($INBOX_DIR/description missing)
help / color / mirror / Atom feed[PATCH 3/9] Optimize allocations in bringetbitmap
282+ messages / 5 participants
[nested] [flat]
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------310A2AE1CC4C2E2E77559E3D
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114-2.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114-2.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------CF71AF65F7C337C37C24B045
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210112.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210112.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------FA974B75A0C3E9CA18CE7207
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210122.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210122.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 4dd29c7b10..470431010d 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------291E0481B4400C0ECC8E497A
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20201108.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20201108.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D21F32E0F9E7408E9BE1AF2E
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------FA974B75A0C3E9CA18CE7207
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210122.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210122.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D21F32E0F9E7408E9BE1AF2E
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210215.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210215.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D21F32E0F9E7408E9BE1AF2E
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------310A2AE1CC4C2E2E77559E3D
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114-2.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114-2.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D21F32E0F9E7408E9BE1AF2E
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------CF71AF65F7C337C37C24B045
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210112.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210112.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------310A2AE1CC4C2E2E77559E3D
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114-2.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114-2.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------FA974B75A0C3E9CA18CE7207
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210122.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210122.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------FA974B75A0C3E9CA18CE7207
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210122.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210122.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210215.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210215.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D21F32E0F9E7408E9BE1AF2E
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------310A2AE1CC4C2E2E77559E3D
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114-2.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114-2.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------310A2AE1CC4C2E2E77559E3D
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114-2.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114-2.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------310A2AE1CC4C2E2E77559E3D
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114-2.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114-2.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 4dd29c7b10..470431010d 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------90362739C57C6315B5637FAD
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20201107.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20201107.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------310A2AE1CC4C2E2E77559E3D
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114-2.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114-2.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 4dd29c7b10..470431010d 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------45196B023835614BDFCBD16D
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20201220.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20201220.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210215.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210215.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------FA974B75A0C3E9CA18CE7207
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210122.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210122.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------FA974B75A0C3E9CA18CE7207
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210122.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210122.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------310A2AE1CC4C2E2E77559E3D
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114-2.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114-2.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------FA974B75A0C3E9CA18CE7207
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210122.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210122.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------CF71AF65F7C337C37C24B045
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210112.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210112.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------CF71AF65F7C337C37C24B045
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210112.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210112.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------CF71AF65F7C337C37C24B045
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210112.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210112.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------CF71AF65F7C337C37C24B045
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210112.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210112.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D21F32E0F9E7408E9BE1AF2E
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------310A2AE1CC4C2E2E77559E3D
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114-2.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114-2.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D21F32E0F9E7408E9BE1AF2E
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210215.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210215.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------CF71AF65F7C337C37C24B045
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210112.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210112.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210215.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210215.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------FA974B75A0C3E9CA18CE7207
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210122.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210122.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------310A2AE1CC4C2E2E77559E3D
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114-2.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114-2.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------310A2AE1CC4C2E2E77559E3D
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114-2.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114-2.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------310A2AE1CC4C2E2E77559E3D
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114-2.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114-2.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D21F32E0F9E7408E9BE1AF2E
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------FA974B75A0C3E9CA18CE7207
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210122.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210122.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------CF71AF65F7C337C37C24B045
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210112.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210112.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------FA974B75A0C3E9CA18CE7207
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210122.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210122.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------CF71AF65F7C337C37C24B045
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210112.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210112.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------CF71AF65F7C337C37C24B045
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210112.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210112.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------CF71AF65F7C337C37C24B045
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210112.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210112.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210215.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210215.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------FA974B75A0C3E9CA18CE7207
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210122.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210122.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------310A2AE1CC4C2E2E77559E3D
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114-2.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114-2.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D21F32E0F9E7408E9BE1AF2E
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------FA974B75A0C3E9CA18CE7207
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210122.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210122.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------310A2AE1CC4C2E2E77559E3D
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114-2.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114-2.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------CF71AF65F7C337C37C24B045
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210112.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210112.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D21F32E0F9E7408E9BE1AF2E
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D21F32E0F9E7408E9BE1AF2E
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------FA974B75A0C3E9CA18CE7207
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210122.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210122.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------310A2AE1CC4C2E2E77559E3D
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114-2.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114-2.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------CF71AF65F7C337C37C24B045
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210112.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210112.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210215.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210215.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210215.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210215.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210215.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210215.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------CF71AF65F7C337C37C24B045
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210112.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210112.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D21F32E0F9E7408E9BE1AF2E
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------CF71AF65F7C337C37C24B045
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210112.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210112.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------FA974B75A0C3E9CA18CE7207
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210122.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210122.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210215.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210215.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210215.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210215.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------CF71AF65F7C337C37C24B045
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210112.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210112.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D21F32E0F9E7408E9BE1AF2E
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D21F32E0F9E7408E9BE1AF2E
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D21F32E0F9E7408E9BE1AF2E
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D21F32E0F9E7408E9BE1AF2E
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------310A2AE1CC4C2E2E77559E3D
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114-2.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114-2.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210215.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210215.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------310A2AE1CC4C2E2E77559E3D
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114-2.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114-2.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------310A2AE1CC4C2E2E77559E3D
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114-2.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114-2.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------CF71AF65F7C337C37C24B045
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210112.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210112.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------CF71AF65F7C337C37C24B045
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210112.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210112.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210215.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210215.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210215.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210215.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210215.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210215.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------310A2AE1CC4C2E2E77559E3D
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114-2.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114-2.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------FA974B75A0C3E9CA18CE7207
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210122.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210122.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D21F32E0F9E7408E9BE1AF2E
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210215.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210215.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210215.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210215.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------FA974B75A0C3E9CA18CE7207
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210122.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210122.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210215.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210215.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------CF71AF65F7C337C37C24B045
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210112.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210112.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------FA974B75A0C3E9CA18CE7207
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210122.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210122.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------CF71AF65F7C337C37C24B045
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210112.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210112.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210215.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210215.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------310A2AE1CC4C2E2E77559E3D
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114-2.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114-2.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D21F32E0F9E7408E9BE1AF2E
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210215.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210215.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------FA974B75A0C3E9CA18CE7207
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210122.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210122.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------FA974B75A0C3E9CA18CE7207
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210122.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210122.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------310A2AE1CC4C2E2E77559E3D
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114-2.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114-2.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D21F32E0F9E7408E9BE1AF2E
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210215.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210215.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D21F32E0F9E7408E9BE1AF2E
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------22A4B241170149838D4D1F8F
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210215.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210215.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D21F32E0F9E7408E9BE1AF2E
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210114.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210114.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..c7f7175a2a 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------0E72B707603BED22B4040825
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210211.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 55851376d8..8d23f2864b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -372,6 +372,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -397,11 +400,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -437,9 +479,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -450,17 +492,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------CF71AF65F7C337C37C24B045
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210112.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210112.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/9] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------4B194FF8F3EA3786FF9EAE1F
Content-Type: text/x-patch; charset=UTF-8;
name="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename*0="0002-Move-IS-NOT-NULL-handling-from-BRIN-support-20210203.pa";
filename*1="tch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------FA974B75A0C3E9CA18CE7207
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210122.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210122.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/8] Optimize allocations in bringetbitmap
@ 2020-09-13 10:12 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2020-09-13 10:12 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 62 +++++++++++++++++++++++++++-------
1 file changed, 49 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 14da9ed17f..3735c41788 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -398,11 +401,50 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* Make room for per-attribute lists of scan keys that we'll pass to the
* consistent support procedure. We keep null and regular keys separate,
* so that we can easily pass regular keys to the consistent function.
+ *
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly
+ * the same lifetime, so that's OK.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ /* regular keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ /* NULL keys */
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -438,9 +480,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -451,17 +493,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------FA974B75A0C3E9CA18CE7207
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210122.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210122.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------556A1DC61262AF15641DB204
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308b.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308b.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------556A1DC61262AF15641DB204
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308b.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308b.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------82D677E9F94AC7574E460205
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------82D677E9F94AC7574E460205
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------6A37408A2C42B5B56C3B3D49
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210303.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210303.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------82D677E9F94AC7574E460205
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------556A1DC61262AF15641DB204
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308b.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308b.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 4/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index ce0f525c21..758c47f0f4 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D524CD37E77B36A2883A7617
Content-Type: text/x-patch; charset=UTF-8;
name="0005-BRIN-bloom-indexes-20210311.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0005-BRIN-bloom-indexes-20210311.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 4/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index ce0f525c21..758c47f0f4 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D524CD37E77B36A2883A7617
Content-Type: text/x-patch; charset=UTF-8;
name="0005-BRIN-bloom-indexes-20210311.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0005-BRIN-bloom-indexes-20210311.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------556A1DC61262AF15641DB204
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308b.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308b.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------556A1DC61262AF15641DB204
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308b.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308b.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------AC3FFB734F56C5F52D422EFC
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210305.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210305.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------556A1DC61262AF15641DB204
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308b.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308b.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------82D677E9F94AC7574E460205
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 4/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index ce0f525c21..758c47f0f4 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D524CD37E77B36A2883A7617
Content-Type: text/x-patch; charset=UTF-8;
name="0005-BRIN-bloom-indexes-20210311.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0005-BRIN-bloom-indexes-20210311.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------82D677E9F94AC7574E460205
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------AC3FFB734F56C5F52D422EFC
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210305.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210305.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------6A37408A2C42B5B56C3B3D49
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210303.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210303.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------AC3FFB734F56C5F52D422EFC
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210305.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210305.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------6A37408A2C42B5B56C3B3D49
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210303.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210303.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------AC3FFB734F56C5F52D422EFC
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210305.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210305.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------AC3FFB734F56C5F52D422EFC
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210305.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210305.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------6A37408A2C42B5B56C3B3D49
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210303.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210303.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------556A1DC61262AF15641DB204
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308b.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308b.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------AC3FFB734F56C5F52D422EFC
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210305.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210305.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 4/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index ce0f525c21..758c47f0f4 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D524CD37E77B36A2883A7617
Content-Type: text/x-patch; charset=UTF-8;
name="0005-BRIN-bloom-indexes-20210311.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0005-BRIN-bloom-indexes-20210311.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------82D677E9F94AC7574E460205
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 4/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index ce0f525c21..758c47f0f4 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D524CD37E77B36A2883A7617
Content-Type: text/x-patch; charset=UTF-8;
name="0005-BRIN-bloom-indexes-20210311.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0005-BRIN-bloom-indexes-20210311.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------82D677E9F94AC7574E460205
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------6A37408A2C42B5B56C3B3D49
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210303.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210303.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 4/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index ce0f525c21..758c47f0f4 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D524CD37E77B36A2883A7617
Content-Type: text/x-patch; charset=UTF-8;
name="0005-BRIN-bloom-indexes-20210311.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0005-BRIN-bloom-indexes-20210311.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 4/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index ce0f525c21..758c47f0f4 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D524CD37E77B36A2883A7617
Content-Type: text/x-patch; charset=UTF-8;
name="0005-BRIN-bloom-indexes-20210311.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0005-BRIN-bloom-indexes-20210311.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------6A37408A2C42B5B56C3B3D49
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210303.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210303.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 4/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index ce0f525c21..758c47f0f4 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D524CD37E77B36A2883A7617
Content-Type: text/x-patch; charset=UTF-8;
name="0005-BRIN-bloom-indexes-20210311.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0005-BRIN-bloom-indexes-20210311.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------6A37408A2C42B5B56C3B3D49
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210303.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210303.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------6A37408A2C42B5B56C3B3D49
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210303.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210303.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------6A37408A2C42B5B56C3B3D49
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210303.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210303.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------6A37408A2C42B5B56C3B3D49
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210303.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210303.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------556A1DC61262AF15641DB204
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308b.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308b.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------556A1DC61262AF15641DB204
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308b.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308b.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------82D677E9F94AC7574E460205
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------82D677E9F94AC7574E460205
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------82D677E9F94AC7574E460205
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------556A1DC61262AF15641DB204
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308b.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308b.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------556A1DC61262AF15641DB204
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308b.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308b.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 4/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index ce0f525c21..758c47f0f4 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D524CD37E77B36A2883A7617
Content-Type: text/x-patch; charset=UTF-8;
name="0005-BRIN-bloom-indexes-20210311.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0005-BRIN-bloom-indexes-20210311.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------6A37408A2C42B5B56C3B3D49
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210303.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210303.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 4/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index ce0f525c21..758c47f0f4 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D524CD37E77B36A2883A7617
Content-Type: text/x-patch; charset=UTF-8;
name="0005-BRIN-bloom-indexes-20210311.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0005-BRIN-bloom-indexes-20210311.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 4/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index ce0f525c21..758c47f0f4 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D524CD37E77B36A2883A7617
Content-Type: text/x-patch; charset=UTF-8;
name="0005-BRIN-bloom-indexes-20210311.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0005-BRIN-bloom-indexes-20210311.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------6A37408A2C42B5B56C3B3D49
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210303.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210303.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------AC3FFB734F56C5F52D422EFC
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210305.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210305.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 4/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index ce0f525c21..758c47f0f4 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D524CD37E77B36A2883A7617
Content-Type: text/x-patch; charset=UTF-8;
name="0005-BRIN-bloom-indexes-20210311.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0005-BRIN-bloom-indexes-20210311.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------AC3FFB734F56C5F52D422EFC
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210305.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210305.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------AC3FFB734F56C5F52D422EFC
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210305.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210305.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------AC3FFB734F56C5F52D422EFC
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210305.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210305.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------556A1DC61262AF15641DB204
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308b.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308b.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------82D677E9F94AC7574E460205
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------AC3FFB734F56C5F52D422EFC
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210305.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210305.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 4/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index ce0f525c21..758c47f0f4 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D524CD37E77B36A2883A7617
Content-Type: text/x-patch; charset=UTF-8;
name="0005-BRIN-bloom-indexes-20210311.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0005-BRIN-bloom-indexes-20210311.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------82D677E9F94AC7574E460205
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 4/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index ce0f525c21..758c47f0f4 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D524CD37E77B36A2883A7617
Content-Type: text/x-patch; charset=UTF-8;
name="0005-BRIN-bloom-indexes-20210311.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0005-BRIN-bloom-indexes-20210311.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------82D677E9F94AC7574E460205
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 4/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index ce0f525c21..758c47f0f4 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D524CD37E77B36A2883A7617
Content-Type: text/x-patch; charset=UTF-8;
name="0005-BRIN-bloom-indexes-20210311.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0005-BRIN-bloom-indexes-20210311.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------AC3FFB734F56C5F52D422EFC
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210305.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210305.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------82D677E9F94AC7574E460205
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 4/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index ce0f525c21..758c47f0f4 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D524CD37E77B36A2883A7617
Content-Type: text/x-patch; charset=UTF-8;
name="0005-BRIN-bloom-indexes-20210311.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0005-BRIN-bloom-indexes-20210311.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------AC3FFB734F56C5F52D422EFC
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210305.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210305.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 4/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index ce0f525c21..758c47f0f4 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D524CD37E77B36A2883A7617
Content-Type: text/x-patch; charset=UTF-8;
name="0005-BRIN-bloom-indexes-20210311.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0005-BRIN-bloom-indexes-20210311.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------556A1DC61262AF15641DB204
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308b.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308b.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------6A37408A2C42B5B56C3B3D49
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210303.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210303.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------82D677E9F94AC7574E460205
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------6A37408A2C42B5B56C3B3D49
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210303.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210303.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------556A1DC61262AF15641DB204
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308b.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308b.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------556A1DC61262AF15641DB204
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308b.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308b.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------82D677E9F94AC7574E460205
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------AC3FFB734F56C5F52D422EFC
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210305.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210305.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------6A37408A2C42B5B56C3B3D49
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210303.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210303.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------6A37408A2C42B5B56C3B3D49
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210303.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210303.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------AC3FFB734F56C5F52D422EFC
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210305.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210305.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------82D677E9F94AC7574E460205
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------556A1DC61262AF15641DB204
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308b.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308b.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------82D677E9F94AC7574E460205
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------556A1DC61262AF15641DB204
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308b.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308b.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 4/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index ce0f525c21..758c47f0f4 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D524CD37E77B36A2883A7617
Content-Type: text/x-patch; charset=UTF-8;
name="0005-BRIN-bloom-indexes-20210311.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0005-BRIN-bloom-indexes-20210311.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------6A37408A2C42B5B56C3B3D49
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210303.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210303.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 4/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index ce0f525c21..758c47f0f4 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D524CD37E77B36A2883A7617
Content-Type: text/x-patch; charset=UTF-8;
name="0005-BRIN-bloom-indexes-20210311.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0005-BRIN-bloom-indexes-20210311.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------AC3FFB734F56C5F52D422EFC
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210305.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210305.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------82D677E9F94AC7574E460205
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 4/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index ce0f525c21..758c47f0f4 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D524CD37E77B36A2883A7617
Content-Type: text/x-patch; charset=UTF-8;
name="0005-BRIN-bloom-indexes-20210311.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0005-BRIN-bloom-indexes-20210311.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------556A1DC61262AF15641DB204
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308b.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308b.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------AC3FFB734F56C5F52D422EFC
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210305.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210305.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------AC3FFB734F56C5F52D422EFC
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210305.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210305.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------6A37408A2C42B5B56C3B3D49
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210303.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210303.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------6A37408A2C42B5B56C3B3D49
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210303.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210303.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 4/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index ce0f525c21..758c47f0f4 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D524CD37E77B36A2883A7617
Content-Type: text/x-patch; charset=UTF-8;
name="0005-BRIN-bloom-indexes-20210311.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0005-BRIN-bloom-indexes-20210311.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------82D677E9F94AC7574E460205
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------6A37408A2C42B5B56C3B3D49
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210303.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210303.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------AC3FFB734F56C5F52D422EFC
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210305.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210305.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------AC3FFB734F56C5F52D422EFC
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210305.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210305.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------82D677E9F94AC7574E460205
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 4/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index ce0f525c21..758c47f0f4 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D524CD37E77B36A2883A7617
Content-Type: text/x-patch; charset=UTF-8;
name="0005-BRIN-bloom-indexes-20210311.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0005-BRIN-bloom-indexes-20210311.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------6A37408A2C42B5B56C3B3D49
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210303.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210303.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------6A37408A2C42B5B56C3B3D49
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210303.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210303.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 4/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index ce0f525c21..758c47f0f4 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D524CD37E77B36A2883A7617
Content-Type: text/x-patch; charset=UTF-8;
name="0005-BRIN-bloom-indexes-20210311.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0005-BRIN-bloom-indexes-20210311.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------6A37408A2C42B5B56C3B3D49
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210303.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210303.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 3/6] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------6A37408A2C42B5B56C3B3D49
Content-Type: text/x-patch; charset=UTF-8;
name="0004-BRIN-bloom-indexes-20210303.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0004-BRIN-bloom-indexes-20210303.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------556A1DC61262AF15641DB204
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308b.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308b.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------82D677E9F94AC7574E460205
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------AC3FFB734F56C5F52D422EFC
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210305.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210305.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------AC3FFB734F56C5F52D422EFC
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210305.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210305.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------556A1DC61262AF15641DB204
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308b.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308b.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------AC3FFB734F56C5F52D422EFC
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210305.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210305.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------556A1DC61262AF15641DB204
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308b.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308b.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------AC3FFB734F56C5F52D422EFC
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210305.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210305.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 4/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index ce0f525c21..758c47f0f4 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------D524CD37E77B36A2883A7617
Content-Type: text/x-patch; charset=UTF-8;
name="0005-BRIN-bloom-indexes-20210311.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0005-BRIN-bloom-indexes-20210311.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------82D677E9F94AC7574E460205
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------556A1DC61262AF15641DB204
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308b.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308b.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------556A1DC61262AF15641DB204
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308b.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308b.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------82D677E9F94AC7574E460205
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 5/8] Optimize allocations in bringetbitmap
@ 2021-03-02 18:57 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-02 18:57 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index 9f2656b8d9..1f82e965f9 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -402,15 +405,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest table can have ~1600 attributes, so this may allocate a
* couple kilobytes of memory). We could invent a more compact approach
* (with just space for used attributes) but that would make the matching
* more complicated, so it may not be a win.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/*
* Preprocess the scan keys - split them into per-attribute arrays.
@@ -444,9 +484,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -457,17 +497,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.26.2
--------------556A1DC61262AF15641DB204
Content-Type: text/x-patch; charset=UTF-8;
name="0006-BRIN-bloom-indexes-20210308b.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0006-BRIN-bloom-indexes-20210308b.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 4/8] Optimize allocations in bringetbitmap
@ 2021-03-21 23:24 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-21 23:24 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Reviewed-by: Alvaro Herrera <[email protected]>
Reviewed-by: Mark Dilger <[email protected]>
Reviewed-by: Alexander Korotkov <[email protected]>
Reviewed-by: Masahiko Sawada <[email protected]>
Reviewed-by: John Naylor <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index efda84aba3..0bf25fd05b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -403,15 +406,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest index can have 32 attributes, so the amount of wasted
* memory is negligible. We could invent a more compact approach (with
* just space for used attributes) but that would make the matching more
* complex so it's not a good trade-off.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/* Preprocess the scan keys - split them into per-attribute arrays. */
for (keyno = 0; keyno < scan->numberOfKeys; keyno++)
@@ -443,9 +483,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -456,17 +496,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.30.2
--------------76348C5C2C314FA585652FE5
Content-Type: text/x-patch; charset=UTF-8;
name="0005-BRIN-bloom-indexes-20210322.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0005-BRIN-bloom-indexes-20210322.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* [PATCH 4/8] Optimize allocations in bringetbitmap
@ 2021-03-21 23:24 Tomas Vondra <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Tomas Vondra @ 2021-03-21 23:24 UTC (permalink / raw)
The bringetbitmap function allocates memory for various purposes, which
may be quite expensive, depending on the number of scan keys. Instead of
allocating them separately, allocate one bit chunk of memory an carve it
into smaller pieces as needed - all the pieces have the same lifespan,
and it saves quite a bit of CPU and memory overhead.
Author: Tomas Vondra <[email protected]>
Reviewed-by: Alvaro Herrera <[email protected]>
Reviewed-by: Mark Dilger <[email protected]>
Reviewed-by: Alexander Korotkov <[email protected]>
Reviewed-by: Masahiko Sawada <[email protected]>
Reviewed-by: John Naylor <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/backend/access/brin/brin.c | 60 ++++++++++++++++++++++++++--------
1 file changed, 47 insertions(+), 13 deletions(-)
diff --git a/src/backend/access/brin/brin.c b/src/backend/access/brin/brin.c
index efda84aba3..0bf25fd05b 100644
--- a/src/backend/access/brin/brin.c
+++ b/src/backend/access/brin/brin.c
@@ -373,6 +373,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
int *nkeys,
*nnullkeys;
int keyno;
+ char *ptr;
+ Size len;
+ char *tmp PG_USED_FOR_ASSERTS_ONLY;
opaque = (BrinOpaque *) scan->opaque;
bdesc = opaque->bo_bdesc;
@@ -403,15 +406,52 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
* We keep null and regular keys separate, so that we can pass just the
* regular keys to the consistent function easily.
*
+ * To reduce the allocation overhead, we allocate one big chunk and then
+ * carve it into smaller arrays ourselves. All the pieces have exactly the
+ * same lifetime, so that's OK.
+ *
* XXX The widest index can have 32 attributes, so the amount of wasted
* memory is negligible. We could invent a more compact approach (with
* just space for used attributes) but that would make the matching more
* complex so it's not a good trade-off.
*/
- keys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nullkeys = palloc0(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
- nkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
- nnullkeys = palloc0(sizeof(int) * bdesc->bd_tupdesc->natts);
+ len =
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* regular keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts) +
+ MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts) + /* NULL keys */
+ MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys) * bdesc->bd_tupdesc->natts +
+ MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ ptr = palloc(len);
+ tmp = ptr;
+
+ keys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nullkeys = (ScanKey **) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey *) * bdesc->bd_tupdesc->natts);
+
+ nkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ nnullkeys = (int *) ptr;
+ ptr += MAXALIGN(sizeof(int) * bdesc->bd_tupdesc->natts);
+
+ for (int i = 0; i < bdesc->bd_tupdesc->natts; i++)
+ {
+ keys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+
+ nullkeys[i] = (ScanKey *) ptr;
+ ptr += MAXALIGN(sizeof(ScanKey) * scan->numberOfKeys);
+ }
+
+ Assert(tmp + len == ptr);
+
+ /* zero the number of keys */
+ memset(nkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
+ memset(nnullkeys, 0, sizeof(int) * bdesc->bd_tupdesc->natts);
/* Preprocess the scan keys - split them into per-attribute arrays. */
for (keyno = 0; keyno < scan->numberOfKeys; keyno++)
@@ -443,9 +483,9 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
{
FmgrInfo *tmp;
- /* No key/null arrays for this attribute. */
- Assert((keys[keyattno - 1] == NULL) && (nkeys[keyattno - 1] == 0));
- Assert((nullkeys[keyattno - 1] == NULL) && (nnullkeys[keyattno - 1] == 0));
+ /* First time we see this attribute, so no key/null keys. */
+ Assert(nkeys[keyattno - 1] == 0);
+ Assert(nnullkeys[keyattno - 1] == 0);
tmp = index_getprocinfo(idxRel, keyattno,
BRIN_PROCNUM_CONSISTENT);
@@ -456,17 +496,11 @@ bringetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
/* Add key to the proper per-attribute array. */
if (key->sk_flags & SK_ISNULL)
{
- if (!nullkeys[keyattno - 1])
- nullkeys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
nullkeys[keyattno - 1][nnullkeys[keyattno - 1]] = key;
nnullkeys[keyattno - 1]++;
}
else
{
- if (!keys[keyattno - 1])
- keys[keyattno - 1] = palloc0(sizeof(ScanKey) * scan->numberOfKeys);
-
keys[keyattno - 1][nkeys[keyattno - 1]] = key;
nkeys[keyattno - 1]++;
}
--
2.30.2
--------------76348C5C2C314FA585652FE5
Content-Type: text/x-patch; charset=UTF-8;
name="0005-BRIN-bloom-indexes-20210322.patch"
Content-Transfer-Encoding: 8bit
Content-Disposition: attachment;
filename="0005-BRIN-bloom-indexes-20210322.patch"
^ permalink raw reply [nested|flat] 282+ messages in thread
* Re: Add new error_action COPY ON_ERROR "log"
@ 2024-03-11 05:46 Michael Paquier <[email protected]>
2024-03-11 12:30 ` Re: Add new error_action COPY ON_ERROR "log" Bharath Rupireddy <[email protected]>
0 siblings, 1 reply; 282+ messages in thread
From: Michael Paquier @ 2024-03-11 05:46 UTC (permalink / raw)
To: Bharath Rupireddy <[email protected]>; +Cc: torikoshia <[email protected]>; David G. Johnston <[email protected]>; [email protected]; pgsql-hackers; Alexander Korotkov <[email protected]>
On Sat, Mar 09, 2024 at 12:01:49AM +0530, Bharath Rupireddy wrote:
> If NOTICE per attribute and incrementing num_errors per row is
> implemented, it ends up erroring out with ERROR: missing data for
> column "m" for all-column-empty-row. Shall we treat this ERROR softly
> too if on_error ignore is specified? Or shall we discuss this idea
> separately?
>
> ERROR: missing data for column "m"
> CONTEXT: COPY check_ign_err, line 5: ""
Hmm. I have spent some time looking at the bevahior of ON_ERROR, and
there are two tests in copy2.sql, one for the case where there is more
data than attributes and a second where there is not enough data in a
row that checks for errors.
For example, take this table:
=# create table tab (a int, b int, c int);
CREATE TABLE
This case works, even if the row has clearly not enough attributes.
The first attribute can be parsed, not the second one and this causes
the remaining data of the row to be skipped:
=# copy tab from stdin (delimiter ',', on_error ignore);
Enter data to be copied followed by a newline.
End with a backslash and a period on a line by itself, or an EOF signal.
>> 1,
>> \.
NOTICE: 00000: 1 row was skipped due to data type incompatibility
LOCATION: CopyFrom, copyfrom.c:1314
COPY 0
This case fails. The first and the second attributes can be parsed,
and the line fails because we are missing the last attribute as of a
lack of delimiter:
=# copy tab from stdin (delimiter ',', on_error ignore);
Enter data to be copied followed by a newline.
End with a backslash and a period on a line by itself, or an EOF signal.
>> 1,1
>> \.
ERROR: 22P04: missing data for column "c"
CONTEXT: COPY tab, line 1: "1,1"
This brings a weird edge case for the all-column-empty-row case you
are mentioning once if we try to get information about all the rows we
should expect, but this has as side effect to break a case that's
intended ro work with ON_ERROR, as far as I understand, which is to
skip entirely a raw on the first conversion error we find, even if
there is no data afterwards. I was a bit confused by that first, but
I can also see why it is useful as-is on HEAD.
At the end of the day, this comes down to what is more helpful to the
user. And I'd agree on the side what ON_ERROR does currently, which
is what your patch relies on: on the first conversion failure, give up
and skip the rest of the row because we cannot trust its contents.
That's my way of saying that I am fine with the proposal of your
patch, and that we cannot provide the full state of a row without
making the error stack of COPY more invasive.
Perhaps we could discuss this idea of ensuring that all the attributes
are on a row in a different thread, as you say, but I am not really
convinced that there's a strong need for it at this stage as ON_ERROR
is new to v17. So it does not sound like a bad thing to let it brew
more before implementing more options and make the COPY paths more
complicated than they already are. I suspect that this may trigger
some discussion during the beta/stability phase depending on the
initial feedback. Perhaps I'm wrong, though.
+ <literal>verbose</literal>, a <literal>NOTICE</literal> message
+ containing the line number and column name for each discarded row is
+ emitted.
This should clarify that the column name refers to the attribute where
the input conversion has failed, I guess. Specifying only "column
name" without more context is a bit confusing.
--
Michael
Attachments:
[application/pgp-signature] signature.asc (833B, ../../[email protected]/2-signature.asc)
download
^ permalink raw reply [nested|flat] 282+ messages in thread
* Re: Add new error_action COPY ON_ERROR "log"
2024-03-11 05:46 Re: Add new error_action COPY ON_ERROR "log" Michael Paquier <[email protected]>
@ 2024-03-11 12:30 ` Bharath Rupireddy <[email protected]>
2024-03-12 06:52 ` Re: Add new error_action COPY ON_ERROR "log" Michael Paquier <[email protected]>
0 siblings, 1 reply; 282+ messages in thread
From: Bharath Rupireddy @ 2024-03-11 12:30 UTC (permalink / raw)
To: Michael Paquier <[email protected]>; +Cc: torikoshia <[email protected]>; David G. Johnston <[email protected]>; [email protected]; pgsql-hackers; Alexander Korotkov <[email protected]>
On Mon, Mar 11, 2024 at 11:16 AM Michael Paquier <[email protected]> wrote:
>
> At the end of the day, this comes down to what is more helpful to the
> user. And I'd agree on the side what ON_ERROR does currently, which
> is what your patch relies on: on the first conversion failure, give up
> and skip the rest of the row because we cannot trust its contents.
> That's my way of saying that I am fine with the proposal of your
> patch, and that we cannot provide the full state of a row without
> making the error stack of COPY more invasive.
+1.
> + <literal>verbose</literal>, a <literal>NOTICE</literal> message
> + containing the line number and column name for each discarded row is
> + emitted.
>
> This should clarify that the column name refers to the attribute where
> the input conversion has failed, I guess. Specifying only "column
> name" without more context is a bit confusing.
Done.
Please see the attached v6 patch set.
--
Bharath Rupireddy
PostgreSQL Contributors Team
RDS Open Source Databases
Amazon Web Services: https://aws.amazon.com
Attachments:
[application/octet-stream] v6-0001-Add-LOG_VERBOSITY-option-to-COPY-command.patch (9.5K, ../../CALj2ACUZkGMTz10kC3+LcxjGz6NMN1URTn8+K=k_rtm7=z8zSA@mail.gmail.com/2-v6-0001-Add-LOG_VERBOSITY-option-to-COPY-command.patch)
download | inline diff:
From 78db3e788719fa1f9735f35dad8cc0f7e10a8742 Mon Sep 17 00:00:00 2001
From: Bharath Rupireddy <[email protected]>
Date: Mon, 11 Mar 2024 11:37:49 +0000
Subject: [PATCH v6 1/2] Add LOG_VERBOSITY option to COPY command
This commit adds a new option LOG_VERBOSITY to set the verbosity of
logged messages by COPY command. A value of 'verbose' can be used
to emit more informative messages by the command, while the value
of 'default (which is the default) can be used to not log any
additional messages. More values such as 'terse', 'row_details'
etc. can be added based on the need to the LOG_VERBOSITY option.
An upcoming commit for emitting more info on soft errors by
COPY FROM command with ON_ERROR 'ignore' uses this.
Author: Bharath Rupireddy
Reviewed-by: Michael Paquier, Masahiko Sawada
Reviewed-by: Atsushi Torikoshi
Discussion: https://www.postgresql.org/message-id/CALj2ACXNA0focNeriYRvQQaCGc4CsTuOnFbzF9LqTKNWxuJdhA%40mail.gmail.com
---
doc/src/sgml/ref/copy.sgml | 14 +++++++++++
src/backend/commands/copy.c | 38 +++++++++++++++++++++++++++++
src/bin/psql/tab-complete.c | 6 ++++-
src/include/commands/copy.h | 10 ++++++++
src/test/regress/expected/copy2.out | 8 ++++++
src/test/regress/sql/copy2.sql | 2 ++
src/tools/pgindent/typedefs.list | 1 +
7 files changed, 78 insertions(+), 1 deletion(-)
diff --git a/doc/src/sgml/ref/copy.sgml b/doc/src/sgml/ref/copy.sgml
index 55764fc1f2..67ba6212fe 100644
--- a/doc/src/sgml/ref/copy.sgml
+++ b/doc/src/sgml/ref/copy.sgml
@@ -45,6 +45,7 @@ COPY { <replaceable class="parameter">table_name</replaceable> [ ( <replaceable
FORCE_NULL { ( <replaceable class="parameter">column_name</replaceable> [, ...] ) | * }
ON_ERROR '<replaceable class="parameter">error_action</replaceable>'
ENCODING '<replaceable class="parameter">encoding_name</replaceable>'
+ LOG_VERBOSITY [ <replaceable class="parameter">mode</replaceable> ]
</synopsis>
</refsynopsisdiv>
@@ -415,6 +416,19 @@ COPY { <replaceable class="parameter">table_name</replaceable> [ ( <replaceable
</listitem>
</varlistentry>
+ <varlistentry>
+ <term><literal>LOG_VERBOSITY</literal></term>
+ <listitem>
+ <para>
+ Sets the verbosity of logged messages by <command>COPY</command> command.
+ A <replaceable class="parameter">mode</replaceable> value of
+ <literal>verbose</literal> can be used to emit more informative messages
+ by the command, while value of <literal>default</literal> (which is the
+ default) can be used to not log any additional messages.
+ </para>
+ </listitem>
+ </varlistentry>
+
<varlistentry>
<term><literal>WHERE</literal></term>
<listitem>
diff --git a/src/backend/commands/copy.c b/src/backend/commands/copy.c
index 056b6733c8..23eb8c9c79 100644
--- a/src/backend/commands/copy.c
+++ b/src/backend/commands/copy.c
@@ -428,6 +428,36 @@ defGetCopyOnErrorChoice(DefElem *def, ParseState *pstate, bool is_from)
return COPY_ON_ERROR_STOP; /* keep compiler quiet */
}
+/*
+ * Extract a CopyLogVerbosityChoice value from a DefElem.
+ */
+static CopyLogVerbosityChoice
+defGetCopyLogVerbosityChoice(DefElem *def, ParseState *pstate)
+{
+ char *sval;
+
+ /*
+ * If no parameter value given, assume the default value.
+ */
+ if (def->arg == NULL)
+ return COPY_LOG_VERBOSITY_DEFAULT;
+
+ /*
+ * Allow "default", or "verbose" values.
+ */
+ sval = defGetString(def);
+ if (pg_strcasecmp(sval, "default") == 0)
+ return COPY_LOG_VERBOSITY_DEFAULT;
+ if (pg_strcasecmp(sval, "verbose") == 0)
+ return COPY_LOG_VERBOSITY_VERBOSE;
+
+ ereport(ERROR,
+ (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
+ errmsg("COPY LOG_VERBOSITY \"%s\" not recognized", sval),
+ parser_errposition(pstate, def->location)));
+ return COPY_LOG_VERBOSITY_DEFAULT; /* keep compiler quiet */
+}
+
/*
* Process the statement option list for COPY.
*
@@ -454,6 +484,7 @@ ProcessCopyOptions(ParseState *pstate,
bool freeze_specified = false;
bool header_specified = false;
bool on_error_specified = false;
+ bool log_verbosity_specified = false;
ListCell *option;
/* Support external use for option sanity checking */
@@ -613,6 +644,13 @@ ProcessCopyOptions(ParseState *pstate,
on_error_specified = true;
opts_out->on_error = defGetCopyOnErrorChoice(defel, pstate, is_from);
}
+ else if (strcmp(defel->defname, "log_verbosity") == 0)
+ {
+ if (log_verbosity_specified)
+ errorConflictingDefElem(defel, pstate);
+ log_verbosity_specified = true;
+ opts_out->log_verbosity = defGetCopyLogVerbosityChoice(defel, pstate);
+ }
else
ereport(ERROR,
(errcode(ERRCODE_SYNTAX_ERROR),
diff --git a/src/bin/psql/tab-complete.c b/src/bin/psql/tab-complete.c
index 73133ce735..9305800340 100644
--- a/src/bin/psql/tab-complete.c
+++ b/src/bin/psql/tab-complete.c
@@ -2901,7 +2901,7 @@ psql_completion(const char *text, int start, int end)
COMPLETE_WITH("FORMAT", "FREEZE", "DELIMITER", "NULL",
"HEADER", "QUOTE", "ESCAPE", "FORCE_QUOTE",
"FORCE_NOT_NULL", "FORCE_NULL", "ENCODING", "DEFAULT",
- "ON_ERROR");
+ "ON_ERROR", "LOG_VERBOSITY");
/* Complete COPY <sth> FROM|TO filename WITH (FORMAT */
else if (Matches("COPY|\\copy", MatchAny, "FROM|TO", MatchAny, "WITH", "(", "FORMAT"))
@@ -2911,6 +2911,10 @@ psql_completion(const char *text, int start, int end)
else if (Matches("COPY|\\copy", MatchAny, "FROM|TO", MatchAny, "WITH", "(", "ON_ERROR"))
COMPLETE_WITH("stop", "ignore");
+ /* Complete COPY <sth> FROM filename WITH (LOG_VERBOSITY */
+ else if (Matches("COPY|\\copy", MatchAny, "FROM|TO", MatchAny, "WITH", "(", "LOG_VERBOSITY"))
+ COMPLETE_WITH("default", "verbose");
+
/* Complete COPY <sth> FROM <sth> WITH (<options>) */
else if (Matches("COPY|\\copy", MatchAny, "FROM", MatchAny, "WITH", MatchAny))
COMPLETE_WITH("WHERE");
diff --git a/src/include/commands/copy.h b/src/include/commands/copy.h
index b3da3cb0be..99d183fa4d 100644
--- a/src/include/commands/copy.h
+++ b/src/include/commands/copy.h
@@ -40,6 +40,15 @@ typedef enum CopyOnErrorChoice
COPY_ON_ERROR_IGNORE, /* ignore errors */
} CopyOnErrorChoice;
+/*
+ * Represents verbosity of logged messages by COPY command.
+ */
+typedef enum CopyLogVerbosityChoice
+{
+ COPY_LOG_VERBOSITY_DEFAULT = 0, /* logs no additional messages, default */
+ COPY_LOG_VERBOSITY_VERBOSE, /* logs additional messages */
+} CopyLogVerbosityChoice;
+
/*
* A struct to hold COPY options, in a parsed form. All of these are related
* to formatting, except for 'freeze', which doesn't really belong here, but
@@ -73,6 +82,7 @@ typedef struct CopyFormatOptions
bool *force_null_flags; /* per-column CSV FN flags */
bool convert_selectively; /* do selective binary conversion? */
CopyOnErrorChoice on_error; /* what to do when error happened */
+ CopyLogVerbosityChoice log_verbosity; /* verbosity of logged messages */
List *convert_select; /* list of column names (can be NIL) */
} CopyFormatOptions;
diff --git a/src/test/regress/expected/copy2.out b/src/test/regress/expected/copy2.out
index 25c401ce34..62406ef827 100644
--- a/src/test/regress/expected/copy2.out
+++ b/src/test/regress/expected/copy2.out
@@ -81,6 +81,10 @@ COPY x from stdin (on_error ignore, on_error ignore);
ERROR: conflicting or redundant options
LINE 1: COPY x from stdin (on_error ignore, on_error ignore);
^
+COPY x from stdin (log_verbosity 'default', log_verbosity 'verbose');
+ERROR: conflicting or redundant options
+LINE 1: COPY x from stdin (log_verbosity 'default', log_verbosity 'v...
+ ^
-- incorrect options
COPY x to stdin (format BINARY, delimiter ',');
ERROR: cannot specify DELIMITER in BINARY mode
@@ -108,6 +112,10 @@ COPY x to stdin (format BINARY, on_error unsupported);
ERROR: COPY ON_ERROR cannot be used with COPY TO
LINE 1: COPY x to stdin (format BINARY, on_error unsupported);
^
+COPY x to stdout (log_verbosity 'unsupported');
+ERROR: COPY LOG_VERBOSITY "unsupported" not recognized
+LINE 1: COPY x to stdout (log_verbosity 'unsupported');
+ ^
-- too many columns in column list: should fail
COPY x (a, b, c, d, e, d, c) from stdin;
ERROR: column "d" specified more than once
diff --git a/src/test/regress/sql/copy2.sql b/src/test/regress/sql/copy2.sql
index b5e549e856..5116157cc9 100644
--- a/src/test/regress/sql/copy2.sql
+++ b/src/test/regress/sql/copy2.sql
@@ -67,6 +67,7 @@ COPY x from stdin (force_null (a), force_null (b));
COPY x from stdin (convert_selectively (a), convert_selectively (b));
COPY x from stdin (encoding 'sql_ascii', encoding 'sql_ascii');
COPY x from stdin (on_error ignore, on_error ignore);
+COPY x from stdin (log_verbosity 'default', log_verbosity 'verbose');
-- incorrect options
COPY x to stdin (format BINARY, delimiter ',');
@@ -80,6 +81,7 @@ COPY x to stdin (format CSV, force_not_null(a));
COPY x to stdout (format TEXT, force_null(a));
COPY x to stdin (format CSV, force_null(a));
COPY x to stdin (format BINARY, on_error unsupported);
+COPY x to stdout (log_verbosity 'unsupported');
-- too many columns in column list: should fail
COPY x (a, b, c, d, e, d, c) from stdin;
diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list
index d3a7f75b08..b9b9f8a7f0 100644
--- a/src/tools/pgindent/typedefs.list
+++ b/src/tools/pgindent/typedefs.list
@@ -478,6 +478,7 @@ CopyFromState
CopyFromStateData
CopyHeaderChoice
CopyInsertMethod
+CopyLogVerbosityChoice
CopyMultiInsertBuffer
CopyMultiInsertInfo
CopyOnErrorChoice
--
2.34.1
[application/octet-stream] v6-0002-Add-detailed-info-when-COPY-skips-soft-errors.patch (4.9K, ../../CALj2ACUZkGMTz10kC3+LcxjGz6NMN1URTn8+K=k_rtm7=z8zSA@mail.gmail.com/3-v6-0002-Add-detailed-info-when-COPY-skips-soft-errors.patch)
download | inline diff:
From 6f21d689c73d3d7a2d80fb0e723960832d69d237 Mon Sep 17 00:00:00 2001
From: Bharath Rupireddy <[email protected]>
Date: Mon, 11 Mar 2024 11:53:44 +0000
Subject: [PATCH v6 2/2] Add detailed info when COPY skips soft errors
This commit emits individual info like line number and column name
when COPY skips soft errors. Because, the summary containing the
total rows skipped isn't enough for the users to know what exactly
are the malformed rows in the input data.
Author: Bharath Rupireddy
Reviewed-by: Michael Paquier, Masahiko Sawada
Reviewed-by: Atsushi Torikoshi
Discussion: https://www.postgresql.org/message-id/CALj2ACUk700cYhx1ATRQyRw-fBM%2BaRo6auRAitKGff7XNmYfqQ%40mail.gmail.com
---
doc/src/sgml/ref/copy.sgml | 12 +++++++++---
src/backend/commands/copyfromparse.c | 10 ++++++++++
src/test/regress/expected/copy2.out | 7 ++++++-
src/test/regress/sql/copy2.sql | 4 +++-
4 files changed, 28 insertions(+), 5 deletions(-)
diff --git a/doc/src/sgml/ref/copy.sgml b/doc/src/sgml/ref/copy.sgml
index 67ba6212fe..ed7fdc59fb 100644
--- a/doc/src/sgml/ref/copy.sgml
+++ b/doc/src/sgml/ref/copy.sgml
@@ -398,8 +398,12 @@ COPY { <replaceable class="parameter">table_name</replaceable> [ ( <replaceable
when the <literal>FORMAT</literal> is <literal>text</literal> or <literal>csv</literal>.
</para>
<para>
- A <literal>NOTICE</literal> message containing the ignored row count is emitted at the end
- of the <command>COPY FROM</command> if at least one row was discarded.
+ A <literal>NOTICE</literal> message containing the ignored row count is
+ emitted at the end of the <command>COPY FROM</command> if at least one
+ row was discarded. When <literal>LOG_VERBOSITY</literal> option is set to
+ <literal>verbose</literal>, a <literal>NOTICE</literal> message
+ containing the line number and column name (whose input conversion has
+ failed) is emitted for each discarded row.
</para>
</listitem>
</varlistentry>
@@ -424,7 +428,9 @@ COPY { <replaceable class="parameter">table_name</replaceable> [ ( <replaceable
A <replaceable class="parameter">mode</replaceable> value of
<literal>verbose</literal> can be used to emit more informative messages
by the command, while value of <literal>default</literal> (which is the
- default) can be used to not log any additional messages.
+ default) can be used to not log any additional messages. As an example,
+ see its usage for <command>COPY FROM</command> command when
+ <literal>ON_ERROR</literal> option is set to <literal>ignore</literal>.
</para>
</listitem>
</varlistentry>
diff --git a/src/backend/commands/copyfromparse.c b/src/backend/commands/copyfromparse.c
index 5682d5d054..a7ad6c17c8 100644
--- a/src/backend/commands/copyfromparse.c
+++ b/src/backend/commands/copyfromparse.c
@@ -967,7 +967,17 @@ NextCopyFrom(CopyFromState cstate, ExprContext *econtext,
(Node *) cstate->escontext,
&values[m]))
{
+ Assert(cstate->opts.on_error != COPY_ON_ERROR_STOP);
+
cstate->num_errors++;
+
+ if (cstate->opts.log_verbosity == COPY_LOG_VERBOSITY_VERBOSE)
+ ereport(NOTICE,
+ errmsg("detected data type incompatibility at line number %llu for column %s; COPY %s",
+ (unsigned long long) cstate->cur_lineno,
+ cstate->cur_attname,
+ cstate->cur_relname));
+
return true;
}
diff --git a/src/test/regress/expected/copy2.out b/src/test/regress/expected/copy2.out
index 62406ef827..c6655000e4 100644
--- a/src/test/regress/expected/copy2.out
+++ b/src/test/regress/expected/copy2.out
@@ -737,7 +737,12 @@ CREATE TABLE check_ign_err (n int, m int[], k int);
COPY check_ign_err FROM STDIN WITH (on_error stop);
ERROR: invalid input syntax for type integer: "a"
CONTEXT: COPY check_ign_err, line 2, column n: "a"
-COPY check_ign_err FROM STDIN WITH (on_error ignore);
+-- tests for options on_error and log_verbosity
+COPY check_ign_err FROM STDIN WITH (on_error ignore, log_verbosity 'verbose');
+NOTICE: detected data type incompatibility at line number 2 for column n; COPY check_ign_err
+NOTICE: detected data type incompatibility at line number 3 for column k; COPY check_ign_err
+NOTICE: detected data type incompatibility at line number 4 for column m; COPY check_ign_err
+NOTICE: detected data type incompatibility at line number 5 for column n; COPY check_ign_err
NOTICE: 4 rows were skipped due to data type incompatibility
SELECT * FROM check_ign_err;
n | m | k
diff --git a/src/test/regress/sql/copy2.sql b/src/test/regress/sql/copy2.sql
index 5116157cc9..b637a5b3bb 100644
--- a/src/test/regress/sql/copy2.sql
+++ b/src/test/regress/sql/copy2.sql
@@ -510,7 +510,9 @@ a {2} 2
5 {5} 5
\.
-COPY check_ign_err FROM STDIN WITH (on_error ignore);
+
+-- tests for options on_error and log_verbosity
+COPY check_ign_err FROM STDIN WITH (on_error ignore, log_verbosity 'verbose');
1 {1} 1
a {2} 2
3 {3} 3333333333
--
2.34.1
^ permalink raw reply [nested|flat] 282+ messages in thread
* Re: Add new error_action COPY ON_ERROR "log"
2024-03-11 05:46 Re: Add new error_action COPY ON_ERROR "log" Michael Paquier <[email protected]>
2024-03-11 12:30 ` Re: Add new error_action COPY ON_ERROR "log" Bharath Rupireddy <[email protected]>
@ 2024-03-12 06:52 ` Michael Paquier <[email protected]>
2024-03-12 07:24 ` Re: Add new error_action COPY ON_ERROR "log" Bharath Rupireddy <[email protected]>
0 siblings, 1 reply; 282+ messages in thread
From: Michael Paquier @ 2024-03-12 06:52 UTC (permalink / raw)
To: Bharath Rupireddy <[email protected]>; +Cc: torikoshia <[email protected]>; David G. Johnston <[email protected]>; [email protected]; pgsql-hackers; Alexander Korotkov <[email protected]>
On Mon, Mar 11, 2024 at 06:00:00PM +0530, Bharath Rupireddy wrote:
> Please see the attached v6 patch set.
I am tempted to tweak a few things in the docs, the comments and the
tests (particularly adding more patterns for tuples that fail on
conversion while it's clear that there are not enough attributes after
the incorrect values), but that looks roughly OK.
Wouldn't it be better to squash the patches together, by the way?
--
Michael
Attachments:
[application/pgp-signature] signature.asc (833B, ../../[email protected]/2-signature.asc)
download
^ permalink raw reply [nested|flat] 282+ messages in thread
* Re: Add new error_action COPY ON_ERROR "log"
2024-03-11 05:46 Re: Add new error_action COPY ON_ERROR "log" Michael Paquier <[email protected]>
2024-03-11 12:30 ` Re: Add new error_action COPY ON_ERROR "log" Bharath Rupireddy <[email protected]>
2024-03-12 06:52 ` Re: Add new error_action COPY ON_ERROR "log" Michael Paquier <[email protected]>
@ 2024-03-12 07:24 ` Bharath Rupireddy <[email protected]>
2024-03-12 23:46 ` Re: Add new error_action COPY ON_ERROR "log" Michael Paquier <[email protected]>
0 siblings, 1 reply; 282+ messages in thread
From: Bharath Rupireddy @ 2024-03-12 07:24 UTC (permalink / raw)
To: Michael Paquier <[email protected]>; +Cc: torikoshia <[email protected]>; David G. Johnston <[email protected]>; [email protected]; pgsql-hackers; Alexander Korotkov <[email protected]>
On Tue, Mar 12, 2024 at 12:22 PM Michael Paquier <[email protected]> wrote:
>
> On Mon, Mar 11, 2024 at 06:00:00PM +0530, Bharath Rupireddy wrote:
> > Please see the attached v6 patch set.
>
> I am tempted to tweak a few things in the docs, the comments and the
> tests (particularly adding more patterns for tuples that fail on
> conversion while it's clear that there are not enough attributes after
> the incorrect values), but that looks roughly OK.
+1. But, do you want to add them now or later as a separate
patch/discussion altogether?
> Wouldn't it be better to squash the patches together, by the way?
I guess not. They are two different things IMV.
--
Bharath Rupireddy
PostgreSQL Contributors Team
RDS Open Source Databases
Amazon Web Services: https://aws.amazon.com
^ permalink raw reply [nested|flat] 282+ messages in thread
* Re: Add new error_action COPY ON_ERROR "log"
2024-03-11 05:46 Re: Add new error_action COPY ON_ERROR "log" Michael Paquier <[email protected]>
2024-03-11 12:30 ` Re: Add new error_action COPY ON_ERROR "log" Bharath Rupireddy <[email protected]>
2024-03-12 06:52 ` Re: Add new error_action COPY ON_ERROR "log" Michael Paquier <[email protected]>
2024-03-12 07:24 ` Re: Add new error_action COPY ON_ERROR "log" Bharath Rupireddy <[email protected]>
@ 2024-03-12 23:46 ` Michael Paquier <[email protected]>
2024-03-13 02:38 ` Re: Add new error_action COPY ON_ERROR "log" Bharath Rupireddy <[email protected]>
0 siblings, 1 reply; 282+ messages in thread
From: Michael Paquier @ 2024-03-12 23:46 UTC (permalink / raw)
To: Bharath Rupireddy <[email protected]>; +Cc: torikoshia <[email protected]>; David G. Johnston <[email protected]>; [email protected]; pgsql-hackers; Alexander Korotkov <[email protected]>
On Tue, Mar 12, 2024 at 12:54:29PM +0530, Bharath Rupireddy wrote:
> +1. But, do you want to add them now or later as a separate
> patch/discussion altogether?
The attached 0003 is what I had in mind:
- Simplification of the LOG generated with quotes applied around the
column name, don't see much need to add the relation name, either, for
consistency and because the knowledge is known in the query.
- A few more tests.
- Some doc changes.
>> Wouldn't it be better to squash the patches together, by the way?
>
> I guess not. They are two different things IMV.
Well, 0001 is sitting doing nothing because the COPY code does not
make use of it internally.
--
Michael
Attachments:
[text/x-diff] v6-0001-Add-LOG_VERBOSITY-option-to-COPY-command.patch (9.5K, ../../[email protected]/2-v6-0001-Add-LOG_VERBOSITY-option-to-COPY-command.patch)
download | inline diff:
From c45474726e084faf876a319485995ce84eef8293 Mon Sep 17 00:00:00 2001
From: Bharath Rupireddy <[email protected]>
Date: Mon, 11 Mar 2024 11:37:49 +0000
Subject: [PATCH v6 1/3] Add LOG_VERBOSITY option to COPY command
This commit adds a new option LOG_VERBOSITY to set the verbosity of
logged messages by COPY command. A value of 'verbose' can be used
to emit more informative messages by the command, while the value
of 'default (which is the default) can be used to not log any
additional messages. More values such as 'terse', 'row_details'
etc. can be added based on the need to the LOG_VERBOSITY option.
An upcoming commit for emitting more info on soft errors by
COPY FROM command with ON_ERROR 'ignore' uses this.
Author: Bharath Rupireddy
Reviewed-by: Michael Paquier, Masahiko Sawada
Reviewed-by: Atsushi Torikoshi
Discussion: https://www.postgresql.org/message-id/CALj2ACXNA0focNeriYRvQQaCGc4CsTuOnFbzF9LqTKNWxuJdhA%40mail.gmail.com
---
src/include/commands/copy.h | 10 ++++++++
src/backend/commands/copy.c | 38 +++++++++++++++++++++++++++++
src/bin/psql/tab-complete.c | 6 ++++-
src/test/regress/expected/copy2.out | 8 ++++++
src/test/regress/sql/copy2.sql | 2 ++
doc/src/sgml/ref/copy.sgml | 14 +++++++++++
src/tools/pgindent/typedefs.list | 1 +
7 files changed, 78 insertions(+), 1 deletion(-)
diff --git a/src/include/commands/copy.h b/src/include/commands/copy.h
index b3da3cb0be..99d183fa4d 100644
--- a/src/include/commands/copy.h
+++ b/src/include/commands/copy.h
@@ -40,6 +40,15 @@ typedef enum CopyOnErrorChoice
COPY_ON_ERROR_IGNORE, /* ignore errors */
} CopyOnErrorChoice;
+/*
+ * Represents verbosity of logged messages by COPY command.
+ */
+typedef enum CopyLogVerbosityChoice
+{
+ COPY_LOG_VERBOSITY_DEFAULT = 0, /* logs no additional messages, default */
+ COPY_LOG_VERBOSITY_VERBOSE, /* logs additional messages */
+} CopyLogVerbosityChoice;
+
/*
* A struct to hold COPY options, in a parsed form. All of these are related
* to formatting, except for 'freeze', which doesn't really belong here, but
@@ -73,6 +82,7 @@ typedef struct CopyFormatOptions
bool *force_null_flags; /* per-column CSV FN flags */
bool convert_selectively; /* do selective binary conversion? */
CopyOnErrorChoice on_error; /* what to do when error happened */
+ CopyLogVerbosityChoice log_verbosity; /* verbosity of logged messages */
List *convert_select; /* list of column names (can be NIL) */
} CopyFormatOptions;
diff --git a/src/backend/commands/copy.c b/src/backend/commands/copy.c
index 056b6733c8..23eb8c9c79 100644
--- a/src/backend/commands/copy.c
+++ b/src/backend/commands/copy.c
@@ -428,6 +428,36 @@ defGetCopyOnErrorChoice(DefElem *def, ParseState *pstate, bool is_from)
return COPY_ON_ERROR_STOP; /* keep compiler quiet */
}
+/*
+ * Extract a CopyLogVerbosityChoice value from a DefElem.
+ */
+static CopyLogVerbosityChoice
+defGetCopyLogVerbosityChoice(DefElem *def, ParseState *pstate)
+{
+ char *sval;
+
+ /*
+ * If no parameter value given, assume the default value.
+ */
+ if (def->arg == NULL)
+ return COPY_LOG_VERBOSITY_DEFAULT;
+
+ /*
+ * Allow "default", or "verbose" values.
+ */
+ sval = defGetString(def);
+ if (pg_strcasecmp(sval, "default") == 0)
+ return COPY_LOG_VERBOSITY_DEFAULT;
+ if (pg_strcasecmp(sval, "verbose") == 0)
+ return COPY_LOG_VERBOSITY_VERBOSE;
+
+ ereport(ERROR,
+ (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
+ errmsg("COPY LOG_VERBOSITY \"%s\" not recognized", sval),
+ parser_errposition(pstate, def->location)));
+ return COPY_LOG_VERBOSITY_DEFAULT; /* keep compiler quiet */
+}
+
/*
* Process the statement option list for COPY.
*
@@ -454,6 +484,7 @@ ProcessCopyOptions(ParseState *pstate,
bool freeze_specified = false;
bool header_specified = false;
bool on_error_specified = false;
+ bool log_verbosity_specified = false;
ListCell *option;
/* Support external use for option sanity checking */
@@ -613,6 +644,13 @@ ProcessCopyOptions(ParseState *pstate,
on_error_specified = true;
opts_out->on_error = defGetCopyOnErrorChoice(defel, pstate, is_from);
}
+ else if (strcmp(defel->defname, "log_verbosity") == 0)
+ {
+ if (log_verbosity_specified)
+ errorConflictingDefElem(defel, pstate);
+ log_verbosity_specified = true;
+ opts_out->log_verbosity = defGetCopyLogVerbosityChoice(defel, pstate);
+ }
else
ereport(ERROR,
(errcode(ERRCODE_SYNTAX_ERROR),
diff --git a/src/bin/psql/tab-complete.c b/src/bin/psql/tab-complete.c
index 73133ce735..9305800340 100644
--- a/src/bin/psql/tab-complete.c
+++ b/src/bin/psql/tab-complete.c
@@ -2901,7 +2901,7 @@ psql_completion(const char *text, int start, int end)
COMPLETE_WITH("FORMAT", "FREEZE", "DELIMITER", "NULL",
"HEADER", "QUOTE", "ESCAPE", "FORCE_QUOTE",
"FORCE_NOT_NULL", "FORCE_NULL", "ENCODING", "DEFAULT",
- "ON_ERROR");
+ "ON_ERROR", "LOG_VERBOSITY");
/* Complete COPY <sth> FROM|TO filename WITH (FORMAT */
else if (Matches("COPY|\\copy", MatchAny, "FROM|TO", MatchAny, "WITH", "(", "FORMAT"))
@@ -2911,6 +2911,10 @@ psql_completion(const char *text, int start, int end)
else if (Matches("COPY|\\copy", MatchAny, "FROM|TO", MatchAny, "WITH", "(", "ON_ERROR"))
COMPLETE_WITH("stop", "ignore");
+ /* Complete COPY <sth> FROM filename WITH (LOG_VERBOSITY */
+ else if (Matches("COPY|\\copy", MatchAny, "FROM|TO", MatchAny, "WITH", "(", "LOG_VERBOSITY"))
+ COMPLETE_WITH("default", "verbose");
+
/* Complete COPY <sth> FROM <sth> WITH (<options>) */
else if (Matches("COPY|\\copy", MatchAny, "FROM", MatchAny, "WITH", MatchAny))
COMPLETE_WITH("WHERE");
diff --git a/src/test/regress/expected/copy2.out b/src/test/regress/expected/copy2.out
index 25c401ce34..62406ef827 100644
--- a/src/test/regress/expected/copy2.out
+++ b/src/test/regress/expected/copy2.out
@@ -81,6 +81,10 @@ COPY x from stdin (on_error ignore, on_error ignore);
ERROR: conflicting or redundant options
LINE 1: COPY x from stdin (on_error ignore, on_error ignore);
^
+COPY x from stdin (log_verbosity 'default', log_verbosity 'verbose');
+ERROR: conflicting or redundant options
+LINE 1: COPY x from stdin (log_verbosity 'default', log_verbosity 'v...
+ ^
-- incorrect options
COPY x to stdin (format BINARY, delimiter ',');
ERROR: cannot specify DELIMITER in BINARY mode
@@ -108,6 +112,10 @@ COPY x to stdin (format BINARY, on_error unsupported);
ERROR: COPY ON_ERROR cannot be used with COPY TO
LINE 1: COPY x to stdin (format BINARY, on_error unsupported);
^
+COPY x to stdout (log_verbosity 'unsupported');
+ERROR: COPY LOG_VERBOSITY "unsupported" not recognized
+LINE 1: COPY x to stdout (log_verbosity 'unsupported');
+ ^
-- too many columns in column list: should fail
COPY x (a, b, c, d, e, d, c) from stdin;
ERROR: column "d" specified more than once
diff --git a/src/test/regress/sql/copy2.sql b/src/test/regress/sql/copy2.sql
index b5e549e856..5116157cc9 100644
--- a/src/test/regress/sql/copy2.sql
+++ b/src/test/regress/sql/copy2.sql
@@ -67,6 +67,7 @@ COPY x from stdin (force_null (a), force_null (b));
COPY x from stdin (convert_selectively (a), convert_selectively (b));
COPY x from stdin (encoding 'sql_ascii', encoding 'sql_ascii');
COPY x from stdin (on_error ignore, on_error ignore);
+COPY x from stdin (log_verbosity 'default', log_verbosity 'verbose');
-- incorrect options
COPY x to stdin (format BINARY, delimiter ',');
@@ -80,6 +81,7 @@ COPY x to stdin (format CSV, force_not_null(a));
COPY x to stdout (format TEXT, force_null(a));
COPY x to stdin (format CSV, force_null(a));
COPY x to stdin (format BINARY, on_error unsupported);
+COPY x to stdout (log_verbosity 'unsupported');
-- too many columns in column list: should fail
COPY x (a, b, c, d, e, d, c) from stdin;
diff --git a/doc/src/sgml/ref/copy.sgml b/doc/src/sgml/ref/copy.sgml
index 55764fc1f2..67ba6212fe 100644
--- a/doc/src/sgml/ref/copy.sgml
+++ b/doc/src/sgml/ref/copy.sgml
@@ -45,6 +45,7 @@ COPY { <replaceable class="parameter">table_name</replaceable> [ ( <replaceable
FORCE_NULL { ( <replaceable class="parameter">column_name</replaceable> [, ...] ) | * }
ON_ERROR '<replaceable class="parameter">error_action</replaceable>'
ENCODING '<replaceable class="parameter">encoding_name</replaceable>'
+ LOG_VERBOSITY [ <replaceable class="parameter">mode</replaceable> ]
</synopsis>
</refsynopsisdiv>
@@ -415,6 +416,19 @@ COPY { <replaceable class="parameter">table_name</replaceable> [ ( <replaceable
</listitem>
</varlistentry>
+ <varlistentry>
+ <term><literal>LOG_VERBOSITY</literal></term>
+ <listitem>
+ <para>
+ Sets the verbosity of logged messages by <command>COPY</command> command.
+ A <replaceable class="parameter">mode</replaceable> value of
+ <literal>verbose</literal> can be used to emit more informative messages
+ by the command, while value of <literal>default</literal> (which is the
+ default) can be used to not log any additional messages.
+ </para>
+ </listitem>
+ </varlistentry>
+
<varlistentry>
<term><literal>WHERE</literal></term>
<listitem>
diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list
index aa7a25b8f8..549378c8ad 100644
--- a/src/tools/pgindent/typedefs.list
+++ b/src/tools/pgindent/typedefs.list
@@ -479,6 +479,7 @@ CopyFromState
CopyFromStateData
CopyHeaderChoice
CopyInsertMethod
+CopyLogVerbosityChoice
CopyMultiInsertBuffer
CopyMultiInsertInfo
CopyOnErrorChoice
--
2.43.0
[text/x-diff] v6-0002-Add-detailed-info-when-COPY-skips-soft-errors.patch (4.9K, ../../[email protected]/3-v6-0002-Add-detailed-info-when-COPY-skips-soft-errors.patch)
download | inline diff:
From 1f087fb0cd740e927c6a475996430e2bb12de845 Mon Sep 17 00:00:00 2001
From: Bharath Rupireddy <[email protected]>
Date: Mon, 11 Mar 2024 11:53:44 +0000
Subject: [PATCH v6 2/3] Add detailed info when COPY skips soft errors
This commit emits individual info like line number and column name
when COPY skips soft errors. Because, the summary containing the
total rows skipped isn't enough for the users to know what exactly
are the malformed rows in the input data.
Author: Bharath Rupireddy
Reviewed-by: Michael Paquier, Masahiko Sawada
Reviewed-by: Atsushi Torikoshi
Discussion: https://www.postgresql.org/message-id/CALj2ACUk700cYhx1ATRQyRw-fBM%2BaRo6auRAitKGff7XNmYfqQ%40mail.gmail.com
---
src/backend/commands/copyfromparse.c | 10 ++++++++++
src/test/regress/expected/copy2.out | 7 ++++++-
src/test/regress/sql/copy2.sql | 4 +++-
doc/src/sgml/ref/copy.sgml | 12 +++++++++---
4 files changed, 28 insertions(+), 5 deletions(-)
diff --git a/src/backend/commands/copyfromparse.c b/src/backend/commands/copyfromparse.c
index 5682d5d054..a7ad6c17c8 100644
--- a/src/backend/commands/copyfromparse.c
+++ b/src/backend/commands/copyfromparse.c
@@ -967,7 +967,17 @@ NextCopyFrom(CopyFromState cstate, ExprContext *econtext,
(Node *) cstate->escontext,
&values[m]))
{
+ Assert(cstate->opts.on_error != COPY_ON_ERROR_STOP);
+
cstate->num_errors++;
+
+ if (cstate->opts.log_verbosity == COPY_LOG_VERBOSITY_VERBOSE)
+ ereport(NOTICE,
+ errmsg("detected data type incompatibility at line number %llu for column %s; COPY %s",
+ (unsigned long long) cstate->cur_lineno,
+ cstate->cur_attname,
+ cstate->cur_relname));
+
return true;
}
diff --git a/src/test/regress/expected/copy2.out b/src/test/regress/expected/copy2.out
index 62406ef827..c6655000e4 100644
--- a/src/test/regress/expected/copy2.out
+++ b/src/test/regress/expected/copy2.out
@@ -737,7 +737,12 @@ CREATE TABLE check_ign_err (n int, m int[], k int);
COPY check_ign_err FROM STDIN WITH (on_error stop);
ERROR: invalid input syntax for type integer: "a"
CONTEXT: COPY check_ign_err, line 2, column n: "a"
-COPY check_ign_err FROM STDIN WITH (on_error ignore);
+-- tests for options on_error and log_verbosity
+COPY check_ign_err FROM STDIN WITH (on_error ignore, log_verbosity 'verbose');
+NOTICE: detected data type incompatibility at line number 2 for column n; COPY check_ign_err
+NOTICE: detected data type incompatibility at line number 3 for column k; COPY check_ign_err
+NOTICE: detected data type incompatibility at line number 4 for column m; COPY check_ign_err
+NOTICE: detected data type incompatibility at line number 5 for column n; COPY check_ign_err
NOTICE: 4 rows were skipped due to data type incompatibility
SELECT * FROM check_ign_err;
n | m | k
diff --git a/src/test/regress/sql/copy2.sql b/src/test/regress/sql/copy2.sql
index 5116157cc9..b637a5b3bb 100644
--- a/src/test/regress/sql/copy2.sql
+++ b/src/test/regress/sql/copy2.sql
@@ -510,7 +510,9 @@ a {2} 2
5 {5} 5
\.
-COPY check_ign_err FROM STDIN WITH (on_error ignore);
+
+-- tests for options on_error and log_verbosity
+COPY check_ign_err FROM STDIN WITH (on_error ignore, log_verbosity 'verbose');
1 {1} 1
a {2} 2
3 {3} 3333333333
diff --git a/doc/src/sgml/ref/copy.sgml b/doc/src/sgml/ref/copy.sgml
index 67ba6212fe..ed7fdc59fb 100644
--- a/doc/src/sgml/ref/copy.sgml
+++ b/doc/src/sgml/ref/copy.sgml
@@ -398,8 +398,12 @@ COPY { <replaceable class="parameter">table_name</replaceable> [ ( <replaceable
when the <literal>FORMAT</literal> is <literal>text</literal> or <literal>csv</literal>.
</para>
<para>
- A <literal>NOTICE</literal> message containing the ignored row count is emitted at the end
- of the <command>COPY FROM</command> if at least one row was discarded.
+ A <literal>NOTICE</literal> message containing the ignored row count is
+ emitted at the end of the <command>COPY FROM</command> if at least one
+ row was discarded. When <literal>LOG_VERBOSITY</literal> option is set to
+ <literal>verbose</literal>, a <literal>NOTICE</literal> message
+ containing the line number and column name (whose input conversion has
+ failed) is emitted for each discarded row.
</para>
</listitem>
</varlistentry>
@@ -424,7 +428,9 @@ COPY { <replaceable class="parameter">table_name</replaceable> [ ( <replaceable
A <replaceable class="parameter">mode</replaceable> value of
<literal>verbose</literal> can be used to emit more informative messages
by the command, while value of <literal>default</literal> (which is the
- default) can be used to not log any additional messages.
+ default) can be used to not log any additional messages. As an example,
+ see its usage for <command>COPY FROM</command> command when
+ <literal>ON_ERROR</literal> option is set to <literal>ignore</literal>.
</para>
</listitem>
</varlistentry>
--
2.43.0
[text/x-diff] v6-0003-Some-docs-and-comments-updates.patch (5.0K, ../../[email protected]/4-v6-0003-Some-docs-and-comments-updates.patch)
download | inline diff:
From 96368b5af3c4186a065770cec96fcb0ea60dafd0 Mon Sep 17 00:00:00 2001
From: Michael Paquier <[email protected]>
Date: Wed, 13 Mar 2024 08:45:20 +0900
Subject: [PATCH v6 3/3] Some docs and comments updates
---
src/backend/commands/copyfromparse.c | 6 ++----
src/test/regress/expected/copy2.out | 15 +++++++++------
src/test/regress/sql/copy2.sql | 3 +++
doc/src/sgml/ref/copy.sgml | 18 ++++++++++--------
4 files changed, 24 insertions(+), 18 deletions(-)
diff --git a/src/backend/commands/copyfromparse.c b/src/backend/commands/copyfromparse.c
index a7ad6c17c8..44f615de28 100644
--- a/src/backend/commands/copyfromparse.c
+++ b/src/backend/commands/copyfromparse.c
@@ -973,11 +973,9 @@ NextCopyFrom(CopyFromState cstate, ExprContext *econtext,
if (cstate->opts.log_verbosity == COPY_LOG_VERBOSITY_VERBOSE)
ereport(NOTICE,
- errmsg("detected data type incompatibility at line number %llu for column %s; COPY %s",
+ errmsg("data type incompatibility at line %llu for column \"%s\"",
(unsigned long long) cstate->cur_lineno,
- cstate->cur_attname,
- cstate->cur_relname));
-
+ cstate->cur_attname));
return true;
}
diff --git a/src/test/regress/expected/copy2.out b/src/test/regress/expected/copy2.out
index c6655000e4..af669fedbe 100644
--- a/src/test/regress/expected/copy2.out
+++ b/src/test/regress/expected/copy2.out
@@ -739,17 +739,20 @@ ERROR: invalid input syntax for type integer: "a"
CONTEXT: COPY check_ign_err, line 2, column n: "a"
-- tests for options on_error and log_verbosity
COPY check_ign_err FROM STDIN WITH (on_error ignore, log_verbosity 'verbose');
-NOTICE: detected data type incompatibility at line number 2 for column n; COPY check_ign_err
-NOTICE: detected data type incompatibility at line number 3 for column k; COPY check_ign_err
-NOTICE: detected data type incompatibility at line number 4 for column m; COPY check_ign_err
-NOTICE: detected data type incompatibility at line number 5 for column n; COPY check_ign_err
-NOTICE: 4 rows were skipped due to data type incompatibility
+NOTICE: data type incompatibility at line 2 for column "n"
+NOTICE: data type incompatibility at line 3 for column "k"
+NOTICE: data type incompatibility at line 4 for column "m"
+NOTICE: data type incompatibility at line 5 for column "n"
+NOTICE: data type incompatibility at line 7 for column "m"
+NOTICE: data type incompatibility at line 8 for column "k"
+NOTICE: 6 rows were skipped due to data type incompatibility
SELECT * FROM check_ign_err;
n | m | k
---+-----+---
1 | {1} | 1
5 | {5} | 5
-(2 rows)
+ 8 | {8} | 8
+(3 rows)
-- test datatype error that can't be handled as soft: should fail
CREATE TABLE hard_err(foo widget);
diff --git a/src/test/regress/sql/copy2.sql b/src/test/regress/sql/copy2.sql
index b637a5b3bb..4fb736535d 100644
--- a/src/test/regress/sql/copy2.sql
+++ b/src/test/regress/sql/copy2.sql
@@ -519,6 +519,9 @@ a {2} 2
4 {a, 4} 4
5 {5} 5
+6 a
+7 {7} a
+8 {8} 8
\.
SELECT * FROM check_ign_err;
diff --git a/doc/src/sgml/ref/copy.sgml b/doc/src/sgml/ref/copy.sgml
index ed7fdc59fb..7d594d275e 100644
--- a/doc/src/sgml/ref/copy.sgml
+++ b/doc/src/sgml/ref/copy.sgml
@@ -402,8 +402,8 @@ COPY { <replaceable class="parameter">table_name</replaceable> [ ( <replaceable
emitted at the end of the <command>COPY FROM</command> if at least one
row was discarded. When <literal>LOG_VERBOSITY</literal> option is set to
<literal>verbose</literal>, a <literal>NOTICE</literal> message
- containing the line number and column name (whose input conversion has
- failed) is emitted for each discarded row.
+ containing the line of the input file and the column name whose input
+ conversion has failed is emitted for each discarded row.
</para>
</listitem>
</varlistentry>
@@ -424,13 +424,15 @@ COPY { <replaceable class="parameter">table_name</replaceable> [ ( <replaceable
<term><literal>LOG_VERBOSITY</literal></term>
<listitem>
<para>
- Sets the verbosity of logged messages by <command>COPY</command> command.
+ Sets the verbosity of some of the messages logged by a
+ <command>COPY</command> command.
A <replaceable class="parameter">mode</replaceable> value of
- <literal>verbose</literal> can be used to emit more informative messages
- by the command, while value of <literal>default</literal> (which is the
- default) can be used to not log any additional messages. As an example,
- see its usage for <command>COPY FROM</command> command when
- <literal>ON_ERROR</literal> option is set to <literal>ignore</literal>.
+ <literal>verbose</literal> can be used to emit more informative messages.
+ <literal>default</literal> will not log any additional messages.
+ </para>
+ <para>
+ This is currently used in <command>COPY FROM</command> command when
+ <literal>ON_ERROR</literal> is set to <literal>ignore</literal>.
</para>
</listitem>
</varlistentry>
--
2.43.0
[application/pgp-signature] signature.asc (833B, ../../[email protected]/5-signature.asc)
download
^ permalink raw reply [nested|flat] 282+ messages in thread
* Re: Add new error_action COPY ON_ERROR "log"
2024-03-11 05:46 Re: Add new error_action COPY ON_ERROR "log" Michael Paquier <[email protected]>
2024-03-11 12:30 ` Re: Add new error_action COPY ON_ERROR "log" Bharath Rupireddy <[email protected]>
2024-03-12 06:52 ` Re: Add new error_action COPY ON_ERROR "log" Michael Paquier <[email protected]>
2024-03-12 07:24 ` Re: Add new error_action COPY ON_ERROR "log" Bharath Rupireddy <[email protected]>
2024-03-12 23:46 ` Re: Add new error_action COPY ON_ERROR "log" Michael Paquier <[email protected]>
@ 2024-03-13 02:38 ` Bharath Rupireddy <[email protected]>
2024-03-13 05:39 ` Re: Add new error_action COPY ON_ERROR "log" Michael Paquier <[email protected]>
0 siblings, 1 reply; 282+ messages in thread
From: Bharath Rupireddy @ 2024-03-13 02:38 UTC (permalink / raw)
To: Michael Paquier <[email protected]>; +Cc: torikoshia <[email protected]>; David G. Johnston <[email protected]>; [email protected]; pgsql-hackers; Alexander Korotkov <[email protected]>
On Wed, Mar 13, 2024 at 5:16 AM Michael Paquier <[email protected]> wrote:
>
> On Tue, Mar 12, 2024 at 12:54:29PM +0530, Bharath Rupireddy wrote:
> > +1. But, do you want to add them now or later as a separate
> > patch/discussion altogether?
>
> The attached 0003 is what I had in mind:
> - Simplification of the LOG generated with quotes applied around the
> column name, don't see much need to add the relation name, either, for
> consistency and because the knowledge is known in the query.
> - A few more tests.
> - Some doc changes.
LGMT. So, I've merged those changes into 0001 and 0002.
> >> Wouldn't it be better to squash the patches together, by the way?
> >
> > I guess not. They are two different things IMV.
>
> Well, 0001 is sitting doing nothing because the COPY code does not
> make use of it internally.
Yes. That's why I left a note in the commit message that a future
commit will use it.
--
Bharath Rupireddy
PostgreSQL Contributors Team
RDS Open Source Databases
Amazon Web Services: https://aws.amazon.com
Attachments:
[application/x-patch] v7-0001-Add-LOG_VERBOSITY-option-to-COPY-command.patch (9.5K, ../../CALj2ACW=QJcmqP4VWNAzBmgaG5f5bSjU_hAGzS2NfwRJdy2cJw@mail.gmail.com/2-v7-0001-Add-LOG_VERBOSITY-option-to-COPY-command.patch)
download | inline diff:
From 77b9f28121d6531f40d96f7c00ecdb860550b67f Mon Sep 17 00:00:00 2001
From: Bharath Rupireddy <[email protected]>
Date: Wed, 13 Mar 2024 02:20:42 +0000
Subject: [PATCH v7 1/2] Add LOG_VERBOSITY option to COPY command
This commit adds a new option LOG_VERBOSITY to set the verbosity of
logged messages by COPY command. A value of 'verbose' can be used
to emit more informative messages by the command, while the value
of 'default (which is the default) can be used to not log any
additional messages. More values such as 'terse', 'row_details'
etc. can be added based on the need to the LOG_VERBOSITY option.
An upcoming commit for emitting more info on soft errors by
COPY FROM command with ON_ERROR 'ignore' uses this.
Author: Bharath Rupireddy
Reviewed-by: Michael Paquier, Masahiko Sawada
Reviewed-by: Atsushi Torikoshi
Discussion: https://www.postgresql.org/message-id/CALj2ACXNA0focNeriYRvQQaCGc4CsTuOnFbzF9LqTKNWxuJdhA%40mail.gmail.com
---
doc/src/sgml/ref/copy.sgml | 14 +++++++++++
src/backend/commands/copy.c | 38 +++++++++++++++++++++++++++++
src/bin/psql/tab-complete.c | 6 ++++-
src/include/commands/copy.h | 10 ++++++++
src/test/regress/expected/copy2.out | 8 ++++++
src/test/regress/sql/copy2.sql | 2 ++
src/tools/pgindent/typedefs.list | 1 +
7 files changed, 78 insertions(+), 1 deletion(-)
diff --git a/doc/src/sgml/ref/copy.sgml b/doc/src/sgml/ref/copy.sgml
index 55764fc1f2..eba9b8f64e 100644
--- a/doc/src/sgml/ref/copy.sgml
+++ b/doc/src/sgml/ref/copy.sgml
@@ -45,6 +45,7 @@ COPY { <replaceable class="parameter">table_name</replaceable> [ ( <replaceable
FORCE_NULL { ( <replaceable class="parameter">column_name</replaceable> [, ...] ) | * }
ON_ERROR '<replaceable class="parameter">error_action</replaceable>'
ENCODING '<replaceable class="parameter">encoding_name</replaceable>'
+ LOG_VERBOSITY [ <replaceable class="parameter">mode</replaceable> ]
</synopsis>
</refsynopsisdiv>
@@ -415,6 +416,19 @@ COPY { <replaceable class="parameter">table_name</replaceable> [ ( <replaceable
</listitem>
</varlistentry>
+ <varlistentry>
+ <term><literal>LOG_VERBOSITY</literal></term>
+ <listitem>
+ <para>
+ Sets the verbosity of some of the messages logged by a
+ <command>COPY</command> command.
+ A <replaceable class="parameter">mode</replaceable> value of
+ <literal>verbose</literal> can be used to emit more informative messages.
+ <literal>default</literal> will not log any additional messages.
+ </para>
+ </listitem>
+ </varlistentry>
+
<varlistentry>
<term><literal>WHERE</literal></term>
<listitem>
diff --git a/src/backend/commands/copy.c b/src/backend/commands/copy.c
index 056b6733c8..23eb8c9c79 100644
--- a/src/backend/commands/copy.c
+++ b/src/backend/commands/copy.c
@@ -428,6 +428,36 @@ defGetCopyOnErrorChoice(DefElem *def, ParseState *pstate, bool is_from)
return COPY_ON_ERROR_STOP; /* keep compiler quiet */
}
+/*
+ * Extract a CopyLogVerbosityChoice value from a DefElem.
+ */
+static CopyLogVerbosityChoice
+defGetCopyLogVerbosityChoice(DefElem *def, ParseState *pstate)
+{
+ char *sval;
+
+ /*
+ * If no parameter value given, assume the default value.
+ */
+ if (def->arg == NULL)
+ return COPY_LOG_VERBOSITY_DEFAULT;
+
+ /*
+ * Allow "default", or "verbose" values.
+ */
+ sval = defGetString(def);
+ if (pg_strcasecmp(sval, "default") == 0)
+ return COPY_LOG_VERBOSITY_DEFAULT;
+ if (pg_strcasecmp(sval, "verbose") == 0)
+ return COPY_LOG_VERBOSITY_VERBOSE;
+
+ ereport(ERROR,
+ (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
+ errmsg("COPY LOG_VERBOSITY \"%s\" not recognized", sval),
+ parser_errposition(pstate, def->location)));
+ return COPY_LOG_VERBOSITY_DEFAULT; /* keep compiler quiet */
+}
+
/*
* Process the statement option list for COPY.
*
@@ -454,6 +484,7 @@ ProcessCopyOptions(ParseState *pstate,
bool freeze_specified = false;
bool header_specified = false;
bool on_error_specified = false;
+ bool log_verbosity_specified = false;
ListCell *option;
/* Support external use for option sanity checking */
@@ -613,6 +644,13 @@ ProcessCopyOptions(ParseState *pstate,
on_error_specified = true;
opts_out->on_error = defGetCopyOnErrorChoice(defel, pstate, is_from);
}
+ else if (strcmp(defel->defname, "log_verbosity") == 0)
+ {
+ if (log_verbosity_specified)
+ errorConflictingDefElem(defel, pstate);
+ log_verbosity_specified = true;
+ opts_out->log_verbosity = defGetCopyLogVerbosityChoice(defel, pstate);
+ }
else
ereport(ERROR,
(errcode(ERRCODE_SYNTAX_ERROR),
diff --git a/src/bin/psql/tab-complete.c b/src/bin/psql/tab-complete.c
index 73133ce735..9305800340 100644
--- a/src/bin/psql/tab-complete.c
+++ b/src/bin/psql/tab-complete.c
@@ -2901,7 +2901,7 @@ psql_completion(const char *text, int start, int end)
COMPLETE_WITH("FORMAT", "FREEZE", "DELIMITER", "NULL",
"HEADER", "QUOTE", "ESCAPE", "FORCE_QUOTE",
"FORCE_NOT_NULL", "FORCE_NULL", "ENCODING", "DEFAULT",
- "ON_ERROR");
+ "ON_ERROR", "LOG_VERBOSITY");
/* Complete COPY <sth> FROM|TO filename WITH (FORMAT */
else if (Matches("COPY|\\copy", MatchAny, "FROM|TO", MatchAny, "WITH", "(", "FORMAT"))
@@ -2911,6 +2911,10 @@ psql_completion(const char *text, int start, int end)
else if (Matches("COPY|\\copy", MatchAny, "FROM|TO", MatchAny, "WITH", "(", "ON_ERROR"))
COMPLETE_WITH("stop", "ignore");
+ /* Complete COPY <sth> FROM filename WITH (LOG_VERBOSITY */
+ else if (Matches("COPY|\\copy", MatchAny, "FROM|TO", MatchAny, "WITH", "(", "LOG_VERBOSITY"))
+ COMPLETE_WITH("default", "verbose");
+
/* Complete COPY <sth> FROM <sth> WITH (<options>) */
else if (Matches("COPY|\\copy", MatchAny, "FROM", MatchAny, "WITH", MatchAny))
COMPLETE_WITH("WHERE");
diff --git a/src/include/commands/copy.h b/src/include/commands/copy.h
index b3da3cb0be..99d183fa4d 100644
--- a/src/include/commands/copy.h
+++ b/src/include/commands/copy.h
@@ -40,6 +40,15 @@ typedef enum CopyOnErrorChoice
COPY_ON_ERROR_IGNORE, /* ignore errors */
} CopyOnErrorChoice;
+/*
+ * Represents verbosity of logged messages by COPY command.
+ */
+typedef enum CopyLogVerbosityChoice
+{
+ COPY_LOG_VERBOSITY_DEFAULT = 0, /* logs no additional messages, default */
+ COPY_LOG_VERBOSITY_VERBOSE, /* logs additional messages */
+} CopyLogVerbosityChoice;
+
/*
* A struct to hold COPY options, in a parsed form. All of these are related
* to formatting, except for 'freeze', which doesn't really belong here, but
@@ -73,6 +82,7 @@ typedef struct CopyFormatOptions
bool *force_null_flags; /* per-column CSV FN flags */
bool convert_selectively; /* do selective binary conversion? */
CopyOnErrorChoice on_error; /* what to do when error happened */
+ CopyLogVerbosityChoice log_verbosity; /* verbosity of logged messages */
List *convert_select; /* list of column names (can be NIL) */
} CopyFormatOptions;
diff --git a/src/test/regress/expected/copy2.out b/src/test/regress/expected/copy2.out
index 25c401ce34..62406ef827 100644
--- a/src/test/regress/expected/copy2.out
+++ b/src/test/regress/expected/copy2.out
@@ -81,6 +81,10 @@ COPY x from stdin (on_error ignore, on_error ignore);
ERROR: conflicting or redundant options
LINE 1: COPY x from stdin (on_error ignore, on_error ignore);
^
+COPY x from stdin (log_verbosity 'default', log_verbosity 'verbose');
+ERROR: conflicting or redundant options
+LINE 1: COPY x from stdin (log_verbosity 'default', log_verbosity 'v...
+ ^
-- incorrect options
COPY x to stdin (format BINARY, delimiter ',');
ERROR: cannot specify DELIMITER in BINARY mode
@@ -108,6 +112,10 @@ COPY x to stdin (format BINARY, on_error unsupported);
ERROR: COPY ON_ERROR cannot be used with COPY TO
LINE 1: COPY x to stdin (format BINARY, on_error unsupported);
^
+COPY x to stdout (log_verbosity 'unsupported');
+ERROR: COPY LOG_VERBOSITY "unsupported" not recognized
+LINE 1: COPY x to stdout (log_verbosity 'unsupported');
+ ^
-- too many columns in column list: should fail
COPY x (a, b, c, d, e, d, c) from stdin;
ERROR: column "d" specified more than once
diff --git a/src/test/regress/sql/copy2.sql b/src/test/regress/sql/copy2.sql
index b5e549e856..5116157cc9 100644
--- a/src/test/regress/sql/copy2.sql
+++ b/src/test/regress/sql/copy2.sql
@@ -67,6 +67,7 @@ COPY x from stdin (force_null (a), force_null (b));
COPY x from stdin (convert_selectively (a), convert_selectively (b));
COPY x from stdin (encoding 'sql_ascii', encoding 'sql_ascii');
COPY x from stdin (on_error ignore, on_error ignore);
+COPY x from stdin (log_verbosity 'default', log_verbosity 'verbose');
-- incorrect options
COPY x to stdin (format BINARY, delimiter ',');
@@ -80,6 +81,7 @@ COPY x to stdin (format CSV, force_not_null(a));
COPY x to stdout (format TEXT, force_null(a));
COPY x to stdin (format CSV, force_null(a));
COPY x to stdin (format BINARY, on_error unsupported);
+COPY x to stdout (log_verbosity 'unsupported');
-- too many columns in column list: should fail
COPY x (a, b, c, d, e, d, c) from stdin;
diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list
index aa7a25b8f8..549378c8ad 100644
--- a/src/tools/pgindent/typedefs.list
+++ b/src/tools/pgindent/typedefs.list
@@ -479,6 +479,7 @@ CopyFromState
CopyFromStateData
CopyHeaderChoice
CopyInsertMethod
+CopyLogVerbosityChoice
CopyMultiInsertBuffer
CopyMultiInsertInfo
CopyOnErrorChoice
--
2.34.1
[application/x-patch] v7-0002-Add-detailed-info-when-COPY-skips-soft-errors.patch (5.0K, ../../CALj2ACW=QJcmqP4VWNAzBmgaG5f5bSjU_hAGzS2NfwRJdy2cJw@mail.gmail.com/3-v7-0002-Add-detailed-info-when-COPY-skips-soft-errors.patch)
download | inline diff:
From badf71c273c4a496bcf14701d973f768f24fa7fd Mon Sep 17 00:00:00 2001
From: Bharath Rupireddy <[email protected]>
Date: Wed, 13 Mar 2024 02:30:46 +0000
Subject: [PATCH v7 2/2] Add detailed info when COPY skips soft errors
This commit emits individual info like line number and column name
when COPY skips soft errors. Because, the summary containing the
total rows skipped isn't enough for the users to know what exactly
are the malformed rows in the input data.
Author: Bharath Rupireddy
Reviewed-by: Michael Paquier, Masahiko Sawada
Reviewed-by: Atsushi Torikoshi
Discussion: https://www.postgresql.org/message-id/CALj2ACUk700cYhx1ATRQyRw-fBM%2BaRo6auRAitKGff7XNmYfqQ%40mail.gmail.com
---
doc/src/sgml/ref/copy.sgml | 12 ++++++++++--
src/backend/commands/copyfromparse.c | 9 +++++++++
src/test/regress/expected/copy2.out | 14 +++++++++++---
src/test/regress/sql/copy2.sql | 7 ++++++-
4 files changed, 36 insertions(+), 6 deletions(-)
diff --git a/doc/src/sgml/ref/copy.sgml b/doc/src/sgml/ref/copy.sgml
index eba9b8f64e..bdd6580721 100644
--- a/doc/src/sgml/ref/copy.sgml
+++ b/doc/src/sgml/ref/copy.sgml
@@ -398,8 +398,12 @@ COPY { <replaceable class="parameter">table_name</replaceable> [ ( <replaceable
when the <literal>FORMAT</literal> is <literal>text</literal> or <literal>csv</literal>.
</para>
<para>
- A <literal>NOTICE</literal> message containing the ignored row count is emitted at the end
- of the <command>COPY FROM</command> if at least one row was discarded.
+ A <literal>NOTICE</literal> message containing the ignored row count is
+ emitted at the end of the <command>COPY FROM</command> if at least one
+ row was discarded. When <literal>LOG_VERBOSITY</literal> option is set to
+ <literal>verbose</literal>, a <literal>NOTICE</literal> message
+ containing the line of the input file and the column name whose input
+ conversion has failed is emitted for each discarded row.
</para>
</listitem>
</varlistentry>
@@ -426,6 +430,10 @@ COPY { <replaceable class="parameter">table_name</replaceable> [ ( <replaceable
<literal>verbose</literal> can be used to emit more informative messages.
<literal>default</literal> will not log any additional messages.
</para>
+ <para>
+ This is currently used in <command>COPY FROM</command> command when
+ <literal>ON_ERROR</literal> is set to <literal>ignore</literal>.
+ </para>
</listitem>
</varlistentry>
diff --git a/src/backend/commands/copyfromparse.c b/src/backend/commands/copyfromparse.c
index 5682d5d054..e4a89eef13 100644
--- a/src/backend/commands/copyfromparse.c
+++ b/src/backend/commands/copyfromparse.c
@@ -967,7 +967,16 @@ NextCopyFrom(CopyFromState cstate, ExprContext *econtext,
(Node *) cstate->escontext,
&values[m]))
{
+ Assert(cstate->opts.on_error != COPY_ON_ERROR_STOP);
+
cstate->num_errors++;
+
+ if (cstate->opts.log_verbosity == COPY_LOG_VERBOSITY_VERBOSE)
+ ereport(NOTICE,
+ errmsg("data type incompatibility at line %llu for column \"%s\"",
+ (unsigned long long) cstate->cur_lineno,
+ cstate->cur_attname));
+
return true;
}
diff --git a/src/test/regress/expected/copy2.out b/src/test/regress/expected/copy2.out
index 62406ef827..af669fedbe 100644
--- a/src/test/regress/expected/copy2.out
+++ b/src/test/regress/expected/copy2.out
@@ -737,14 +737,22 @@ CREATE TABLE check_ign_err (n int, m int[], k int);
COPY check_ign_err FROM STDIN WITH (on_error stop);
ERROR: invalid input syntax for type integer: "a"
CONTEXT: COPY check_ign_err, line 2, column n: "a"
-COPY check_ign_err FROM STDIN WITH (on_error ignore);
-NOTICE: 4 rows were skipped due to data type incompatibility
+-- tests for options on_error and log_verbosity
+COPY check_ign_err FROM STDIN WITH (on_error ignore, log_verbosity 'verbose');
+NOTICE: data type incompatibility at line 2 for column "n"
+NOTICE: data type incompatibility at line 3 for column "k"
+NOTICE: data type incompatibility at line 4 for column "m"
+NOTICE: data type incompatibility at line 5 for column "n"
+NOTICE: data type incompatibility at line 7 for column "m"
+NOTICE: data type incompatibility at line 8 for column "k"
+NOTICE: 6 rows were skipped due to data type incompatibility
SELECT * FROM check_ign_err;
n | m | k
---+-----+---
1 | {1} | 1
5 | {5} | 5
-(2 rows)
+ 8 | {8} | 8
+(3 rows)
-- test datatype error that can't be handled as soft: should fail
CREATE TABLE hard_err(foo widget);
diff --git a/src/test/regress/sql/copy2.sql b/src/test/regress/sql/copy2.sql
index 5116157cc9..4fb736535d 100644
--- a/src/test/regress/sql/copy2.sql
+++ b/src/test/regress/sql/copy2.sql
@@ -510,13 +510,18 @@ a {2} 2
5 {5} 5
\.
-COPY check_ign_err FROM STDIN WITH (on_error ignore);
+
+-- tests for options on_error and log_verbosity
+COPY check_ign_err FROM STDIN WITH (on_error ignore, log_verbosity 'verbose');
1 {1} 1
a {2} 2
3 {3} 3333333333
4 {a, 4} 4
5 {5} 5
+6 a
+7 {7} a
+8 {8} 8
\.
SELECT * FROM check_ign_err;
--
2.34.1
^ permalink raw reply [nested|flat] 282+ messages in thread
* Re: Add new error_action COPY ON_ERROR "log"
2024-03-11 05:46 Re: Add new error_action COPY ON_ERROR "log" Michael Paquier <[email protected]>
2024-03-11 12:30 ` Re: Add new error_action COPY ON_ERROR "log" Bharath Rupireddy <[email protected]>
2024-03-12 06:52 ` Re: Add new error_action COPY ON_ERROR "log" Michael Paquier <[email protected]>
2024-03-12 07:24 ` Re: Add new error_action COPY ON_ERROR "log" Bharath Rupireddy <[email protected]>
2024-03-12 23:46 ` Re: Add new error_action COPY ON_ERROR "log" Michael Paquier <[email protected]>
2024-03-13 02:38 ` Re: Add new error_action COPY ON_ERROR "log" Bharath Rupireddy <[email protected]>
@ 2024-03-13 05:39 ` Michael Paquier <[email protected]>
0 siblings, 0 replies; 282+ messages in thread
From: Michael Paquier @ 2024-03-13 05:39 UTC (permalink / raw)
To: Bharath Rupireddy <[email protected]>; +Cc: torikoshia <[email protected]>; David G. Johnston <[email protected]>; [email protected]; pgsql-hackers; Alexander Korotkov <[email protected]>
On Wed, Mar 13, 2024 at 08:08:42AM +0530, Bharath Rupireddy wrote:
> On Wed, Mar 13, 2024 at 5:16 AM Michael Paquier <[email protected]> wrote:
>> The attached 0003 is what I had in mind:
>> - Simplification of the LOG generated with quotes applied around the
>> column name, don't see much need to add the relation name, either, for
>> consistency and because the knowledge is known in the query.
>> - A few more tests.
>> - Some doc changes.
>
> LGMT. So, I've merged those changes into 0001 and 0002.
I've applied the extra tests for now, as this was really confusing.
Hmm. This NOTICE is really bugging me. It is true that the clients
would get more information, but the information is duplicated on the
server side because the error context provides the same information as
the NOTICE itself:
NOTICE: data type incompatibility at line 1 for column "a"
CONTEXT: COPY aa, line 1, column a: "a"
STATEMENT: copy aa from stdin with (on_error ignore, log_verbosity verbose);
--
Michael
Attachments:
[application/pgp-signature] signature.asc (833B, ../../[email protected]/2-signature.asc)
download
^ permalink raw reply [nested|flat] 282+ messages in thread
end of thread, other threads:[~2024-03-13 05:39 UTC | newest]
Thread overview: 282+ messages (download: mbox mbox.gz follow: Atom feed)
-- links below jump to the message on this page --
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/9] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2020-09-13 10:12 [PATCH 3/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 4/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 4/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 4/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 4/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 4/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 4/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 4/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 4/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 4/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 4/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 4/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 4/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 4/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 4/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 4/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 4/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 4/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 4/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 4/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 4/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 4/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 4/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 4/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 3/6] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 4/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-02 18:57 [PATCH 5/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-21 23:24 [PATCH 4/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2021-03-21 23:24 [PATCH 4/8] Optimize allocations in bringetbitmap Tomas Vondra <[email protected]>
2024-03-11 05:46 Re: Add new error_action COPY ON_ERROR "log" Michael Paquier <[email protected]>
2024-03-11 12:30 ` Re: Add new error_action COPY ON_ERROR "log" Bharath Rupireddy <[email protected]>
2024-03-12 06:52 ` Re: Add new error_action COPY ON_ERROR "log" Michael Paquier <[email protected]>
2024-03-12 07:24 ` Re: Add new error_action COPY ON_ERROR "log" Bharath Rupireddy <[email protected]>
2024-03-12 23:46 ` Re: Add new error_action COPY ON_ERROR "log" Michael Paquier <[email protected]>
2024-03-13 02:38 ` Re: Add new error_action COPY ON_ERROR "log" Bharath Rupireddy <[email protected]>
2024-03-13 05:39 ` Re: Add new error_action COPY ON_ERROR "log" Michael Paquier <[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