend_oid(l, prev);
+ }
+ else
+ {
+ prev = val;
+ l = lappend_oid(l, val);
+ }
}
result = (Node *) l;
}
else if (tok_len == 1 && token[0] == 'x')
{
+ TransactionId prev = 0;
/* List of TransactionIds */
for (;;)
{
@@ -459,7 +482,16 @@ nodeRead(const char *token, int tok_len)
if (endptr != token + tok_len)
elog(ERROR, "unrecognized Xid: \"%.*s\"",
tok_len, token);
- l = lappend_xid(l, val);
+ if (token[0] == '+')
+ {
+ for (int i = 0; i < val; i++)
+ l = lappend_xid(l, ++prev);
+ }
+ else
+ {
+ prev = val;
+ l = lappend_xid(l, val);
+ }
}
result = (Node *) l;
}
diff --git a/src/backend/nodes/readfuncs.c b/src/backend/nodes/readfuncs.c
index 19c023fb32..ad96ff6646 100644
--- a/src/backend/nodes/readfuncs.c
+++ b/src/backend/nodes/readfuncs.c
@@ -298,6 +298,7 @@ static Bitmapset *
_readBitmapset(void)
{
Bitmapset *result = NULL;
+ int prev = 0;
READ_TEMP_LOCALS();
@@ -326,7 +327,19 @@ _readBitmapset(void)
val = (int) strtol(token, &endptr, 10);
if (endptr != token + length)
elog(ERROR, "unrecognized integer: \"%.*s\"", length, token);
- result = bms_add_member(result, val);
+ if (val > 0 && token[0] == '+')
+ {
+ for (int i = 0; i < val; i++)
+ {
+ prev++;
+ result = bms_add_member(result, prev);
+ }
+ }
+ else
+ {
+ prev += val;
+ result = bms_add_member(result, prev);
+ }
}
return result;
--
2.40.1
[application/octet-stream] v1-0006-NodeSupport-Apply-some-more-defaults-serializatio.patch (16.5K, ../../CAEze2Wigkd1+J4s=7wUqW8Y4g9mDWSC28119ukbKkf799WBpzg@mail.gmail.com/8-v1-0006-NodeSupport-Apply-some-more-defaults-serializatio.patch)
download | inline diff:
From 202203bed6bb5c1965c72710e8df59193b4b5932 Mon Sep 17 00:00:00 2001
From: Matthias van de Meent <[email protected]>
Date: Wed, 3 Jan 2024 15:24:15 +0100
Subject: [PATCH v1 6/7] NodeSupport: Apply some more defaults/serialization
optimizations
This reduces raw serialized size of the initdb pg_rewrite.ev_action
dataset by 7.5%, it's pglz-compressed dataset by 4%/12kB, and the size
of the template0 database by 2 blocks.
---
src/backend/nodes/outfuncs.c | 32 +++++++++++++--------------
src/backend/nodes/readfuncs.c | 37 +++++++++++++++++--------------
src/include/nodes/parsenodes.h | 17 +++++++++------
src/include/nodes/pathnodes.h | 2 +-
src/include/nodes/primnodes.h | 40 +++++++++++++++++++---------------
5 files changed, 70 insertions(+), 58 deletions(-)
diff --git a/src/backend/nodes/outfuncs.c b/src/backend/nodes/outfuncs.c
index f6cfcc8a4c..83f7e43242 100644
--- a/src/backend/nodes/outfuncs.c
+++ b/src/backend/nodes/outfuncs.c
@@ -457,18 +457,18 @@ _outConst(StringInfo str, const Const *node)
WRITE_NODE_TYPE("CONST");
WRITE_OID_FIELD(consttype);
- WRITE_INT_FIELD(consttypmod);
+ WRITE_INT_FIELD_DEFAULT(consttypmod, -1);
WRITE_OID_FIELD(constcollid);
- WRITE_INT_FIELD(constlen);
+ WRITE_INT_FIELD_DEFAULT(constlen, -1);
WRITE_BOOL_FIELD(constbyval);
WRITE_BOOL_FIELD(constisnull);
WRITE_LOCATION_FIELD(location);
- appendStringInfoString(str, " :constvalue ");
- if (node->constisnull)
- appendStringInfoString(str, "<>");
- else
+ if (!node->constisnull)
+ {
+ appendStringInfoString(str, " :constvalue ");
outDatum(str, node->constvalue, node->constlen, node->constbyval);
+ }
}
static void
@@ -579,10 +579,10 @@ _outRangeTblEntry(StringInfo str, const RangeTblEntry *node)
{
case RTE_RELATION:
WRITE_OID_FIELD(relid);
- WRITE_CHAR_FIELD(relkind);
- WRITE_INT_FIELD(rellockmode);
+ WRITE_CHAR_FIELD_DEFAULT(relkind, 'r'); /* default: 'r'elation */
+ WRITE_INT_FIELD_DEFAULT(rellockmode, 1); /* AccessShareLock */
WRITE_NODE_FIELD(tablesample);
- WRITE_UINT_FIELD(perminfoindex);
+ WRITE_UINT_FIELD_DEFAULT(perminfoindex, 1); /* Index starts at 1 */
break;
case RTE_SUBQUERY:
WRITE_NODE_FIELD(subquery);
@@ -591,7 +591,7 @@ _outRangeTblEntry(StringInfo str, const RangeTblEntry *node)
WRITE_OID_FIELD(relid);
WRITE_CHAR_FIELD(relkind);
WRITE_INT_FIELD(rellockmode);
- WRITE_UINT_FIELD(perminfoindex);
+ WRITE_UINT_FIELD_DEFAULT(perminfoindex, 1); /* Index starts at 1 */
break;
case RTE_JOIN:
WRITE_ENUM_FIELD(jointype, JoinType);
@@ -640,8 +640,8 @@ _outRangeTblEntry(StringInfo str, const RangeTblEntry *node)
}
WRITE_BOOL_FIELD(lateral);
- WRITE_BOOL_FIELD(inh);
- WRITE_BOOL_FIELD(inFromCl);
+ WRITE_BOOL_FIELD_DEFAULT(inh, true); /* safe assumption */
+ WRITE_BOOL_FIELD_DEFAULT(inFromCl, true); /* safe assumption */
WRITE_NODE_FIELD(securityQuals);
}
@@ -798,7 +798,7 @@ _outConstraint(StringInfo str, const Constraint *node)
WRITE_INT_FIELD(inhcount);
WRITE_BOOL_FIELD(is_no_inherit);
WRITE_BOOL_FIELD(skip_validation);
- WRITE_BOOL_FIELD(initially_valid);
+ WRITE_BOOL_FIELD_DEFAULT(initially_valid, true);
break;
case CONSTR_DEFAULT:
@@ -826,7 +826,7 @@ _outConstraint(StringInfo str, const Constraint *node)
WRITE_NODE_FIELD(raw_expr);
WRITE_STRING_FIELD(cooked_expr);
WRITE_BOOL_FIELD(skip_validation);
- WRITE_BOOL_FIELD(initially_valid);
+ WRITE_BOOL_FIELD_DEFAULT(initially_valid, true);
break;
case CONSTR_PRIMARY:
@@ -842,7 +842,7 @@ _outConstraint(StringInfo str, const Constraint *node)
case CONSTR_UNIQUE:
appendStringInfoString(str, "UNIQUE");
- WRITE_BOOL_FIELD(nulls_not_distinct);
+ WRITE_BOOL_FIELD_DEFAULT(nulls_not_distinct, true);
WRITE_NODE_FIELD(keys);
WRITE_NODE_FIELD(including);
WRITE_NODE_FIELD(options);
@@ -876,7 +876,7 @@ _outConstraint(StringInfo str, const Constraint *node)
WRITE_NODE_FIELD(old_conpfeqop);
WRITE_OID_FIELD(old_pktable_oid);
WRITE_BOOL_FIELD(skip_validation);
- WRITE_BOOL_FIELD(initially_valid);
+ WRITE_BOOL_FIELD_DEFAULT(initially_valid, true);
break;
case CONSTR_ATTR_DEFERRABLE:
diff --git a/src/backend/nodes/readfuncs.c b/src/backend/nodes/readfuncs.c
index 4287a7ec6e..19c023fb32 100644
--- a/src/backend/nodes/readfuncs.c
+++ b/src/backend/nodes/readfuncs.c
@@ -356,18 +356,23 @@ _readConst(void)
READ_LOCALS(Const);
READ_OID_FIELD(consttype);
- READ_INT_FIELD(consttypmod);
+ READ_INT_FIELD_DEFAULT(consttypmod, -1);
READ_OID_FIELD(constcollid);
- READ_INT_FIELD(constlen);
+ READ_INT_FIELD_DEFAULT(constlen, -1);
READ_BOOL_FIELD(constbyval);
READ_BOOL_FIELD(constisnull);
READ_LOCATION_FIELD(location);
- token = pg_strtok(&length); /* skip :constvalue */
- if (local_node->constisnull)
- token = pg_strtok(&length); /* skip "<>" */
- else
+ if (pg_strtoken_next(":constvalue"))
+ {
+ token = pg_strtok(&length); /* skip :constvalue */
+ Assert(strncmp(token, ":constvalue", sizeof(":constvalue") - 1) == 0);
local_node->constvalue = readDatum(local_node->constbyval);
+ }
+ else
+ {
+ /* value was omitted */
+ }
READ_DONE();
}
@@ -493,7 +498,7 @@ _readConstraint(void)
READ_INT_FIELD(inhcount);
READ_BOOL_FIELD(is_no_inherit);
READ_BOOL_FIELD(skip_validation);
- READ_BOOL_FIELD(initially_valid);
+ READ_BOOL_FIELD_DEFAULT(initially_valid, true);
break;
case CONSTR_DEFAULT:
@@ -517,7 +522,7 @@ _readConstraint(void)
READ_NODE_FIELD(raw_expr);
READ_STRING_FIELD(cooked_expr);
READ_BOOL_FIELD(skip_validation);
- READ_BOOL_FIELD(initially_valid);
+ READ_BOOL_FIELD_DEFAULT(initially_valid, true);
break;
case CONSTR_PRIMARY:
@@ -531,7 +536,7 @@ _readConstraint(void)
break;
case CONSTR_UNIQUE:
- READ_BOOL_FIELD(nulls_not_distinct);
+ READ_BOOL_FIELD_DEFAULT(nulls_not_distinct, true);
READ_NODE_FIELD(keys);
READ_NODE_FIELD(including);
READ_NODE_FIELD(options);
@@ -563,7 +568,7 @@ _readConstraint(void)
READ_NODE_FIELD(old_conpfeqop);
READ_OID_FIELD(old_pktable_oid);
READ_BOOL_FIELD(skip_validation);
- READ_BOOL_FIELD(initially_valid);
+ READ_BOOL_FIELD_DEFAULT(initially_valid, true);
break;
case CONSTR_ATTR_DEFERRABLE:
@@ -595,10 +600,10 @@ _readRangeTblEntry(void)
{
case RTE_RELATION:
READ_OID_FIELD(relid);
- READ_CHAR_FIELD(relkind);
- READ_INT_FIELD(rellockmode);
+ READ_CHAR_FIELD_DEFAULT(relkind, 'r');
+ READ_INT_FIELD_DEFAULT(rellockmode, 1);
READ_NODE_FIELD(tablesample);
- READ_UINT_FIELD(perminfoindex);
+ READ_UINT_FIELD_DEFAULT(perminfoindex, 1);
break;
case RTE_SUBQUERY:
READ_NODE_FIELD(subquery);
@@ -607,7 +612,7 @@ _readRangeTblEntry(void)
READ_OID_FIELD(relid);
READ_CHAR_FIELD(relkind);
READ_INT_FIELD(rellockmode);
- READ_UINT_FIELD(perminfoindex);
+ READ_UINT_FIELD_DEFAULT(perminfoindex, 1);
break;
case RTE_JOIN:
READ_ENUM_FIELD(jointype, JoinType);
@@ -666,8 +671,8 @@ _readRangeTblEntry(void)
}
READ_BOOL_FIELD(lateral);
- READ_BOOL_FIELD(inh);
- READ_BOOL_FIELD(inFromCl);
+ READ_BOOL_FIELD_DEFAULT(inh, true); /* safe assumption */
+ READ_BOOL_FIELD_DEFAULT(inFromCl, true); /* safe assumption */
READ_NODE_FIELD(securityQuals);
READ_DONE();
diff --git a/src/include/nodes/parsenodes.h b/src/include/nodes/parsenodes.h
index e11b044b89..dd6508ae84 100644
--- a/src/include/nodes/parsenodes.h
+++ b/src/include/nodes/parsenodes.h
@@ -261,7 +261,7 @@ typedef struct TypeName
bool setof; /* is a set? */
bool pct_type; /* %TYPE specified? */
List *typmods; /* type modifier expression(s) */
- int32 typemod; /* prespecified type modifier */
+ int32 typemod pg_node_attr(default(-1)); /* prespecified type modifier */
List *arrayBounds; /* array bounds */
int location; /* token location, or -1 if unknown */
} TypeName;
@@ -556,7 +556,8 @@ typedef struct WindowDef
char *refname; /* referenced window name, if any */
List *partitionClause; /* PARTITION BY expression list */
List *orderClause; /* ORDER BY (list of SortBy) */
- int frameOptions; /* frame_clause options, see below */
+ /* frame_clause options, see below */
+ int frameOptions pg_node_attr(default(FRAMEOPTION_DEFAULTS));
Node *startOffset; /* expression for starting bound, if any */
Node *endOffset; /* expression for ending bound, if any */
int location; /* parse location, or -1 if none/unknown */
@@ -1235,7 +1236,8 @@ typedef struct RTEPermissionInfo
NodeTag type;
Oid relid; /* relation OID */
- bool inh; /* separately check inheritance children? */
+ /* separately check inheritance children? */
+ bool inh pg_node_attr(default(true));
/* bitmask of required access permissions, in views usually ACL_SELECT */
AclMode requiredPerms pg_node_attr(default(ACL_SELECT));
Oid checkAsUser; /* if valid, check access as this role */
@@ -1388,8 +1390,8 @@ typedef struct SortGroupClause
Oid eqop; /* the equality operator ('=' op) */
Oid sortop; /* the ordering operator ('<' op), or 0 */
bool nulls_first; /* do NULLs come before normal values? */
- /* can eqop be implemented by hashing? */
- bool hashable pg_node_attr(query_jumble_ignore);
+ /* can eqop be implemented by hashing? (often true for builtin operators) */
+ bool hashable pg_node_attr(query_jumble_ignore, default(true));
} SortGroupClause;
/*
@@ -1491,7 +1493,8 @@ typedef struct WindowClause
List *partitionClause; /* PARTITION BY list */
/* ORDER BY list */
List *orderClause;
- int frameOptions; /* frame_clause options, see WindowDef */
+ /* frame_clause options, see WindowDef */
+ int frameOptions pg_node_attr(default(FRAMEOPTION_DEFAULTS));
Node *startOffset; /* expression for starting bound, if any */
Node *endOffset; /* expression for ending bound, if any */
/* qual to help short-circuit execution */
@@ -2025,7 +2028,7 @@ typedef struct SelectStmt
typedef struct SetOperationStmt
{
NodeTag type;
- SetOperation op; /* type of set op */
+ SetOperation op pg_node_attr(default(1)); /* type of set op */
bool all; /* ALL specified? */
Node *larg; /* left child */
Node *rarg; /* right child */
diff --git a/src/include/nodes/pathnodes.h b/src/include/nodes/pathnodes.h
index e688fc2b79..78ee667704 100644
--- a/src/include/nodes/pathnodes.h
+++ b/src/include/nodes/pathnodes.h
@@ -3380,7 +3380,7 @@ typedef struct AggTransInfo
/* Additional data about transtype */
int32 aggtranstypmod pg_node_attr(default(-1));
- int transtypeLen;
+ int transtypeLen pg_node_attr(default(-1));
bool transtypeByVal;
/* Space-consumption estimate */
diff --git a/src/include/nodes/primnodes.h b/src/include/nodes/primnodes.h
index acb275900f..bb11231554 100644
--- a/src/include/nodes/primnodes.h
+++ b/src/include/nodes/primnodes.h
@@ -245,7 +245,7 @@ typedef struct Var
/*
* attribute number of this var, or zero for all attrs ("whole-row Var")
*/
- AttrNumber varattno;
+ AttrNumber varattno pg_node_attr(default(1));
/* pg_type OID for the type of this var */
Oid vartype pg_node_attr(query_jumble_ignore);
@@ -305,7 +305,7 @@ typedef struct Const
/* OID of collation, or InvalidOid if none */
Oid constcollid pg_node_attr(query_jumble_ignore);
/* typlen of the constant's datatype */
- int constlen pg_node_attr(query_jumble_ignore);
+ int constlen pg_node_attr(query_jumble_ignore, default(-1));
/* the constant's value */
Datum constvalue pg_node_attr(query_jumble_ignore);
/* whether the constant is null (if true, constvalue is undefined) */
@@ -364,7 +364,8 @@ typedef struct Param
{
Expr xpr;
ParamKind paramkind; /* kind of parameter. See above */
- int paramid; /* numeric ID for parameter */
+ /* numeric ID for parameter */
+ int paramid pg_node_attr(default(1));
Oid paramtype; /* pg_type OID of parameter's datatype */
/* typmod value, if known */
int32 paramtypmod pg_node_attr(query_jumble_ignore, default(-1));
@@ -476,7 +477,7 @@ typedef struct Aggref
bool aggvariadic pg_node_attr(query_jumble_ignore);
/* aggregate kind (see pg_aggregate.h) */
- char aggkind pg_node_attr(query_jumble_ignore);
+ char aggkind pg_node_attr(query_jumble_ignore, default('n'));
/* aggregate input already sorted */
bool aggpresorted pg_node_attr(equal_ignore, query_jumble_ignore);
@@ -760,7 +761,7 @@ typedef struct OpExpr
Oid opfuncid pg_node_attr(equal_ignore_if_zero, query_jumble_ignore);
/* PG_TYPE OID of result value */
- Oid opresulttype pg_node_attr(query_jumble_ignore);
+ Oid opresulttype pg_node_attr(query_jumble_ignore, default(BOOLOID));
/* true if operator returns set */
bool opretset pg_node_attr(query_jumble_ignore);
@@ -845,8 +846,8 @@ typedef struct ScalarArrayOpExpr
/* PG_PROC OID of negator of opfuncid function or InvalidOid. See above */
Oid negfuncid pg_node_attr(equal_ignore_if_zero, query_jumble_ignore);
- /* true for ANY, false for ALL */
- bool useOr;
+ /* true for ANY, false for ALL - usually users use ANY */
+ bool useOr pg_node_attr(default(true));
/* OID of collation that operator should use */
Oid inputcollid pg_node_attr(query_jumble_ignore);
@@ -1127,7 +1128,7 @@ typedef struct RelabelType
/* OID of collation, or InvalidOid if none */
Oid resultcollid pg_node_attr(query_jumble_ignore);
/* how to display this node */
- CoercionForm relabelformat pg_node_attr(query_jumble_ignore);
+ CoercionForm relabelformat pg_node_attr(query_jumble_ignore, default(COERCE_EXPLICIT_CAST));
int location; /* token location, or -1 if unknown */
} RelabelType;
@@ -1148,8 +1149,8 @@ typedef struct CoerceViaIO
/* output typmod is not stored, but is presumed -1 */
/* OID of collation, or InvalidOid if none */
Oid resultcollid pg_node_attr(query_jumble_ignore);
- /* how to display this node */
- CoercionForm coerceformat pg_node_attr(query_jumble_ignore);
+ /* how to display this node, in pg_rewrite this is generally an implicit cast */
+ CoercionForm coerceformat pg_node_attr(query_jumble_ignore, default(COERCE_IMPLICIT_CAST));
int location; /* token location, or -1 if unknown */
} CoerceViaIO;
@@ -1176,8 +1177,9 @@ typedef struct ArrayCoerceExpr
int32 resulttypmod pg_node_attr(query_jumble_ignore, default(-1));
/* OID of collation, or InvalidOid if none */
Oid resultcollid pg_node_attr(query_jumble_ignore);
- /* how to display this node */
- CoercionForm coerceformat pg_node_attr(query_jumble_ignore);
+ /* How to display this node. Arrays don't often have implicit casts, so
+ * we default to explicit casts */
+ CoercionForm coerceformat pg_node_attr(query_jumble_ignore, default(COERCE_EXPLICIT_CAST));
int location; /* token location, or -1 if unknown */
} ArrayCoerceExpr;
@@ -1293,7 +1295,7 @@ typedef struct CaseTestExpr
Expr xpr;
Oid typeId; /* type for substituted value */
/* typemod for substituted value */
- int32 typeMod pg_node_attr(query_jumble_ignore);
+ int32 typeMod pg_node_attr(query_jumble_ignore, default(-1));
/* collation for the substituted value */
Oid collation pg_node_attr(query_jumble_ignore);
} CaseTestExpr;
@@ -1702,7 +1704,8 @@ typedef struct NullTest
{
Expr xpr;
Expr *arg; /* input expression */
- NullTestType nulltesttype; /* IS NULL, IS NOT NULL */
+ /* IS NULL, IS NOT NULL */
+ NullTestType nulltesttype pg_node_attr(default(IS_NOT_NULL));
/* T to perform field-by-field null checks */
bool argisrow pg_node_attr(query_jumble_ignore);
int location; /* token location, or -1 if unknown */
@@ -1789,7 +1792,7 @@ typedef struct CoerceToDomainValue
/* type for substituted value */
Oid typeId;
/* typemod for substituted value */
- int32 typeMod pg_node_attr(query_jumble_ignore);
+ int32 typeMod pg_node_attr(query_jumble_ignore, default(-1));
/* collation for the substituted value */
Oid collation pg_node_attr(query_jumble_ignore);
/* token location, or -1 if unknown */
@@ -1809,7 +1812,7 @@ typedef struct SetToDefault
/* type for substituted value */
Oid typeId;
/* typemod for substituted value */
- int32 typeMod pg_node_attr(query_jumble_ignore);
+ int32 typeMod pg_node_attr(query_jumble_ignore, default(-1));
/* collation for the substituted value */
Oid collation pg_node_attr(query_jumble_ignore);
/* token location, or -1 if unknown */
@@ -1926,8 +1929,9 @@ typedef struct TargetEntry
Expr xpr;
/* expression to evaluate */
Expr *expr;
- /* attribute number (see notes above) */
- AttrNumber resno;
+ /* attribute number (see notes above).
+ * Counting starts at 1, so that's the default for serialization, too. */
+ AttrNumber resno pg_node_attr(default(1));
/* name of the column (could be NULL) */
char *resname pg_node_attr(query_jumble_ignore);
/* nonzero if referenced by a sort/group clause */
--
2.40.1
^ permalink raw reply [nested|flat] 70+ messages in thread
end of thread, other threads:[~2024-01-03 23:23 UTC | newest]
Thread overview: 70+ messages (download: mbox mbox.gz follow: Atom feed)
-- links below jump to the message on this page --
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2021-02-02 00:57 [PATCH 7/9] Remove the special batch mode, use a larger buffer always Tomas Vondra <[email protected]>
2023-12-06 21:08 Reducing output size of nodeToString Matthias van de Meent <[email protected]>
2023-12-07 12:09 ` Re: Reducing output size of nodeToString David Rowley <[email protected]>
2024-01-02 10:30 ` Re: Reducing output size of nodeToString Peter Eisentraut <[email protected]>
2024-01-03 23:23 ` Re: Reducing output size of nodeToString Matthias van de Meent <[email protected]>
This inbox is served by agora; see mirroring instructions
for how to clone and mirror all data and code used for this inbox