agora inbox for [email protected]help / color / mirror / Atom feed
[PATCH v1 2/2] WIP: Fix EEOP_JSON_CONSTRUCTOR and EEOP_JSONEXPR size. 80+ messages / 3 participants [nested] [flat]
* [PATCH v1 2/2] WIP: Fix EEOP_JSON_CONSTRUCTOR and EEOP_JSONEXPR size. @ 2022-06-17 01:33 Andres Freund <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Andres Freund @ 2022-06-17 01:33 UTC (permalink / raw) Author: Reviewed-By: Discussion: https://postgr.es/m/ Backpatch: --- src/include/executor/execExpr.h | 2 ++ 1 file changed, 2 insertions(+) diff --git a/src/include/executor/execExpr.h b/src/include/executor/execExpr.h index 2fba1450c65..5bfa7e9bdd4 100644 --- a/src/include/executor/execExpr.h +++ b/src/include/executor/execExpr.h @@ -22,6 +22,8 @@ struct ExprEvalStep; struct SubscriptingRefState; struct ScalarArrayOpExprHashTable; struct JsonbValue; +struct JsonExprState; +struct JsonConstructorExprState; /* Bits in ExprState->flags (see also execnodes.h for public flag bits): */ /* expression's interpreter has been initialized */ -- 2.35.1.677.gabf474a5dd --rftf6si7y32exfgc-- ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v2 4/7] Row pattern recognition patch (executor). @ 2023-06-26 08:05 Tatsuo Ishii <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Tatsuo Ishii @ 2023-06-26 08:05 UTC (permalink / raw) --- src/backend/executor/nodeWindowAgg.c | 225 +++++++++++++++++++- src/backend/utils/adt/windowfuncs.c | 302 ++++++++++++++++++++++++++- src/include/catalog/pg_proc.dat | 9 + src/include/nodes/execnodes.h | 13 ++ src/include/windowapi.h | 9 + 5 files changed, 548 insertions(+), 10 deletions(-) diff --git a/src/backend/executor/nodeWindowAgg.c b/src/backend/executor/nodeWindowAgg.c index 310ac23e3a..bef2bc62b2 100644 --- a/src/backend/executor/nodeWindowAgg.c +++ b/src/backend/executor/nodeWindowAgg.c @@ -48,6 +48,7 @@ #include "utils/acl.h" #include "utils/builtins.h" #include "utils/datum.h" +#include "utils/fmgroids.h" #include "utils/expandeddatum.h" #include "utils/lsyscache.h" #include "utils/memutils.h" @@ -159,6 +160,14 @@ typedef struct WindowStatePerAggData bool restart; /* need to restart this agg in this cycle? */ } WindowStatePerAggData; +/* + * Map between Var attno in a target list and the parsed attno. + */ +typedef struct AttnoMap { + List *attno; /* att number in target list (list of AttNumber) */ + List *attnosyn; /* parsed att number (list of AttNumber) */ +} AttnoMap; + static void initialize_windowaggregate(WindowAggState *winstate, WindowStatePerFunc perfuncstate, WindowStatePerAgg peraggstate); @@ -195,9 +204,9 @@ static Datum GetAggInitVal(Datum textInitVal, Oid transtype); static bool are_peers(WindowAggState *winstate, TupleTableSlot *slot1, TupleTableSlot *slot2); -static bool window_gettupleslot(WindowObject winobj, int64 pos, - TupleTableSlot *slot); +static void attno_map(Node *node, AttnoMap *map); +static bool attno_map_walker(Node *node, void *context); /* * initialize_windowaggregate @@ -2388,6 +2397,12 @@ ExecInitWindowAgg(WindowAgg *node, EState *estate, int eflags) TupleDesc scanDesc; ListCell *l; + TargetEntry *te; + Expr *expr; + Var *var; + int nargs; + AttnoMap attnomap; + /* check for unsupported flags */ Assert(!(eflags & (EXEC_FLAG_BACKWARD | EXEC_FLAG_MARK))); @@ -2483,6 +2498,16 @@ ExecInitWindowAgg(WindowAgg *node, EState *estate, int eflags) winstate->temp_slot_2 = ExecInitExtraTupleSlot(estate, scanDesc, &TTSOpsMinimalTuple); + winstate->prev_slot = ExecInitExtraTupleSlot(estate, scanDesc, + &TTSOpsMinimalTuple); + + winstate->next_slot = ExecInitExtraTupleSlot(estate, scanDesc, + &TTSOpsMinimalTuple); + + winstate->null_slot = ExecInitExtraTupleSlot(estate, scanDesc, + &TTSOpsMinimalTuple); + winstate->null_slot = ExecStoreAllNullTuple(winstate->null_slot); + /* * create frame head and tail slots only if needed (must create slots in * exactly the same cases that update_frameheadpos and update_frametailpos @@ -2667,6 +2692,71 @@ ExecInitWindowAgg(WindowAgg *node, EState *estate, int eflags) winstate->inRangeAsc = node->inRangeAsc; winstate->inRangeNullsFirst = node->inRangeNullsFirst; + /* Set up SKIP TO type */ + winstate->rpSkipTo = node->rpSkipTo; + /* Set up row pattern recognition PATTERN clause */ + winstate->patternVariableList = node->patternVariable; + winstate->patternRegexpList = node->patternRegexp; + + /* Set up row pattern recognition DEFINE clause */ + + /* + * Collect mapping between varattno and varattnosyn in the targetlist. + * XXX: For now we only check RPR's argument. Eventually we have to + * recurse the targetlist to find out all mappings in Var nodes. + */ + attnomap.attno = NIL; + attnomap.attnosyn = NIL; + + foreach (l, node->plan.targetlist) + { + te = lfirst(l); + if (IsA(te->expr, WindowFunc)) + { + WindowFunc *func = (WindowFunc *)te->expr; + if (func->winfnoid != F_RPR) + continue; + + /* sanity check */ + nargs = list_length(func->args); + if (list_length(func->args) != 1) + elog(ERROR, "RPR must have 1 argument but function %d has %d args", func->winfnoid, nargs); + + expr = (Expr *) lfirst(list_head(func->args)); + if (!IsA(expr, Var)) + elog(ERROR, "RPR's arg is not Var"); + + var = (Var *)expr; + elog(DEBUG1, "resname: %s varattno: %d varattnosyn: %d", + te->resname, var->varattno, var->varattnosyn); + attnomap.attno = lappend_int(attnomap.attno, var->varattno); + attnomap.attnosyn = lappend_int(attnomap.attnosyn, var->varattnosyn); + } + } + + winstate->defineVariableList = NIL; + winstate->defineClauseList = NIL; + if (node->defineClause != NIL) + { + foreach(l, node->defineClause) + { + char *name; + ExprState *exps; + + te = lfirst(l); + name = te->resname; + expr = te->expr; + + elog(DEBUG1, "defineVariable name: %s", name); + winstate->defineVariableList = lappend(winstate->defineVariableList, + makeString(pstrdup(name))); + /* tweak expr so that it referes to outer slot */ + attno_map((Node *)expr, &attnomap); + exps = ExecInitExpr(expr, (PlanState *) winstate); + winstate->defineClauseList = lappend(winstate->defineClauseList, exps); + } + } + winstate->all_first = true; winstate->partition_spooled = false; winstate->more_partitions = false; @@ -2674,6 +2764,76 @@ ExecInitWindowAgg(WindowAgg *node, EState *estate, int eflags) return winstate; } +/* + * Rewrite Var node's varattno to the varattno which is used in the target + * list using AttnoMap. We also rewrite varno so that it sees outer tuple + * (PREV) or inner tuple (NEXT). + */ +static void +attno_map(Node *node, AttnoMap *map) +{ + (void) expression_tree_walker(node, attno_map_walker, (void *) map); +} + +static bool +attno_map_walker(Node *node, void *context) +{ + FuncExpr *func; + int nargs; + Expr *expr; + Var *var; + AttnoMap *attnomap; + ListCell *lc1, *lc2; + + if (node == NULL) + return false; + + attnomap = (AttnoMap *) context; + + if (IsA(node, FuncExpr)) + { + func = (FuncExpr *)node; + + if (func->funcid == F_PREV || func->funcid == F_NEXT) + { + /* sanity check */ + nargs = list_length(func->args); + if (list_length(func->args) != 1) + elog(ERROR, "PREV/NEXT must have 1 argument but function %d has %d args", func->funcid, nargs); + + expr = (Expr *) lfirst(list_head(func->args)); + if (!IsA(expr, Var)) + elog(ERROR, "PREV/NEXT's arg is not Var"); /* XXX: is it possible that arg type is Const? */ + var = (Var *)expr; + + if (func->funcid == F_PREV) + var->varno = OUTER_VAR; + else + var->varno = INNER_VAR; + } + return expression_tree_walker(node, attno_map_walker, (void *) context); + } + else if (IsA(node, Var)) + { + var = (Var *)node; + + elog(DEBUG1, "original varno: %d varattno: %d", var->varno, var->varattno); + + forboth(lc1, attnomap->attno, lc2, attnomap->attnosyn) + { + int attno = lfirst_int(lc1); + int attnosyn = lfirst_int(lc2); + + if (var->varattno == attnosyn) + { + elog(DEBUG1, "loc: %d rewrite varattno from: %d to %d", var->location, attnosyn, attno); + var->varattno = attno; + } + } + } + return expression_tree_walker(node, attno_map_walker, (void *) context); +} + /* ----------------- * ExecEndWindowAgg * ----------------- @@ -2691,6 +2851,8 @@ ExecEndWindowAgg(WindowAggState *node) ExecClearTuple(node->agg_row_slot); ExecClearTuple(node->temp_slot_1); ExecClearTuple(node->temp_slot_2); + ExecClearTuple(node->prev_slot); + ExecClearTuple(node->next_slot); if (node->framehead_slot) ExecClearTuple(node->framehead_slot); if (node->frametail_slot) @@ -2740,6 +2902,8 @@ ExecReScanWindowAgg(WindowAggState *node) ExecClearTuple(node->agg_row_slot); ExecClearTuple(node->temp_slot_1); ExecClearTuple(node->temp_slot_2); + ExecClearTuple(node->prev_slot); + ExecClearTuple(node->next_slot); if (node->framehead_slot) ExecClearTuple(node->framehead_slot); if (node->frametail_slot) @@ -3080,7 +3244,7 @@ are_peers(WindowAggState *winstate, TupleTableSlot *slot1, * * Returns true if successful, false if no such row */ -static bool +bool window_gettupleslot(WindowObject winobj, int64 pos, TupleTableSlot *slot) { WindowAggState *winstate = winobj->winstate; @@ -3420,14 +3584,53 @@ WinGetFuncArgInFrame(WindowObject winobj, int argno, WindowAggState *winstate; ExprContext *econtext; TupleTableSlot *slot; - int64 abs_pos; - int64 mark_pos; Assert(WindowObjectIsValid(winobj)); winstate = winobj->winstate; econtext = winstate->ss.ps.ps_ExprContext; slot = winstate->temp_slot_1; + if (WinGetSlotInFrame(winobj, slot, + relpos, seektype, set_mark, + isnull, isout) == 0) + { + econtext->ecxt_outertuple = slot; + return ExecEvalExpr((ExprState *) list_nth(winobj->argstates, argno), + econtext, isnull); + } + + if (isout) + *isout = true; + *isnull = true; + return (Datum) 0; +} + +/* + * WinGetSlotInFrame + * slot: TupleTableSlot to store the result + * relpos: signed rowcount offset from the seek position + * seektype: WINDOW_SEEK_HEAD or WINDOW_SEEK_TAIL + * set_mark: If the row is found/in frame and set_mark is true, the mark is + * moved to the row as a side-effect. + * isnull: output argument, receives isnull status of result + * isout: output argument, set to indicate whether target row position + * is out of frame (can pass NULL if caller doesn't care about this) + * + * Returns 0 if we successfullt got the slot. false if out of frame. + * (also isout is set) + */ +int +WinGetSlotInFrame(WindowObject winobj, TupleTableSlot *slot, + int relpos, int seektype, bool set_mark, + bool *isnull, bool *isout) +{ + WindowAggState *winstate; + int64 abs_pos; + int64 mark_pos; + + Assert(WindowObjectIsValid(winobj)); + winstate = winobj->winstate; + switch (seektype) { case WINDOW_SEEK_CURRENT: @@ -3583,15 +3786,13 @@ WinGetFuncArgInFrame(WindowObject winobj, int argno, *isout = false; if (set_mark) WinSetMarkPosition(winobj, mark_pos); - econtext->ecxt_outertuple = slot; - return ExecEvalExpr((ExprState *) list_nth(winobj->argstates, argno), - econtext, isnull); + return 0; out_of_frame: if (isout) *isout = true; *isnull = true; - return (Datum) 0; + return -1; } /* @@ -3622,3 +3823,9 @@ WinGetFuncArgCurrent(WindowObject winobj, int argno, bool *isnull) return ExecEvalExpr((ExprState *) list_nth(winobj->argstates, argno), econtext, isnull); } + +WindowAggState * +WinGetAggState(WindowObject winobj) +{ + return winobj->winstate; +} diff --git a/src/backend/utils/adt/windowfuncs.c b/src/backend/utils/adt/windowfuncs.c index b87a624fb2..74ef11ce55 100644 --- a/src/backend/utils/adt/windowfuncs.c +++ b/src/backend/utils/adt/windowfuncs.c @@ -13,6 +13,9 @@ */ #include "postgres.h" +#include "catalog/pg_collation_d.h" +#include "executor/executor.h" +#include "nodes/execnodes.h" #include "nodes/supportnodes.h" #include "utils/builtins.h" #include "windowapi.h" @@ -36,10 +39,21 @@ typedef struct int64 remainder; /* (total rows) % (bucket num) */ } ntile_context; +/* + * rpr process information. + * Used for AFTER MATCH SKIP PAST LAST ROW + */ +typedef struct SkipContext +{ + int64 pos; /* last row absolute position */ +} SkipContext; + static bool rank_up(WindowObject winobj); static Datum leadlag_common(FunctionCallInfo fcinfo, bool forward, bool withoffset, bool withdefault); - +static bool get_slots(WindowObject winobj, WindowAggState *winstate, int current_pos); +static int evaluate_pattern(WindowObject winobj, WindowAggState *winstate, + int relpos, char *vname, char *quantifier, bool *result); /* * utility routine for *_rank functions. @@ -713,3 +727,289 @@ window_nth_value(PG_FUNCTION_ARGS) PG_RETURN_DATUM(result); } + +/* + * rpr + * allow to use "Row pattern recognition: WINDOW clause" (SQL:2016 R020) in + * the target list. + * Usage: SELECT rpr(colname) OVER (..) + * where colname is defined in PATTERN clause. + */ +Datum +window_rpr(PG_FUNCTION_ARGS) +{ +#define MAX_PATTERNS 16 /* max variables in PATTERN clause */ + + WindowObject winobj = PG_WINDOW_OBJECT(); + WindowAggState *winstate = WinGetAggState(winobj); + Datum result; + bool expression_result; + bool isnull; + int relpos; + int64 curr_pos, markpos; + ListCell *lc, *lc1; + SkipContext *context = NULL; + + curr_pos = WinGetCurrentPosition(winobj); + elog(DEBUG1, "rpr is called. row: " INT64_FORMAT, curr_pos); + + if (winstate->rpSkipTo == ST_PAST_LAST_ROW) + { + context = (SkipContext *) WinGetPartitionLocalMemory(winobj, sizeof(SkipContext)); + if (curr_pos < context->pos) + { + elog(DEBUG1, "skip this row: curr_pos: " INT64_FORMAT "context->pos: " INT64_FORMAT, + curr_pos, context->pos); + PG_RETURN_NULL(); + } + } + + /* + * Evaluate PATTERN until one of expressions is not true or out of frame. + */ + relpos = 0; + + forboth(lc, winstate->patternVariableList, lc1, winstate->patternRegexpList) + { + char *vname = strVal(lfirst(lc)); + char *quantifier = strVal(lfirst(lc1)); + + elog(DEBUG1, "relpos: %d pattern vname: %s quantifier: %s", relpos, vname, quantifier); + + /* evaluate row pattern against current row */ + relpos = evaluate_pattern(winobj, winstate, relpos, vname, quantifier, &expression_result); + + /* + * If the expression did not match, we are done. + */ + if (!expression_result) + break; + + /* out of frame? */ + if (relpos < 0) + break; + + /* count up relative row position */ + relpos++; + } + + elog(DEBUG1, "relpos: %d", relpos); + + /* + * If current row satified the pattern, return argument expression. + */ + if (expression_result) + { + result = WinGetFuncArgInFrame(winobj, 0, + 0, WINDOW_SEEK_HEAD, false, + &isnull, NULL); + } + + /* + * At this point we can set mark down to current pos -2. + */ + markpos = curr_pos -2; + elog(DEBUG1, "markpos: " INT64_FORMAT, markpos); + if (markpos >= 0) + WinSetMarkPosition(winobj, markpos); + + if (expression_result) + { + if (winstate->rpSkipTo == ST_PAST_LAST_ROW) + { + context->pos += relpos; + elog(DEBUG1, "context->pos: " INT64_FORMAT, context->pos); + } + PG_RETURN_DATUM(result); + } + + PG_RETURN_NULL(); +} + +/* + * Evaluate expression associated with PATTERN variable vname. + * relpos is relative row position in a frame (starting from 0). + * "quantifier" is the quatifier part of the PATTERN regular expression. + * Currently only '+' is allowed. + * result is out paramater representing the expression evaluation result + * is true of false. + * Return values are: + * >=0: the last match relative row position + * -1: current row is out of frame + */ +static +int evaluate_pattern(WindowObject winobj, WindowAggState *winstate, + int relpos, char *vname, char *quantifier, bool *result) +{ + ExprContext *econtext = winstate->ss.ps.ps_ExprContext; + ListCell *lc1, *lc2; + ExprState *pat; + Datum eval_result; + int sts; + bool out_of_frame = false; + bool isnull; + StringInfo encoded_str = makeStringInfo(); + char pattern_str[128]; + + forboth (lc1, winstate->defineVariableList, lc2, winstate->defineClauseList) + { + char *name = strVal(lfirst(lc1)); + bool second_try_match = false; + + if (strcmp(vname, name)) + continue; + + /* set expression to evaluate */ + pat = lfirst(lc2); + + for (;;) + { + if (!get_slots(winobj, winstate, relpos)) + { + out_of_frame = true; + break; /* current row is out of frame */ + } + + /* evaluate the expression */ + eval_result = ExecEvalExpr(pat, econtext, &isnull); + if (isnull) + { + /* expression is NULL */ + elog(DEBUG1, "expression for %s is NULL at row: %d", vname, relpos); + break; + } + else + { + if (!DatumGetBool(eval_result)) + { + /* expression is false */ + elog(DEBUG1, "expression for %s is false at row: %d", vname, relpos); + break; + } + else + { + /* expression is true */ + elog(DEBUG1, "expression for %s is true at row: %d", vname, relpos); + appendStringInfoChar(encoded_str, vname[0]); + + /* If quantifier is "+", we need to look for more matching row */ + if (quantifier && !strcmp(quantifier, "+")) + { + /* remember that we want to try another row */ + second_try_match = true; + relpos++; + } + else + break; + } + } + } + if (second_try_match) + relpos--; + + if (out_of_frame) + { + *result = false; + return -1; + } + + /* build regular expression */ + snprintf(pattern_str, sizeof(pattern_str), "%c%s", vname[0], quantifier); + + /* + * Do regular expression matching against sequence of rows satisfying + * the expression using regexp_instr(). + */ + sts = DatumGetInt32(DirectFunctionCall2Coll(regexp_instr, DEFAULT_COLLATION_OID, + PointerGetDatum(cstring_to_text(encoded_str->data)), + PointerGetDatum(cstring_to_text(pattern_str)))); + elog(DEBUG1, "regexp_instr returned: %d. str: %s regexp: %s", + sts, encoded_str->data, pattern_str); + *result = (sts > 0)? true : false; + } + return relpos; +} + +/* + * Get current, previous and next tuple. + * Returns true if still within frame. + */ +static bool +get_slots(WindowObject winobj, WindowAggState *winstate, int current_pos) +{ + TupleTableSlot *slot; + bool isnull, isout; + int sts; + ExprContext *econtext; + + econtext = winstate->ss.ps.ps_ExprContext; + + /* for current row */ + slot = winstate->temp_slot_1; + sts = WinGetSlotInFrame(winobj, slot, + current_pos, WINDOW_SEEK_HEAD, false, + &isnull, &isout); + if (sts < 0) + { + elog(DEBUG1, "current row is out of frame"); + econtext->ecxt_scantuple = winstate->null_slot; + return false; + } + else + econtext->ecxt_scantuple = slot; + + /* for PREV */ + if (current_pos > 0) + { + slot = winstate->prev_slot; + sts = WinGetSlotInFrame(winobj, slot, + current_pos - 1, WINDOW_SEEK_HEAD, false, + &isnull, &isout); + if (sts < 0) + { + elog(DEBUG1, "previous row out of frame at: %d", current_pos); + econtext->ecxt_outertuple = winstate->null_slot; + } + econtext->ecxt_outertuple = slot; + } + else + econtext->ecxt_outertuple = winstate->null_slot; + + /* for NEXT */ + slot = winstate->next_slot; + sts = WinGetSlotInFrame(winobj, slot, + current_pos + 1, WINDOW_SEEK_HEAD, false, + &isnull, &isout); + if (sts < 0) + { + elog(DEBUG1, "next row out of frame at: %d", current_pos); + econtext->ecxt_innertuple = winstate->null_slot; + } + else + econtext->ecxt_innertuple = slot; + + return true; +} + +/* + * prev + * Dummy function to invoke RPR's navigation operator "PREV". + * This is *not* a window function. + */ +Datum +window_prev(PG_FUNCTION_ARGS) +{ + PG_RETURN_DATUM(PG_GETARG_DATUM(0)); +} + +/* + * next + * Dummy function to invoke RPR's navigation operation "NEXT". + * This is *not* a window function. + */ +Datum +window_next(PG_FUNCTION_ARGS) +{ + PG_RETURN_DATUM(PG_GETARG_DATUM(0)); +} + diff --git a/src/include/catalog/pg_proc.dat b/src/include/catalog/pg_proc.dat index 6996073989..e3a9e0ffeb 100644 --- a/src/include/catalog/pg_proc.dat +++ b/src/include/catalog/pg_proc.dat @@ -10397,6 +10397,15 @@ { oid => '3114', descr => 'fetch the Nth row value', proname => 'nth_value', prokind => 'w', prorettype => 'anyelement', proargtypes => 'anyelement int4', prosrc => 'window_nth_value' }, +{ oid => '6122', descr => 'row pattern recognition in window', + proname => 'rpr', prokind => 'w', prorettype => 'anyelement', + proargtypes => 'anyelement', prosrc => 'window_rpr' }, +{ oid => '6123', descr => 'previous value', + proname => 'prev', provolatile => 's', prorettype => 'anyelement', + proargtypes => 'anyelement', prosrc => 'window_prev' }, +{ oid => '6124', descr => 'next value', + proname => 'next', provolatile => 's', prorettype => 'anyelement', + proargtypes => 'anyelement', prosrc => 'window_next' }, # functions for range types { oid => '3832', descr => 'I/O', diff --git a/src/include/nodes/execnodes.h b/src/include/nodes/execnodes.h index cb714f4a19..1643eaa6f1 100644 --- a/src/include/nodes/execnodes.h +++ b/src/include/nodes/execnodes.h @@ -2519,6 +2519,14 @@ typedef struct WindowAggState int64 groupheadpos; /* current row's peer group head position */ int64 grouptailpos; /* " " " " tail position (group end+1) */ + /* these fields are used in Row pattern recognition: */ + RPSkipTo rpSkipTo; /* Row Pattern Skip To type */ + List *patternVariableList; /* list of row pattern variables names (list of String) */ + List *patternRegexpList; /* list of row pattern regular expressions ('+' or ''. list of String) */ + List *defineVariableList; /* list of row pattern definition variables (list of String) */ + List *defineClauseList; /* expression for row pattern definition + * search conditions ExprState list */ + MemoryContext partcontext; /* context for partition-lifespan data */ MemoryContext aggcontext; /* shared context for aggregate working data */ MemoryContext curaggcontext; /* current aggregate's working data */ @@ -2555,6 +2563,11 @@ typedef struct WindowAggState TupleTableSlot *agg_row_slot; TupleTableSlot *temp_slot_1; TupleTableSlot *temp_slot_2; + + /* temporary slots for RPR */ + TupleTableSlot *prev_slot; /* PREV row navigation operator */ + TupleTableSlot *next_slot; /* NEXT row navigation operator */ + TupleTableSlot *null_slot; /* all NULL slot */ } WindowAggState; /* ---------------- diff --git a/src/include/windowapi.h b/src/include/windowapi.h index b8c2c565d1..a0facf38fe 100644 --- a/src/include/windowapi.h +++ b/src/include/windowapi.h @@ -58,7 +58,16 @@ extern Datum WinGetFuncArgInFrame(WindowObject winobj, int argno, int relpos, int seektype, bool set_mark, bool *isnull, bool *isout); +extern int WinGetSlotInFrame(WindowObject winobj, TupleTableSlot *slot, + int relpos, int seektype, bool set_mark, + bool *isnull, bool *isout); + extern Datum WinGetFuncArgCurrent(WindowObject winobj, int argno, bool *isnull); +extern WindowAggState *WinGetAggState(WindowObject winobj); + +extern bool window_gettupleslot(WindowObject winobj, int64 pos, TupleTableSlot *slot); + + #endif /* WINDOWAPI_H */ -- 2.25.1 ----Next_Part(Mon_Jun_26_17_45_07_2023_724)-- Content-Type: Text/X-Patch; charset=us-ascii Content-Transfer-Encoding: 7bit Content-Disposition: inline; filename="v2-0005-Row-pattern-recognition-patch-docs.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 80+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 80+ messages in thread
end of thread, other threads:[~2026-04-06 22:07 UTC | newest] Thread overview: 80+ messages (download: mbox mbox.gz follow: Atom feed) -- links below jump to the message on this page -- 2022-06-17 01:33 [PATCH v1 2/2] WIP: Fix EEOP_JSON_CONSTRUCTOR and EEOP_JSONEXPR size. Andres Freund <[email protected]> 2023-06-26 08:05 [PATCH v2 4/7] Row pattern recognition patch (executor). Tatsuo Ishii <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[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