agora inbox for [email protected]  

help / color / mirror / Atom feed

[PATCH v5 4/7] Row pattern recognition patch (executor).
664+ messages / 3 participants
[nested] [flat]

* [PATCH v5 4/7] Row pattern recognition patch (executor).
@ 2023-09-02 06:32 Tatsuo Ishii <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Tatsuo Ishii @ 2023-09-02 06:32 UTC (permalink / raw)

---
 src/backend/executor/nodeWindowAgg.c | 671 ++++++++++++++++++++++++++-
 src/backend/utils/adt/windowfuncs.c  |  38 +-
 src/include/catalog/pg_proc.dat      |   6 +
 src/include/nodes/execnodes.h        |  19 +
 4 files changed, 722 insertions(+), 12 deletions(-)

diff --git a/src/backend/executor/nodeWindowAgg.c b/src/backend/executor/nodeWindowAgg.c
index 310ac23e3a..2e59369a71 100644
--- a/src/backend/executor/nodeWindowAgg.c
+++ b/src/backend/executor/nodeWindowAgg.c
@@ -36,6 +36,7 @@
 #include "access/htup_details.h"
 #include "catalog/objectaccess.h"
 #include "catalog/pg_aggregate.h"
+#include "catalog/pg_collation_d.h"
 #include "catalog/pg_proc.h"
 #include "executor/executor.h"
 #include "executor/nodeWindowAgg.h"
@@ -48,6 +49,7 @@
 #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"
@@ -182,8 +184,9 @@ static void begin_partition(WindowAggState *winstate);
 static void spool_tuples(WindowAggState *winstate, int64 pos);
 static void release_partition(WindowAggState *winstate);
 
-static int	row_is_in_frame(WindowAggState *winstate, int64 pos,
+static int  row_is_in_frame(WindowAggState *winstate, int64 pos,
 							TupleTableSlot *slot);
+
 static void update_frameheadpos(WindowAggState *winstate);
 static void update_frametailpos(WindowAggState *winstate);
 static void update_grouptailpos(WindowAggState *winstate);
@@ -195,9 +198,25 @@ 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 int	WinGetSlotInFrame(WindowObject winobj, TupleTableSlot *slot,
+							  int relpos, int seektype, bool set_mark,
+							  bool *isnull, bool *isout);
+static bool window_gettupleslot(WindowObject winobj, int64 pos, TupleTableSlot *slot);
+
+static void attno_map(Node *node);
+static bool attno_map_walker(Node *node, void *context);
+static int row_is_in_reduced_frame(WindowObject winobj, int64 pos);
+
+static int64 evaluate_pattern(WindowObject winobj, int64 current_pos, 
+							  char *vname, StringInfo encoded_str, bool *result);
+
+static bool get_slots(WindowObject winobj, int64 current_pos);
+
+static int search_str_set(char *pattern, StringInfo *str_set, int set_size);
+static void search_str_set_recurse(char *pattern, StringInfo *str_set, int set_size, int set_index,
+								   char *encoded_str, int *resultlen);
+static char	pattern_initial(WindowAggState *winstate, char *vname);
 
 /*
  * initialize_windowaggregate
@@ -673,6 +692,9 @@ eval_windowaggregates(WindowAggState *winstate)
 	WindowObject agg_winobj;
 	TupleTableSlot *agg_row_slot;
 	TupleTableSlot *temp_slot;
+	bool		reduced_frame_set;
+	bool		check_reduced_frame;
+	int			num_rows_in_reduced_frame;
 
 	numaggs = winstate->numaggs;
 	if (numaggs == 0)
@@ -790,6 +812,7 @@ eval_windowaggregates(WindowAggState *winstate)
 			(winstate->frameOptions & FRAMEOPTION_EXCLUSION) ||
 			winstate->aggregatedupto <= winstate->frameheadpos)
 		{
+			elog(DEBUG1, "peraggstate->restart  is set");
 			peraggstate->restart = true;
 			numaggs_restart++;
 		}
@@ -861,8 +884,10 @@ eval_windowaggregates(WindowAggState *winstate)
 	 * If we created a mark pointer for aggregates, keep it pushed up to frame
 	 * head, so that tuplestore can discard unnecessary rows.
 	 */
+#ifdef NOT_USED
 	if (agg_winobj->markptr >= 0)
 		WinSetMarkPosition(agg_winobj, winstate->frameheadpos);
+#endif
 
 	/*
 	 * Now restart the aggregates that require it.
@@ -919,6 +944,10 @@ eval_windowaggregates(WindowAggState *winstate)
 		ExecClearTuple(agg_row_slot);
 	}
 
+	reduced_frame_set = false;
+	check_reduced_frame = false;
+	num_rows_in_reduced_frame = 0;
+
 	/*
 	 * Advance until we reach a row not in frame (or end of partition).
 	 *
@@ -930,12 +959,18 @@ eval_windowaggregates(WindowAggState *winstate)
 	{
 		int			ret;
 
+		elog(DEBUG1, "===== loop in frame starts: " INT64_FORMAT, winstate->aggregatedupto);
+
 		/* Fetch next row if we didn't already */
 		if (TupIsNull(agg_row_slot))
 		{
 			if (!window_gettupleslot(agg_winobj, winstate->aggregatedupto,
 									 agg_row_slot))
+			{
+				if (check_reduced_frame)
+					winstate->aggregatedupto--;
 				break;			/* must be end of partition */
+			}
 		}
 
 		/*
@@ -944,10 +979,47 @@ eval_windowaggregates(WindowAggState *winstate)
 		 */
 		ret = row_is_in_frame(winstate, winstate->aggregatedupto, agg_row_slot);
 		if (ret < 0)
+		{
+			if (winstate->patternVariableList != NIL && check_reduced_frame)
+				winstate->aggregatedupto--;
 			break;
+		}
 		if (ret == 0)
 			goto next_tuple;
 
+		if (winstate->patternVariableList != NIL)
+		{
+			if (!reduced_frame_set)
+			{
+				num_rows_in_reduced_frame = row_is_in_reduced_frame(winstate->agg_winobj, winstate->aggregatedupto);
+				reduced_frame_set = true;
+				elog(DEBUG1, "set num_rows_in_reduced_frame: %d pos: " INT64_FORMAT,
+					 num_rows_in_reduced_frame, winstate->aggregatedupto);
+
+				if (num_rows_in_reduced_frame <= 0)
+					break;
+
+				else if (num_rows_in_reduced_frame > 0)
+					check_reduced_frame = true;
+			}
+
+			if (check_reduced_frame)
+			{
+				elog(DEBUG1, "decrease num_rows_in_reduced_frame: %d pos: " INT64_FORMAT,
+					 num_rows_in_reduced_frame, winstate->aggregatedupto);
+				num_rows_in_reduced_frame--;
+				if (num_rows_in_reduced_frame < 0)
+				{
+					/*
+					 * No more rows remain in the reduced frame. Finish
+					 * accumulating row into the aggregates.
+					 */
+					winstate->aggregatedupto--;
+					break;
+				}
+			}
+		}
+
 		/* Set tuple context for evaluation of aggregate arguments */
 		winstate->tmpcontext->ecxt_outertuple = agg_row_slot;
 
@@ -976,6 +1048,8 @@ next_tuple:
 		ExecClearTuple(agg_row_slot);
 	}
 
+	elog(DEBUG1, "===== break loop in frame starts: " INT64_FORMAT, winstate->aggregatedupto);
+
 	/* The frame's end is not supposed to move backwards, ever */
 	Assert(aggregatedupto_nonrestarted <= winstate->aggregatedupto);
 
@@ -2053,6 +2127,8 @@ ExecWindowAgg(PlanState *pstate)
 
 	CHECK_FOR_INTERRUPTS();
 
+	elog(DEBUG1, "ExecWindowAgg called. pos: " INT64_FORMAT , winstate->currentpos);
+
 	if (winstate->status == WINDOWAGG_DONE)
 		return NULL;
 
@@ -2388,6 +2464,9 @@ ExecInitWindowAgg(WindowAgg *node, EState *estate, int eflags)
 	TupleDesc	scanDesc;
 	ListCell   *l;
 
+	TargetEntry	*te;
+	Expr		*expr;
+
 	/* check for unsupported flags */
 	Assert(!(eflags & (EXEC_FLAG_BACKWARD | EXEC_FLAG_MARK)));
 
@@ -2483,6 +2562,16 @@ 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 +2756,39 @@ 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 */
+	winstate->defineInitial = node->defineInitial;
+	winstate->defineVariableList = NIL;
+	winstate->defineClauseList = NIL;
+	if (node->defineClause != NIL)
+	{
+		/*
+		 * Tweak arg var of PREV/NEXT so that it refers to scan/inner slot.
+		 */
+		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)));
+			attno_map((Node *)expr);
+			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 +2796,57 @@ ExecInitWindowAgg(WindowAgg *node, EState *estate, int eflags)
 	return winstate;
 }
 
+/*
+ * Rewrite varno of Var node that is the argument of PREV/NET so that it sees
+ * scan tuple (PREV) or inner tuple (NEXT).
+ */
+static void
+attno_map(Node *node)
+{
+	(void) expression_tree_walker(node, attno_map_walker, NULL);
+}
+
+static bool
+attno_map_walker(Node *node, void *context)
+{
+	FuncExpr	*func;
+	int			nargs;
+	Expr		*expr;
+	Var			*var;
+	
+	if (node == NULL)
+		return false;
+
+	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)
+				/*
+				 * Rewrite varno from OUTER_VAR to regular var no so that the
+				 * var references scan tuple.
+				 */
+				var->varno = var->varnosyn;
+			else
+				var->varno = INNER_VAR;
+			elog(DEBUG1, "PREV/NEXT's varno is rewritten to: %d", var->varno);
+		}
+	}
+	return expression_tree_walker(node, attno_map_walker, NULL);
+}
+
 /* -----------------
  * ExecEndWindowAgg
  * -----------------
@@ -2691,6 +2864,8 @@ 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 +2915,8 @@ 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)
@@ -3100,7 +3277,7 @@ window_gettupleslot(WindowObject winobj, int64 pos, TupleTableSlot *slot)
 		return false;
 
 	if (pos < winobj->markpos)
-		elog(ERROR, "cannot fetch row before WindowObject's mark position");
+		elog(ERROR, "cannot fetch row: " INT64_FORMAT " before WindowObject's mark position: " INT64_FORMAT,  pos, winobj->markpos );
 
 	oldcontext = MemoryContextSwitchTo(winstate->ss.ps.ps_ExprContext->ecxt_per_query_memory);
 
@@ -3420,14 +3597,54 @@ 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)
+ */
+static 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;
+	int			num_reduced_frame;
+
+	Assert(WindowObjectIsValid(winobj));
+	winstate = winobj->winstate;
+
 	switch (seektype)
 	{
 		case WINDOW_SEEK_CURRENT:
@@ -3494,6 +3711,12 @@ WinGetFuncArgInFrame(WindowObject winobj, int argno,
 						 winstate->frameOptions);
 					break;
 			}
+			num_reduced_frame = row_is_in_reduced_frame(winobj, winstate->frameheadpos);
+			if (num_reduced_frame < 0)
+				goto out_of_frame;
+			else if (num_reduced_frame > 0)
+				if (relpos >= num_reduced_frame)
+					goto out_of_frame;
 			break;
 		case WINDOW_SEEK_TAIL:
 			/* rejecting relpos > 0 is easy and simplifies code below */
@@ -3565,6 +3788,12 @@ WinGetFuncArgInFrame(WindowObject winobj, int argno,
 					mark_pos = 0;	/* keep compiler quiet */
 					break;
 			}
+
+			num_reduced_frame = row_is_in_reduced_frame(winobj, winstate->frameheadpos + relpos);
+			if (num_reduced_frame < 0)
+				goto out_of_frame;
+			else if (num_reduced_frame > 0)
+				abs_pos = winstate->frameheadpos + relpos + num_reduced_frame - 1;
 			break;
 		default:
 			elog(ERROR, "unrecognized window seek type: %d", seektype);
@@ -3583,15 +3812,13 @@ 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 +3849,427 @@ WinGetFuncArgCurrent(WindowObject winobj, int argno, bool *isnull)
 	return ExecEvalExpr((ExprState *) list_nth(winobj->argstates, argno),
 						econtext, isnull);
 }
+
+/*
+ * row_is_in_reduced_frame
+ * Determine whether a row is in the current row's reduced window frame according
+ * to row pattern matching
+ *
+ * The row must has been already determined that it is in a full window frame
+ * and fetched it into slot.
+ *
+ * Returns:
+ * = 0, RPR is not defined.
+ * >0, if the row is the first in the reduced frame. Return the number of rows in the reduced frame.
+ * -1, if the row is unmatched row
+ * -2, if the row is in the reduced frame but needed to be skipped because of
+ * AFTER MATCH SKIP PAST LAST ROW
+ */
+static
+int row_is_in_reduced_frame(WindowObject winobj, int64 pos)
+{
+	WindowAggState *winstate = winobj->winstate;
+	ListCell	*lc1, *lc2;
+	bool		expression_result;
+	int			num_matched_rows;
+	int64		original_pos;
+	bool		anymatch;
+	StringInfo	encoded_str;
+	StringInfo	pattern_str = makeStringInfo();
+
+	/*
+	 * Array of pattern variables evaluted to true.
+	 * Each character corresponds to pattern variable.
+	 * Example:
+	 * str_set[0] = "AB";
+	 * str_set[1] = "AC";
+	 * In this case at row 0 A and B are true, and A and C are true in row 1.
+	 */
+	#define ENCODED_STR_ARRAY_ALLOC_SIZE	128
+	StringInfo	*str_set = NULL;
+	int		str_set_index;
+	int		str_set_size;
+
+	if (winstate->patternVariableList == NIL)
+	{
+		/*
+		 * RPR is not defined. Assume that we are always in the the reduced
+		 * window frame.
+		 */
+		return 0;
+	}
+
+	/* save original pos */
+	original_pos = pos;
+
+	/*
+	 * Check whether the row speicied by pos is in the reduced frame. The
+	 * second and subsequent rows need to be recognized as "unmatched" rows if
+	 * AFTER MATCH SKIP PAST LAST ROW is defined.
+	 */
+	if (winstate->rpSkipTo == ST_PAST_LAST_ROW &&
+		pos > winstate->headpos_in_reduced_frame &&
+		pos < (winstate->headpos_in_reduced_frame + winstate->num_rows_in_reduced_frame))
+		return -2;
+		
+	/*
+	 * Loop over until none of pattern matches or encounters end of frame.
+	 */
+	for (;;)
+	{
+		int64	result_pos = -1;
+
+		/*
+		 * Loop over each PATTERN variable.
+		 */
+		anymatch = false;
+		encoded_str = makeStringInfo();
+
+		forboth(lc1, winstate->patternVariableList, lc2, winstate->patternRegexpList)
+		{
+			char	*vname = strVal(lfirst(lc1));
+			char	*quantifier = strVal(lfirst(lc2));
+
+			elog(DEBUG1, "pos: " INT64_FORMAT " pattern vname: %s quantifier: %s", pos, vname, quantifier);
+
+			expression_result = false;
+
+			/* evaluate row pattern against current row */
+			result_pos = evaluate_pattern(winobj, pos, vname, encoded_str, &expression_result);
+			if (expression_result)
+			{
+				elog(DEBUG1, "expression result is true");
+				anymatch = true;
+			}
+
+			/*
+			 * If out of frame, we are done.
+			 */
+			 if (result_pos < 0)
+				 break;
+		}
+
+		if (!anymatch)
+		{
+			/* none of patterns matched. */
+			break;
+		}
+
+		/* build encoded string array */
+		if (str_set == NULL)
+		{
+			str_set_index = 0;
+			str_set_size = ENCODED_STR_ARRAY_ALLOC_SIZE * sizeof(StringInfo);
+			str_set = palloc(str_set_size);
+		}
+
+		str_set[str_set_index++] = encoded_str;
+
+		elog(DEBUG1, "pos: " INT64_FORMAT " str_set_index: %d encoded_str: %s", pos, str_set_index, encoded_str->data);
+
+		if (str_set_index >= str_set_size)
+		{
+			str_set_size *= 2;
+			str_set = repalloc(str_set, str_set_size);
+		}
+
+		/* move to next row */
+		pos++;
+
+		if (result_pos < 0)
+		{
+			/* out of frame */
+			break;
+		}
+	}
+
+	if (str_set == NULL)
+	{
+		/* no matches found in the first row */
+		return -1;
+	}
+
+	elog(DEBUG1, "pos: " INT64_FORMAT " encoded_str: %s", pos, encoded_str->data);
+
+	/* build regular expression */
+	pattern_str = makeStringInfo();
+	appendStringInfoChar(pattern_str, '^');
+	forboth (lc1, winstate->patternVariableList, lc2, winstate->patternRegexpList)
+	{
+		char	*vname = strVal(lfirst(lc1));
+		char	*quantifier = strVal(lfirst(lc2));
+		char	 initial;
+
+		initial = pattern_initial(winstate, vname);
+		Assert(initial != 0);
+		appendStringInfoChar(pattern_str, initial);
+		if (quantifier[0])
+			appendStringInfoChar(pattern_str, quantifier[0]);
+		elog(DEBUG1, "vname: %s initial: %c quantifier: %s", vname, initial, quantifier);
+	}
+
+	elog(DEBUG1, "pos: " INT64_FORMAT " pattern: %s", pos, pattern_str->data);
+
+	/* look for matching pattern variable sequence */
+	num_matched_rows = search_str_set(pattern_str->data, str_set, str_set_index);
+	if (num_matched_rows <= 0)
+		return -1;
+
+	/*
+	 * We are at the first row in the reduced frame.  Save the number of
+	 * matched rows as the number of rows in the reduced frame.
+	 */
+	winstate->headpos_in_reduced_frame = original_pos;
+	winstate->num_rows_in_reduced_frame = num_matched_rows;
+
+	return num_matched_rows;
+}
+
+/*
+ * search set of encode_str.
+ * set_size: size of set_str array.
+ */
+static
+int search_str_set(char *pattern, StringInfo *str_set, int set_size)
+{
+	char		*encoded_str = palloc0(set_size+1);
+	int			resultlen = 0;
+
+	search_str_set_recurse(pattern, str_set, set_size, 0, encoded_str, &resultlen);
+	elog(DEBUG1, "search_str_set returns %d", resultlen);
+	return resultlen;
+}
+
+static
+void search_str_set_recurse(char *pattern, StringInfo *str_set,
+							int set_size, int set_index, char *encoded_str, int *resultlen)
+{
+	char	*p;
+
+	if (set_index >= set_size)
+	{
+		Datum	d;
+		text	*res;
+		char	*substr;
+
+		/*
+		 * We first perform pattern matching using regexp_instr, then call
+		 * textregexsubstr to get matched substring to know how log the
+		 * matched string is. That is the number of rows in the reduced window
+		 * frame.  The reason why we can't call textregexsubstr is, it error
+		 * out if pattern is not match.
+		 */
+		if (DatumGetInt32(DirectFunctionCall2Coll(regexp_instr, DEFAULT_COLLATION_OID,
+												  PointerGetDatum(cstring_to_text(encoded_str)),
+												  PointerGetDatum(cstring_to_text(pattern)))) > 0)
+		{
+			d = DirectFunctionCall2Coll(textregexsubstr,
+										DEFAULT_COLLATION_OID,
+										PointerGetDatum(cstring_to_text(encoded_str)),
+										PointerGetDatum(cstring_to_text(pattern)));
+			if (d != 0)
+			{
+				int		len;
+
+				res = DatumGetTextPP(d);
+				substr = text_to_cstring(res);
+				len = strlen(substr);
+				if (len > *resultlen)
+					/* remember the longest match */
+					*resultlen = len;
+			}
+		}
+		return;
+	}
+
+	p = str_set[set_index]->data;
+	while (*p)
+	{
+		encoded_str[set_index] = *p;
+		p++;
+		search_str_set_recurse(pattern, str_set, set_size, set_index + 1, encoded_str, resultlen);
+	}
+}
+
+
+/*
+ * 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 absolute row position
+ * other wise out of frame.
+ */
+static
+int64 evaluate_pattern(WindowObject winobj, int64 current_pos, 
+						char *vname, StringInfo encoded_str, bool *result)
+{
+	WindowAggState	*winstate = winobj->winstate;
+	ExprContext		*econtext = winstate->ss.ps.ps_ExprContext;
+	ListCell		*lc1, *lc2, *lc3;
+	ExprState		*pat;
+	Datum			eval_result;
+	bool			out_of_frame = false;
+	bool			isnull;
+
+	forthree (lc1, winstate->defineVariableList, lc2, winstate->defineClauseList, lc3, winstate->defineInitial)
+	{
+		char	initial;
+		char	*name = strVal(lfirst(lc1));
+
+		elog(DEBUG1, "evaluate_pattern: define variable: %s, pattern variable: %s", name, vname);
+
+		if (strcmp(vname, name))
+			continue;
+
+		initial = *(strVal(lfirst(lc3)));
+
+		/* set expression to evaluate */
+		pat = lfirst(lc2);
+
+		/* get current, previous and next tuples */
+		if (!get_slots(winobj, current_pos))
+		{
+			out_of_frame = true;
+		}
+		else
+		{
+			/* evaluate the expression */
+			eval_result = ExecEvalExpr(pat, econtext, &isnull);
+			if (isnull)
+			{
+				/* expression is NULL */
+				elog(DEBUG1, "expression for %s is NULL at row: " INT64_FORMAT, vname, current_pos);
+				*result = false;
+			}
+			else
+			{
+				if (!DatumGetBool(eval_result))
+				{
+					/* expression is false */
+					elog(DEBUG1, "expression for %s is false at row: " INT64_FORMAT, vname, current_pos);
+					*result = false;
+				}
+				else
+				{
+					/* expression is true */
+					elog(DEBUG1, "expression for %s is true at row: " INT64_FORMAT, vname, current_pos);
+					appendStringInfoChar(encoded_str, initial);
+					*result = true;
+				}
+			}
+			break;
+		}
+
+		if (out_of_frame)
+		{
+			*result = false;
+			return -1;
+		}
+	}
+	return current_pos;
+}
+
+/*
+ * Get current, previous and next tuples.
+ * Returns false if current row is out of partition/full frame.
+ */
+static
+bool get_slots(WindowObject winobj, int64 current_pos)
+{
+	WindowAggState *winstate = winobj->winstate;
+	TupleTableSlot *slot;
+	int		ret;
+	ExprContext *econtext;
+
+	econtext = winstate->ss.ps.ps_ExprContext;
+
+	/* set up current row tuple slot */
+	slot = winstate->temp_slot_1;
+	if (!window_gettupleslot(winobj, current_pos, slot))
+	{
+		elog(DEBUG1, "current row is out of partition at:" INT64_FORMAT, current_pos);
+		return false;
+
+		ret = row_is_in_frame(winstate, current_pos, slot);
+		if (ret <= 0)
+		{
+			elog(DEBUG1, "current row is out of frame at: " INT64_FORMAT, current_pos);
+			return false;
+		}
+	}
+	econtext->ecxt_outertuple = slot;
+
+	/* for PREV */
+	if (current_pos > 0)
+	{
+		slot = winstate->prev_slot;
+		if (!window_gettupleslot(winobj, current_pos - 1, slot))
+		{
+			elog(DEBUG1, "previous row is out of partition at: " INT64_FORMAT, current_pos - 1);
+			econtext->ecxt_scantuple = winstate->null_slot;
+		}
+		else
+		{
+			ret = row_is_in_frame(winstate, current_pos - 1, slot);
+			if (ret <= 0)
+			{
+				elog(DEBUG1, "previous row is out of frame at: " INT64_FORMAT, current_pos - 1);
+				econtext->ecxt_scantuple = winstate->null_slot;
+			}
+			else
+			{
+				econtext->ecxt_scantuple = slot;
+			}
+		}
+	}
+	else
+		econtext->ecxt_scantuple = winstate->null_slot;
+
+	/* for NEXT */
+	slot = winstate->next_slot;
+	if (!window_gettupleslot(winobj, current_pos + 1, slot))
+	{
+		elog(DEBUG1, "next row is out of partiton at: " INT64_FORMAT, current_pos + 1);
+		econtext->ecxt_innertuple = winstate->null_slot;
+	}
+	else
+	{
+		ret = row_is_in_frame(winstate, current_pos + 1, slot);
+		if (ret <= 0)
+		{
+			elog(DEBUG1, "next row is out of frame at: " INT64_FORMAT, current_pos + 1);
+			econtext->ecxt_innertuple = winstate->null_slot;
+		}
+		else
+			econtext->ecxt_innertuple = slot;
+	}
+	return true;
+}
+
+/*
+ * Return pattern variable initial character
+ * matching with pattern variable name vname.
+ * If not found, return 0.
+ */
+static
+char	pattern_initial(WindowAggState *winstate, char *vname)
+{
+	char		initial;
+	char		*name;
+	ListCell	*lc1, *lc2;
+
+	forboth (lc1, winstate->defineVariableList, lc2, winstate->defineInitial)
+	{
+		name = strVal(lfirst(lc1));				/* DEFINE variable name */
+		initial = *(strVal(lfirst(lc2)));		/* DEFINE variable initial */
+
+
+		if (!strcmp(name, vname))
+				return initial;					/* found */
+	}
+	return 0;
+}
diff --git a/src/backend/utils/adt/windowfuncs.c b/src/backend/utils/adt/windowfuncs.c
index b87a624fb2..e4cab36ec9 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"
@@ -36,11 +39,19 @@ 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);
 
-
 /*
  * utility routine for *_rank functions.
  */
@@ -673,7 +684,7 @@ window_last_value(PG_FUNCTION_ARGS)
 	bool		isnull;
 
 	result = WinGetFuncArgInFrame(winobj, 0,
-								  0, WINDOW_SEEK_TAIL, true,
+								  0, WINDOW_SEEK_TAIL, false,
 								  &isnull, NULL);
 	if (isnull)
 		PG_RETURN_NULL();
@@ -713,3 +724,26 @@ window_nth_value(PG_FUNCTION_ARGS)
 
 	PG_RETURN_DATUM(result);
 }
+
+/*
+ * 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 9805bc6118..d20f803cf5 100644
--- a/src/include/catalog/pg_proc.dat
+++ b/src/include/catalog/pg_proc.dat
@@ -10416,6 +10416,12 @@
 { 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 => 'previous value',
+  proname => 'prev', provolatile => 's', prorettype => 'anyelement',
+  proargtypes => 'anyelement', prosrc => 'window_prev' },
+{ oid => '6123', 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..2bd6fcb5e1 100644
--- a/src/include/nodes/execnodes.h
+++ b/src/include/nodes/execnodes.h
@@ -2519,6 +2519,15 @@ 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 */
+	List	   *defineInitial;		/* list of row pattern definition variable initials (list of String) */
+
 	MemoryContext partcontext;	/* context for partition-lifespan data */
 	MemoryContext aggcontext;	/* shared context for aggregate working data */
 	MemoryContext curaggcontext;	/* current aggregate's working data */
@@ -2555,6 +2564,16 @@ 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 */
+
+	/* head of the reduced window frame */
+	int64		headpos_in_reduced_frame;
+	/* number of rows in the reduced window frame */
+	int64		num_rows_in_reduced_frame;
 } WindowAggState;
 
 /* ----------------
-- 
2.25.1


----Next_Part(Sat_Sep__2_15_52_35_2023_273)--
Content-Type: Text/X-Patch; charset=us-ascii
Content-Transfer-Encoding: 7bit
Content-Disposition: inline;
 filename="v5-0005-Row-pattern-recognition-patch-docs.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread

* [PATCH v45 7/8] Error out any process that would block at REPACK
@ 2026-03-25 19:35 Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Álvaro Herrera @ 2026-03-25 19:35 UTC (permalink / raw)

Any process waiting on REPACK to release its lock would actually cause
it to deadlock when it tries to upgrade its lock to AEL, losing all work
done to that point.  We avoid this by teaching the deadlock detector to
raise an error when this condition is detected.
---
 src/backend/commands/repack.c                 | 53 ++++++++----
 src/backend/storage/lmgr/deadlock.c           | 15 ++++
 src/include/storage/proc.h                    |  6 +-
 src/test/modules/injection_points/Makefile    |  1 +
 .../expected/repack_deadlock.out              | 63 ++++++++++++++
 src/test/modules/injection_points/meson.build |  1 +
 .../specs/repack_deadlock.spec                | 83 +++++++++++++++++++
 7 files changed, 202 insertions(+), 20 deletions(-)
 create mode 100644 src/test/modules/injection_points/expected/repack_deadlock.out
 create mode 100644 src/test/modules/injection_points/specs/repack_deadlock.spec

diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c
index d25ab43da7b..7291b4180ab 100644
--- a/src/backend/commands/repack.c
+++ b/src/backend/commands/repack.c
@@ -276,6 +276,16 @@ ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
 	/* Determine the lock mode to use. */
 	lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
 
+	/*
+	 * If in concurrent mode, set the PROC_IN_CONCURRENT_REPACK flag.  This
+	 * makes the deadlock checker cause anyone that would conflict with us
+	 * to error out.  It's important to set this flag ahead of actually locking
+	 * the relation; it won't of course affect anyone until we do have a lock
+	 * that others can conflict with.
+	 */
+	if ((params.options & CLUOPT_CONCURRENT) != 0)
+		MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK;
+
 	/*
 	 * If a single relation is specified, process it and we're done ... unless
 	 * the relation is a partitioned table, in which case we fall through.
@@ -476,11 +486,8 @@ RepackLockLevel(bool concurrent)
  * If indexOid is InvalidOid, the table will be rewritten in physical order
  * instead of index order.
  *
- * Note that, in the concurrent case, the function releases the lock at some
- * point, in order to get AccessExclusiveLock for the final steps (i.e. to
- * swap the relation files). To make things simpler, the caller should expect
- * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
- * AccessExclusiveLock is kept till the end of the transaction.)
+ * On return, OldHeap is closed but locked with AccessExclusiveLock - the lock
+ * will be released at end of the transaction.
  *
  * 'cmd' indicates which command is being executed, to be used for error
  * messages.
@@ -512,10 +519,12 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
 		/*
 		 * Make sure we're not in a transaction block.
 		 *
-		 * The reason is that repack_setup_logical_decoding() could deadlock
-		 * if there's an XID already assigned.  It would be possible to run in
-		 * a transaction block if we had no XID, but this restriction is
-		 * simpler for users to understand and we don't lose anything.
+		 * The reason is that repack_setup_logical_decoding() could wait
+		 * indefinitely for our XID to complete. (The deadlock detector would
+		 * not recognize it because we'd be waiting for ourselves, i.e. no
+		 * real lock conflict.) It would be possible to run in a transaction
+		 * block if we had no XID, but this restriction is simpler for users
+		 * to understand and we don't lose anything.
 		 */
 		PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
 
@@ -998,10 +1007,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose,
 		 * Note that the worker has to wait for all transactions with XID
 		 * already assigned to finish. If some of those transactions is
 		 * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
-		 * table (e.g.  it runs CREATE INDEX), we can end up in a deadlock.
-		 * Not sure this risk is worth unlocking/locking the table (and its
-		 * clustering index) and checking again if it's still eligible for
-		 * REPACK CONCURRENTLY.
+		 * table (e.g. it runs CREATE INDEX), it should encounter ERROR in the
+		 * deadlock checking code.
 		 */
 		start_repack_decoding_worker(tableOid);
 
@@ -3119,10 +3126,18 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
 		LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
 
 	/*
-	 * Tuples and pages of the old heap will be gone, but the heap will stay.
+	 * Now that we have all access-exclusive locks on all relations, we no
+	 * longer want other processes to error out when trying to acquire a
+	 * conflicting lock.  Therefore, unset our flag.
+	 */
+	MyProc->statusFlags &= ~PROC_IN_CONCURRENT_REPACK;
+
+	/*
+	 * Tuples and pages of the old heap will be gone, but the heap itself will
+	 * stay.  In order for predicate locks to continue to work, convert them
+	 * to relation-level locks.  We do this both for table and indexes.
 	 */
 	TransferPredicateLocksToHeapRelation(OldHeap);
-	/* The same for indexes. */
 	for (int i = 0; i < nind; i++)
 	{
 		Relation	index = ind_refs[i];
@@ -3342,9 +3357,11 @@ start_repack_decoding_worker(Oid relid)
 
 	/*
 	 * The decoding setup must be done before the caller can have XID assigned
-	 * for any reason, otherwise the worker might end up in a deadlock,
-	 * waiting for the caller's transaction to end. Therefore wait here until
-	 * the worker indicates that it has the logical decoding initialized.
+	 * for any reason, otherwise the worker might end up waiting for the
+	 * caller's transaction to end. (Deadlock detector does not consider this
+	 * a conflict because the worker is in the same locking group as the
+	 * backend that launched it.) Therefore wait here until the worker
+	 * indicates that it has the logical decoding initialized.
 	 */
 	ConditionVariablePrepareToSleep(&shared->cv);
 	for (;;)
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index b8962d875b6..3d3ec0da111 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -620,6 +620,21 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 						proc->statusFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
+					/*
+					 * Similarly, if we note that we're blocked by some process
+					 * running REPACK (CONCURRENTLY), just fail.  That process
+					 * is going to upgrade its lock at some point, and it would
+					 * be inappropriate for any other process to cause that
+					 * to fail.
+					 */
+					if (checkProc == MyProc &&
+						proc->statusFlags & PROC_IN_CONCURRENT_REPACK)
+						ereport(ERROR,
+								errcode(ERRCODE_OBJECT_IN_USE),
+								errmsg("could not wait for concurrent REPACK"),
+								errdetail("Process %d waits for REPACK running on process %d",
+									   MyProc->pid, proc->pid));
+
 					/* We're done looking at this proclock */
 					break;
 				}
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1dad125706e..8ad9718f3d6 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -69,10 +69,12 @@ struct XidCache
 #define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be
 												 * included in vacuum horizons
 												 * in all databases */
+#define		PROC_IN_CONCURRENT_REPACK	0x40	/* REPACK (CONCURRENTLY) */
 
-/* flags reset at EOXact */
+/* flags reset at EOXact.  A bit of a misnomer ... */
 #define		PROC_VACUUM_STATE_MASK \
-	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)
+	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | \
+	 PROC_IN_CONCURRENT_REPACK)
 
 /*
  * Xmin-related flags. Make sure any flags that affect how the process' Xmin
diff --git a/src/test/modules/injection_points/Makefile b/src/test/modules/injection_points/Makefile
index 2cd7d87c533..f7663859fe2 100644
--- a/src/test/modules/injection_points/Makefile
+++ b/src/test/modules/injection_points/Makefile
@@ -15,6 +15,7 @@ REGRESS_OPTS = --dlpath=$(top_builddir)/src/test/regress
 ISOLATION = basic \
 	    inplace \
 	    repack \
+	    repack_deadlock \
 	    repack_toast \
 	    syscache-update-pruned \
 	    heap_lock_update
diff --git a/src/test/modules/injection_points/expected/repack_deadlock.out b/src/test/modules/injection_points/expected/repack_deadlock.out
new file mode 100644
index 00000000000..a86e4767536
--- /dev/null
+++ b/src/test/modules/injection_points/expected/repack_deadlock.out
@@ -0,0 +1,63 @@
+Parsed test spec with 2 sessions
+
+starting permutation: wait_before_lock add_column wakeup_before_lock check1
+injection_points_attach
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: 
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+ <waiting ...>
+step add_column: 
+	alter table repack_deadlock add column noise text;
+ <waiting ...>
+step add_column: <... completed>
+ERROR:  could not wait for concurrent REPACK
+step wakeup_before_lock: 
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+
+injection_points_wakeup
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: <... completed>
+step check1: 
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+
+count
+-----
+    1
+(1 row)
+
+i|j
+-+-
+1|1
+2|2
+3|3
+4|4
+(4 rows)
+
+count
+-----
+    4
+(1 row)
+
+injection_points_detach
+-----------------------
+                       
+(1 row)
+
diff --git a/src/test/modules/injection_points/meson.build b/src/test/modules/injection_points/meson.build
index a414abb924b..1cd88d6db65 100644
--- a/src/test/modules/injection_points/meson.build
+++ b/src/test/modules/injection_points/meson.build
@@ -46,6 +46,7 @@ tests += {
       'basic',
       'inplace',
       'repack',
+      'repack_deadlock',
       'repack_toast',
       'syscache-update-pruned',
       'heap_lock_update',
diff --git a/src/test/modules/injection_points/specs/repack_deadlock.spec b/src/test/modules/injection_points/specs/repack_deadlock.spec
new file mode 100644
index 00000000000..9d23a6588c2
--- /dev/null
+++ b/src/test/modules/injection_points/specs/repack_deadlock.spec
@@ -0,0 +1,83 @@
+# Test REPACK with a concurrent transaction that would cause a deadlock
+setup
+{
+	CREATE EXTENSION injection_points;
+
+	CREATE TABLE repack_deadlock(i int PRIMARY KEY, j int);
+	INSERT INTO repack_deadlock(i, j) VALUES (1, 1), (2, 2), (3, 3), (4, 4);
+
+	CREATE TABLE relfilenodes(node oid);
+
+	CREATE TABLE data_s1(i int, j int);
+	CREATE TABLE data_s2(i int, j int);
+}
+
+teardown
+{
+	DROP TABLE repack_deadlock;
+	DROP EXTENSION injection_points;
+
+	DROP TABLE relfilenodes;
+	DROP TABLE data_s1;
+	DROP TABLE data_s2;
+}
+
+session s1
+setup
+{
+	SELECT injection_points_set_local();
+	SELECT injection_points_attach('repack-concurrently-before-lock', 'wait');
+}
+# Perform the initial load and wait for s2 to do some data changes.
+step wait_before_lock
+{
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+}
+# Check the table from the perspective of s1.
+#
+# Besides the contents, we also check that relfilenode has changed.
+
+# Have each session write the contents into a table and use FULL JOIN to check
+# if the outputs are identical.
+step check1
+{
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+}
+teardown
+{
+	SELECT injection_points_detach('repack-concurrently-before-lock');
+}
+
+session s2
+# Change the existing data. UPDATE changes both key and non-key columns. Also
+# update one row twice to test whether tuple version generated by this session
+# can be found.
+step add_column
+{
+	alter table repack_deadlock add column noise text;
+}
+
+step wakeup_before_lock
+{
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+}
+
+# Test if data changes introduced while one session is performing REPACK
+# CONCURRENTLY find their way into the table.
+permutation
+	wait_before_lock
+	add_column
+	wakeup_before_lock
+	check1
-- 
2.47.3


--cldir6umqisr673d
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v45-0008-Teach-snapshot-builder-to-skip-transactions-runn.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread

* [PATCH v47 8/9] Error out any process that would block at REPACK
@ 2026-03-25 19:35 Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Álvaro Herrera @ 2026-03-25 19:35 UTC (permalink / raw)

Any process waiting on REPACK to release its lock would actually cause
it to deadlock when it tries to upgrade its lock to AEL, losing all work
done to that point.  We avoid this by teaching the deadlock detector to
raise an error when this condition is detected.
---
 src/backend/commands/repack.c                 | 52 ++++++++----
 src/backend/storage/lmgr/deadlock.c           | 15 ++++
 src/include/storage/proc.h                    |  6 +-
 src/test/modules/injection_points/Makefile    |  1 +
 .../expected/repack_deadlock.out              | 63 ++++++++++++++
 src/test/modules/injection_points/meson.build |  1 +
 .../specs/repack_deadlock.spec                | 83 +++++++++++++++++++
 7 files changed, 202 insertions(+), 19 deletions(-)
 create mode 100644 src/test/modules/injection_points/expected/repack_deadlock.out
 create mode 100644 src/test/modules/injection_points/specs/repack_deadlock.spec

diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c
index 9c807f75d71..3acb2e1d2ba 100644
--- a/src/backend/commands/repack.c
+++ b/src/backend/commands/repack.c
@@ -276,6 +276,16 @@ ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
 	/* Determine the lock mode to use. */
 	lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
 
+	/*
+	 * If in concurrent mode, set the PROC_IN_CONCURRENT_REPACK flag.  This
+	 * makes the deadlock checker cause anyone that would conflict with us
+	 * to error out.  It's important to set this flag ahead of actually locking
+	 * the relation; it won't of course affect anyone until we do have a lock
+	 * that others can conflict with.
+	 */
+	if ((params.options & CLUOPT_CONCURRENT) != 0)
+		MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK;
+
 	/*
 	 * If a single relation is specified, process it and we're done ... unless
 	 * the relation is a partitioned table, in which case we fall through.
@@ -476,11 +486,8 @@ RepackLockLevel(bool concurrent)
  * If indexOid is InvalidOid, the table will be rewritten in physical order
  * instead of index order.
  *
- * Note that, in the concurrent case, the function releases the lock at some
- * point, in order to get AccessExclusiveLock for the final steps (i.e. to
- * swap the relation files). To make things simpler, the caller should expect
- * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
- * AccessExclusiveLock is kept till the end of the transaction.)
+ * On return, OldHeap is closed but locked with AccessExclusiveLock - the lock
+ * will be released at end of the transaction.
  *
  * 'cmd' indicates which command is being executed, to be used for error
  * messages.
@@ -512,10 +519,12 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
 		/*
 		 * Make sure we're not in a transaction block.
 		 *
-		 * The reason is that repack_setup_logical_decoding() could deadlock
-		 * if there's an XID already assigned.  It would be possible to run in
-		 * a transaction block if we had no XID, but this restriction is
-		 * simpler for users to understand and we don't lose anything.
+		 * The reason is that repack_setup_logical_decoding() could wait
+		 * indefinitely for our XID to complete. (The deadlock detector would
+		 * not recognize it because we'd be waiting for ourselves, i.e. no
+		 * real lock conflict.) It would be possible to run in a transaction
+		 * block if we had no XID, but this restriction is simpler for users
+		 * to understand and we don't lose anything.
 		 */
 		PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
 
@@ -998,10 +1007,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose,
 		 * Note that the worker has to wait for all transactions with XID
 		 * already assigned to finish. If some of those transactions is
 		 * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
-		 * table (e.g.  it runs CREATE INDEX), we can end up in a deadlock.
-		 * Not sure this risk is worth unlocking/locking the table (and its
-		 * clustering index) and checking again if it's still eligible for
-		 * REPACK CONCURRENTLY.
+		 * table (e.g. it runs CREATE INDEX), it should encounter ERROR in the
+		 * deadlock checking code.
 		 */
 		start_repack_decoding_worker(tableOid);
 
@@ -3090,7 +3097,16 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
 		LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
 
 	/*
-	 * Tuples and pages of the old heap will be gone, but the heap will stay.
+	 * Now that we have all access-exclusive locks on all relations, we no
+	 * longer want other processes to error out when trying to acquire a
+	 * conflicting lock.  Therefore, unset our flag.
+	 */
+	MyProc->statusFlags &= ~PROC_IN_CONCURRENT_REPACK;
+
+	/*
+	 * Tuples and pages of the old heap will be gone, but the heap itself will
+	 * stay.  In order for predicate locks to continue to work, convert them
+	 * to relation-level locks.  We do this both for table and indexes.
 	 */
 	TransferPredicateLocksToHeapRelation(OldHeap);
 	foreach_ptr(RelationData, index, indexrels)
@@ -3287,9 +3303,11 @@ start_repack_decoding_worker(Oid relid)
 
 	/*
 	 * The decoding setup must be done before the caller can have XID assigned
-	 * for any reason, otherwise the worker might end up in a deadlock,
-	 * waiting for the caller's transaction to end. Therefore wait here until
-	 * the worker indicates that it has the logical decoding initialized.
+	 * for any reason, otherwise the worker might end up waiting for the
+	 * caller's transaction to end. (Deadlock detector does not consider this
+	 * a conflict because the worker is in the same locking group as the
+	 * backend that launched it.) Therefore wait here until the worker
+	 * indicates that it has the logical decoding initialized.
 	 */
 	ConditionVariablePrepareToSleep(&shared->cv);
 	for (;;)
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index b8962d875b6..3d3ec0da111 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -620,6 +620,21 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 						proc->statusFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
+					/*
+					 * Similarly, if we note that we're blocked by some process
+					 * running REPACK (CONCURRENTLY), just fail.  That process
+					 * is going to upgrade its lock at some point, and it would
+					 * be inappropriate for any other process to cause that
+					 * to fail.
+					 */
+					if (checkProc == MyProc &&
+						proc->statusFlags & PROC_IN_CONCURRENT_REPACK)
+						ereport(ERROR,
+								errcode(ERRCODE_OBJECT_IN_USE),
+								errmsg("could not wait for concurrent REPACK"),
+								errdetail("Process %d waits for REPACK running on process %d",
+									   MyProc->pid, proc->pid));
+
 					/* We're done looking at this proclock */
 					break;
 				}
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1dad125706e..8ad9718f3d6 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -69,10 +69,12 @@ struct XidCache
 #define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be
 												 * included in vacuum horizons
 												 * in all databases */
+#define		PROC_IN_CONCURRENT_REPACK	0x40	/* REPACK (CONCURRENTLY) */
 
-/* flags reset at EOXact */
+/* flags reset at EOXact.  A bit of a misnomer ... */
 #define		PROC_VACUUM_STATE_MASK \
-	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)
+	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | \
+	 PROC_IN_CONCURRENT_REPACK)
 
 /*
  * Xmin-related flags. Make sure any flags that affect how the process' Xmin
diff --git a/src/test/modules/injection_points/Makefile b/src/test/modules/injection_points/Makefile
index 2cd7d87c533..f7663859fe2 100644
--- a/src/test/modules/injection_points/Makefile
+++ b/src/test/modules/injection_points/Makefile
@@ -15,6 +15,7 @@ REGRESS_OPTS = --dlpath=$(top_builddir)/src/test/regress
 ISOLATION = basic \
 	    inplace \
 	    repack \
+	    repack_deadlock \
 	    repack_toast \
 	    syscache-update-pruned \
 	    heap_lock_update
diff --git a/src/test/modules/injection_points/expected/repack_deadlock.out b/src/test/modules/injection_points/expected/repack_deadlock.out
new file mode 100644
index 00000000000..a86e4767536
--- /dev/null
+++ b/src/test/modules/injection_points/expected/repack_deadlock.out
@@ -0,0 +1,63 @@
+Parsed test spec with 2 sessions
+
+starting permutation: wait_before_lock add_column wakeup_before_lock check1
+injection_points_attach
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: 
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+ <waiting ...>
+step add_column: 
+	alter table repack_deadlock add column noise text;
+ <waiting ...>
+step add_column: <... completed>
+ERROR:  could not wait for concurrent REPACK
+step wakeup_before_lock: 
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+
+injection_points_wakeup
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: <... completed>
+step check1: 
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+
+count
+-----
+    1
+(1 row)
+
+i|j
+-+-
+1|1
+2|2
+3|3
+4|4
+(4 rows)
+
+count
+-----
+    4
+(1 row)
+
+injection_points_detach
+-----------------------
+                       
+(1 row)
+
diff --git a/src/test/modules/injection_points/meson.build b/src/test/modules/injection_points/meson.build
index a414abb924b..1cd88d6db65 100644
--- a/src/test/modules/injection_points/meson.build
+++ b/src/test/modules/injection_points/meson.build
@@ -46,6 +46,7 @@ tests += {
       'basic',
       'inplace',
       'repack',
+      'repack_deadlock',
       'repack_toast',
       'syscache-update-pruned',
       'heap_lock_update',
diff --git a/src/test/modules/injection_points/specs/repack_deadlock.spec b/src/test/modules/injection_points/specs/repack_deadlock.spec
new file mode 100644
index 00000000000..9d23a6588c2
--- /dev/null
+++ b/src/test/modules/injection_points/specs/repack_deadlock.spec
@@ -0,0 +1,83 @@
+# Test REPACK with a concurrent transaction that would cause a deadlock
+setup
+{
+	CREATE EXTENSION injection_points;
+
+	CREATE TABLE repack_deadlock(i int PRIMARY KEY, j int);
+	INSERT INTO repack_deadlock(i, j) VALUES (1, 1), (2, 2), (3, 3), (4, 4);
+
+	CREATE TABLE relfilenodes(node oid);
+
+	CREATE TABLE data_s1(i int, j int);
+	CREATE TABLE data_s2(i int, j int);
+}
+
+teardown
+{
+	DROP TABLE repack_deadlock;
+	DROP EXTENSION injection_points;
+
+	DROP TABLE relfilenodes;
+	DROP TABLE data_s1;
+	DROP TABLE data_s2;
+}
+
+session s1
+setup
+{
+	SELECT injection_points_set_local();
+	SELECT injection_points_attach('repack-concurrently-before-lock', 'wait');
+}
+# Perform the initial load and wait for s2 to do some data changes.
+step wait_before_lock
+{
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+}
+# Check the table from the perspective of s1.
+#
+# Besides the contents, we also check that relfilenode has changed.
+
+# Have each session write the contents into a table and use FULL JOIN to check
+# if the outputs are identical.
+step check1
+{
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+}
+teardown
+{
+	SELECT injection_points_detach('repack-concurrently-before-lock');
+}
+
+session s2
+# Change the existing data. UPDATE changes both key and non-key columns. Also
+# update one row twice to test whether tuple version generated by this session
+# can be found.
+step add_column
+{
+	alter table repack_deadlock add column noise text;
+}
+
+step wakeup_before_lock
+{
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+}
+
+# Test if data changes introduced while one session is performing REPACK
+# CONCURRENTLY find their way into the table.
+permutation
+	wait_before_lock
+	add_column
+	wakeup_before_lock
+	check1
-- 
2.47.3


--2fxmamo6mu2qbxgv
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v47-0009-Teach-snapshot-builder-to-skip-transactions-runn.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread

* [PATCH v48 6/7] Error out any process that would block at REPACK
@ 2026-03-25 19:35 Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Álvaro Herrera @ 2026-03-25 19:35 UTC (permalink / raw)

Any process waiting on REPACK to release its lock would actually cause
it to deadlock when it tries to upgrade its lock to AEL, losing all work
done to that point.  We avoid this by teaching the deadlock detector to
raise an error when this condition is detected.
---
 src/backend/commands/repack.c                 | 52 ++++++++----
 src/backend/storage/lmgr/deadlock.c           | 15 ++++
 src/include/storage/proc.h                    |  6 +-
 src/test/modules/injection_points/Makefile    |  1 +
 .../expected/repack_deadlock.out              | 63 ++++++++++++++
 src/test/modules/injection_points/meson.build |  1 +
 .../specs/repack_deadlock.spec                | 83 +++++++++++++++++++
 7 files changed, 202 insertions(+), 19 deletions(-)
 create mode 100644 src/test/modules/injection_points/expected/repack_deadlock.out
 create mode 100644 src/test/modules/injection_points/specs/repack_deadlock.spec

diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c
index a514053b777..781a471fc1d 100644
--- a/src/backend/commands/repack.c
+++ b/src/backend/commands/repack.c
@@ -276,6 +276,16 @@ ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
 	/* Determine the lock mode to use. */
 	lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
 
+	/*
+	 * If in concurrent mode, set the PROC_IN_CONCURRENT_REPACK flag.  This
+	 * makes the deadlock checker cause anyone that would conflict with us to
+	 * error out.  It's important to set this flag ahead of actually locking
+	 * the relation; it won't of course affect anyone until we do have a lock
+	 * that others can conflict with.
+	 */
+	if ((params.options & CLUOPT_CONCURRENT) != 0)
+		MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK;
+
 	/*
 	 * If a single relation is specified, process it and we're done ... unless
 	 * the relation is a partitioned table, in which case we fall through.
@@ -476,11 +486,8 @@ RepackLockLevel(bool concurrent)
  * If indexOid is InvalidOid, the table will be rewritten in physical order
  * instead of index order.
  *
- * Note that, in the concurrent case, the function releases the lock at some
- * point, in order to get AccessExclusiveLock for the final steps (i.e. to
- * swap the relation files). To make things simpler, the caller should expect
- * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
- * AccessExclusiveLock is kept till the end of the transaction.)
+ * On return, OldHeap is closed but locked with AccessExclusiveLock - the lock
+ * will be released at end of the transaction.
  *
  * 'cmd' indicates which command is being executed, to be used for error
  * messages.
@@ -512,10 +519,12 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
 		/*
 		 * Make sure we're not in a transaction block.
 		 *
-		 * The reason is that repack_setup_logical_decoding() could deadlock
-		 * if there's an XID already assigned.  It would be possible to run in
-		 * a transaction block if we had no XID, but this restriction is
-		 * simpler for users to understand and we don't lose anything.
+		 * The reason is that repack_setup_logical_decoding() could wait
+		 * indefinitely for our XID to complete. (The deadlock detector would
+		 * not recognize it because we'd be waiting for ourselves, i.e. no
+		 * real lock conflict.) It would be possible to run in a transaction
+		 * block if we had no XID, but this restriction is simpler for users
+		 * to understand and we don't lose anything.
 		 */
 		PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
 
@@ -998,10 +1007,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose,
 		 * Note that the worker has to wait for all transactions with XID
 		 * already assigned to finish. If some of those transactions is
 		 * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
-		 * table (e.g.  it runs CREATE INDEX), we can end up in a deadlock.
-		 * Not sure this risk is worth unlocking/locking the table (and its
-		 * clustering index) and checking again if it's still eligible for
-		 * REPACK CONCURRENTLY.
+		 * table (e.g. it runs CREATE INDEX), it should encounter ERROR in the
+		 * deadlock checking code.
 		 */
 		start_repack_decoding_worker(tableOid);
 
@@ -3090,7 +3097,16 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
 		LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
 
 	/*
-	 * Tuples and pages of the old heap will be gone, but the heap will stay.
+	 * Now that we have all access-exclusive locks on all relations, we no
+	 * longer want other processes to error out when trying to acquire a
+	 * conflicting lock.  Therefore, unset our flag.
+	 */
+	MyProc->statusFlags &= ~PROC_IN_CONCURRENT_REPACK;
+
+	/*
+	 * Tuples and pages of the old heap will be gone, but the heap itself will
+	 * stay.  In order for predicate locks to continue to work, convert them
+	 * to relation-level locks.  We do this both for table and indexes.
 	 */
 	TransferPredicateLocksToHeapRelation(OldHeap);
 	foreach_ptr(RelationData, index, indexrels)
@@ -3288,9 +3304,11 @@ start_repack_decoding_worker(Oid relid)
 
 	/*
 	 * The decoding setup must be done before the caller can have XID assigned
-	 * for any reason, otherwise the worker might end up in a deadlock,
-	 * waiting for the caller's transaction to end. Therefore wait here until
-	 * the worker indicates that it has the logical decoding initialized.
+	 * for any reason, otherwise the worker might end up waiting for the
+	 * caller's transaction to end. (Deadlock detector does not consider this
+	 * a conflict because the worker is in the same locking group as the
+	 * backend that launched it.) Therefore wait here until the worker
+	 * indicates that it has the logical decoding initialized.
 	 */
 	ConditionVariablePrepareToSleep(&shared->cv);
 	for (;;)
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index b8962d875b6..c20ac682b0d 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -620,6 +620,21 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 						proc->statusFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
+					/*
+					 * Similarly, if we note that we're blocked by some
+					 * process running REPACK (CONCURRENTLY), just fail.  That
+					 * process is going to upgrade its lock at some point, and
+					 * it would be inappropriate for any other process to
+					 * cause that to fail.
+					 */
+					if (checkProc == MyProc &&
+						proc->statusFlags & PROC_IN_CONCURRENT_REPACK)
+						ereport(ERROR,
+								errcode(ERRCODE_OBJECT_IN_USE),
+								errmsg("could not wait for concurrent REPACK"),
+								errdetail("Process %d waits for REPACK running on process %d",
+										  MyProc->pid, proc->pid));
+
 					/* We're done looking at this proclock */
 					break;
 				}
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1dad125706e..8ad9718f3d6 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -69,10 +69,12 @@ struct XidCache
 #define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be
 												 * included in vacuum horizons
 												 * in all databases */
+#define		PROC_IN_CONCURRENT_REPACK	0x40	/* REPACK (CONCURRENTLY) */
 
-/* flags reset at EOXact */
+/* flags reset at EOXact.  A bit of a misnomer ... */
 #define		PROC_VACUUM_STATE_MASK \
-	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)
+	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | \
+	 PROC_IN_CONCURRENT_REPACK)
 
 /*
  * Xmin-related flags. Make sure any flags that affect how the process' Xmin
diff --git a/src/test/modules/injection_points/Makefile b/src/test/modules/injection_points/Makefile
index 2cd7d87c533..f7663859fe2 100644
--- a/src/test/modules/injection_points/Makefile
+++ b/src/test/modules/injection_points/Makefile
@@ -15,6 +15,7 @@ REGRESS_OPTS = --dlpath=$(top_builddir)/src/test/regress
 ISOLATION = basic \
 	    inplace \
 	    repack \
+	    repack_deadlock \
 	    repack_toast \
 	    syscache-update-pruned \
 	    heap_lock_update
diff --git a/src/test/modules/injection_points/expected/repack_deadlock.out b/src/test/modules/injection_points/expected/repack_deadlock.out
new file mode 100644
index 00000000000..a86e4767536
--- /dev/null
+++ b/src/test/modules/injection_points/expected/repack_deadlock.out
@@ -0,0 +1,63 @@
+Parsed test spec with 2 sessions
+
+starting permutation: wait_before_lock add_column wakeup_before_lock check1
+injection_points_attach
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: 
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+ <waiting ...>
+step add_column: 
+	alter table repack_deadlock add column noise text;
+ <waiting ...>
+step add_column: <... completed>
+ERROR:  could not wait for concurrent REPACK
+step wakeup_before_lock: 
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+
+injection_points_wakeup
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: <... completed>
+step check1: 
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+
+count
+-----
+    1
+(1 row)
+
+i|j
+-+-
+1|1
+2|2
+3|3
+4|4
+(4 rows)
+
+count
+-----
+    4
+(1 row)
+
+injection_points_detach
+-----------------------
+                       
+(1 row)
+
diff --git a/src/test/modules/injection_points/meson.build b/src/test/modules/injection_points/meson.build
index a414abb924b..1cd88d6db65 100644
--- a/src/test/modules/injection_points/meson.build
+++ b/src/test/modules/injection_points/meson.build
@@ -46,6 +46,7 @@ tests += {
       'basic',
       'inplace',
       'repack',
+      'repack_deadlock',
       'repack_toast',
       'syscache-update-pruned',
       'heap_lock_update',
diff --git a/src/test/modules/injection_points/specs/repack_deadlock.spec b/src/test/modules/injection_points/specs/repack_deadlock.spec
new file mode 100644
index 00000000000..9d23a6588c2
--- /dev/null
+++ b/src/test/modules/injection_points/specs/repack_deadlock.spec
@@ -0,0 +1,83 @@
+# Test REPACK with a concurrent transaction that would cause a deadlock
+setup
+{
+	CREATE EXTENSION injection_points;
+
+	CREATE TABLE repack_deadlock(i int PRIMARY KEY, j int);
+	INSERT INTO repack_deadlock(i, j) VALUES (1, 1), (2, 2), (3, 3), (4, 4);
+
+	CREATE TABLE relfilenodes(node oid);
+
+	CREATE TABLE data_s1(i int, j int);
+	CREATE TABLE data_s2(i int, j int);
+}
+
+teardown
+{
+	DROP TABLE repack_deadlock;
+	DROP EXTENSION injection_points;
+
+	DROP TABLE relfilenodes;
+	DROP TABLE data_s1;
+	DROP TABLE data_s2;
+}
+
+session s1
+setup
+{
+	SELECT injection_points_set_local();
+	SELECT injection_points_attach('repack-concurrently-before-lock', 'wait');
+}
+# Perform the initial load and wait for s2 to do some data changes.
+step wait_before_lock
+{
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+}
+# Check the table from the perspective of s1.
+#
+# Besides the contents, we also check that relfilenode has changed.
+
+# Have each session write the contents into a table and use FULL JOIN to check
+# if the outputs are identical.
+step check1
+{
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+}
+teardown
+{
+	SELECT injection_points_detach('repack-concurrently-before-lock');
+}
+
+session s2
+# Change the existing data. UPDATE changes both key and non-key columns. Also
+# update one row twice to test whether tuple version generated by this session
+# can be found.
+step add_column
+{
+	alter table repack_deadlock add column noise text;
+}
+
+step wakeup_before_lock
+{
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+}
+
+# Test if data changes introduced while one session is performing REPACK
+# CONCURRENTLY find their way into the table.
+permutation
+	wait_before_lock
+	add_column
+	wakeup_before_lock
+	check1
-- 
2.47.3


--qfkt2ktdpcfeypib
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v48-0007-Teach-snapshot-builder-to-skip-transactions-runn.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread

* [PATCH v47 8/9] Error out any process that would block at REPACK
@ 2026-03-25 19:35 Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Álvaro Herrera @ 2026-03-25 19:35 UTC (permalink / raw)

Any process waiting on REPACK to release its lock would actually cause
it to deadlock when it tries to upgrade its lock to AEL, losing all work
done to that point.  We avoid this by teaching the deadlock detector to
raise an error when this condition is detected.
---
 src/backend/commands/repack.c                 | 52 ++++++++----
 src/backend/storage/lmgr/deadlock.c           | 15 ++++
 src/include/storage/proc.h                    |  6 +-
 src/test/modules/injection_points/Makefile    |  1 +
 .../expected/repack_deadlock.out              | 63 ++++++++++++++
 src/test/modules/injection_points/meson.build |  1 +
 .../specs/repack_deadlock.spec                | 83 +++++++++++++++++++
 7 files changed, 202 insertions(+), 19 deletions(-)
 create mode 100644 src/test/modules/injection_points/expected/repack_deadlock.out
 create mode 100644 src/test/modules/injection_points/specs/repack_deadlock.spec

diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c
index 9c807f75d71..3acb2e1d2ba 100644
--- a/src/backend/commands/repack.c
+++ b/src/backend/commands/repack.c
@@ -276,6 +276,16 @@ ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
 	/* Determine the lock mode to use. */
 	lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
 
+	/*
+	 * If in concurrent mode, set the PROC_IN_CONCURRENT_REPACK flag.  This
+	 * makes the deadlock checker cause anyone that would conflict with us
+	 * to error out.  It's important to set this flag ahead of actually locking
+	 * the relation; it won't of course affect anyone until we do have a lock
+	 * that others can conflict with.
+	 */
+	if ((params.options & CLUOPT_CONCURRENT) != 0)
+		MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK;
+
 	/*
 	 * If a single relation is specified, process it and we're done ... unless
 	 * the relation is a partitioned table, in which case we fall through.
@@ -476,11 +486,8 @@ RepackLockLevel(bool concurrent)
  * If indexOid is InvalidOid, the table will be rewritten in physical order
  * instead of index order.
  *
- * Note that, in the concurrent case, the function releases the lock at some
- * point, in order to get AccessExclusiveLock for the final steps (i.e. to
- * swap the relation files). To make things simpler, the caller should expect
- * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
- * AccessExclusiveLock is kept till the end of the transaction.)
+ * On return, OldHeap is closed but locked with AccessExclusiveLock - the lock
+ * will be released at end of the transaction.
  *
  * 'cmd' indicates which command is being executed, to be used for error
  * messages.
@@ -512,10 +519,12 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
 		/*
 		 * Make sure we're not in a transaction block.
 		 *
-		 * The reason is that repack_setup_logical_decoding() could deadlock
-		 * if there's an XID already assigned.  It would be possible to run in
-		 * a transaction block if we had no XID, but this restriction is
-		 * simpler for users to understand and we don't lose anything.
+		 * The reason is that repack_setup_logical_decoding() could wait
+		 * indefinitely for our XID to complete. (The deadlock detector would
+		 * not recognize it because we'd be waiting for ourselves, i.e. no
+		 * real lock conflict.) It would be possible to run in a transaction
+		 * block if we had no XID, but this restriction is simpler for users
+		 * to understand and we don't lose anything.
 		 */
 		PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
 
@@ -998,10 +1007,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose,
 		 * Note that the worker has to wait for all transactions with XID
 		 * already assigned to finish. If some of those transactions is
 		 * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
-		 * table (e.g.  it runs CREATE INDEX), we can end up in a deadlock.
-		 * Not sure this risk is worth unlocking/locking the table (and its
-		 * clustering index) and checking again if it's still eligible for
-		 * REPACK CONCURRENTLY.
+		 * table (e.g. it runs CREATE INDEX), it should encounter ERROR in the
+		 * deadlock checking code.
 		 */
 		start_repack_decoding_worker(tableOid);
 
@@ -3090,7 +3097,16 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
 		LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
 
 	/*
-	 * Tuples and pages of the old heap will be gone, but the heap will stay.
+	 * Now that we have all access-exclusive locks on all relations, we no
+	 * longer want other processes to error out when trying to acquire a
+	 * conflicting lock.  Therefore, unset our flag.
+	 */
+	MyProc->statusFlags &= ~PROC_IN_CONCURRENT_REPACK;
+
+	/*
+	 * Tuples and pages of the old heap will be gone, but the heap itself will
+	 * stay.  In order for predicate locks to continue to work, convert them
+	 * to relation-level locks.  We do this both for table and indexes.
 	 */
 	TransferPredicateLocksToHeapRelation(OldHeap);
 	foreach_ptr(RelationData, index, indexrels)
@@ -3287,9 +3303,11 @@ start_repack_decoding_worker(Oid relid)
 
 	/*
 	 * The decoding setup must be done before the caller can have XID assigned
-	 * for any reason, otherwise the worker might end up in a deadlock,
-	 * waiting for the caller's transaction to end. Therefore wait here until
-	 * the worker indicates that it has the logical decoding initialized.
+	 * for any reason, otherwise the worker might end up waiting for the
+	 * caller's transaction to end. (Deadlock detector does not consider this
+	 * a conflict because the worker is in the same locking group as the
+	 * backend that launched it.) Therefore wait here until the worker
+	 * indicates that it has the logical decoding initialized.
 	 */
 	ConditionVariablePrepareToSleep(&shared->cv);
 	for (;;)
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index b8962d875b6..3d3ec0da111 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -620,6 +620,21 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 						proc->statusFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
+					/*
+					 * Similarly, if we note that we're blocked by some process
+					 * running REPACK (CONCURRENTLY), just fail.  That process
+					 * is going to upgrade its lock at some point, and it would
+					 * be inappropriate for any other process to cause that
+					 * to fail.
+					 */
+					if (checkProc == MyProc &&
+						proc->statusFlags & PROC_IN_CONCURRENT_REPACK)
+						ereport(ERROR,
+								errcode(ERRCODE_OBJECT_IN_USE),
+								errmsg("could not wait for concurrent REPACK"),
+								errdetail("Process %d waits for REPACK running on process %d",
+									   MyProc->pid, proc->pid));
+
 					/* We're done looking at this proclock */
 					break;
 				}
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1dad125706e..8ad9718f3d6 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -69,10 +69,12 @@ struct XidCache
 #define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be
 												 * included in vacuum horizons
 												 * in all databases */
+#define		PROC_IN_CONCURRENT_REPACK	0x40	/* REPACK (CONCURRENTLY) */
 
-/* flags reset at EOXact */
+/* flags reset at EOXact.  A bit of a misnomer ... */
 #define		PROC_VACUUM_STATE_MASK \
-	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)
+	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | \
+	 PROC_IN_CONCURRENT_REPACK)
 
 /*
  * Xmin-related flags. Make sure any flags that affect how the process' Xmin
diff --git a/src/test/modules/injection_points/Makefile b/src/test/modules/injection_points/Makefile
index 2cd7d87c533..f7663859fe2 100644
--- a/src/test/modules/injection_points/Makefile
+++ b/src/test/modules/injection_points/Makefile
@@ -15,6 +15,7 @@ REGRESS_OPTS = --dlpath=$(top_builddir)/src/test/regress
 ISOLATION = basic \
 	    inplace \
 	    repack \
+	    repack_deadlock \
 	    repack_toast \
 	    syscache-update-pruned \
 	    heap_lock_update
diff --git a/src/test/modules/injection_points/expected/repack_deadlock.out b/src/test/modules/injection_points/expected/repack_deadlock.out
new file mode 100644
index 00000000000..a86e4767536
--- /dev/null
+++ b/src/test/modules/injection_points/expected/repack_deadlock.out
@@ -0,0 +1,63 @@
+Parsed test spec with 2 sessions
+
+starting permutation: wait_before_lock add_column wakeup_before_lock check1
+injection_points_attach
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: 
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+ <waiting ...>
+step add_column: 
+	alter table repack_deadlock add column noise text;
+ <waiting ...>
+step add_column: <... completed>
+ERROR:  could not wait for concurrent REPACK
+step wakeup_before_lock: 
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+
+injection_points_wakeup
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: <... completed>
+step check1: 
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+
+count
+-----
+    1
+(1 row)
+
+i|j
+-+-
+1|1
+2|2
+3|3
+4|4
+(4 rows)
+
+count
+-----
+    4
+(1 row)
+
+injection_points_detach
+-----------------------
+                       
+(1 row)
+
diff --git a/src/test/modules/injection_points/meson.build b/src/test/modules/injection_points/meson.build
index a414abb924b..1cd88d6db65 100644
--- a/src/test/modules/injection_points/meson.build
+++ b/src/test/modules/injection_points/meson.build
@@ -46,6 +46,7 @@ tests += {
       'basic',
       'inplace',
       'repack',
+      'repack_deadlock',
       'repack_toast',
       'syscache-update-pruned',
       'heap_lock_update',
diff --git a/src/test/modules/injection_points/specs/repack_deadlock.spec b/src/test/modules/injection_points/specs/repack_deadlock.spec
new file mode 100644
index 00000000000..9d23a6588c2
--- /dev/null
+++ b/src/test/modules/injection_points/specs/repack_deadlock.spec
@@ -0,0 +1,83 @@
+# Test REPACK with a concurrent transaction that would cause a deadlock
+setup
+{
+	CREATE EXTENSION injection_points;
+
+	CREATE TABLE repack_deadlock(i int PRIMARY KEY, j int);
+	INSERT INTO repack_deadlock(i, j) VALUES (1, 1), (2, 2), (3, 3), (4, 4);
+
+	CREATE TABLE relfilenodes(node oid);
+
+	CREATE TABLE data_s1(i int, j int);
+	CREATE TABLE data_s2(i int, j int);
+}
+
+teardown
+{
+	DROP TABLE repack_deadlock;
+	DROP EXTENSION injection_points;
+
+	DROP TABLE relfilenodes;
+	DROP TABLE data_s1;
+	DROP TABLE data_s2;
+}
+
+session s1
+setup
+{
+	SELECT injection_points_set_local();
+	SELECT injection_points_attach('repack-concurrently-before-lock', 'wait');
+}
+# Perform the initial load and wait for s2 to do some data changes.
+step wait_before_lock
+{
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+}
+# Check the table from the perspective of s1.
+#
+# Besides the contents, we also check that relfilenode has changed.
+
+# Have each session write the contents into a table and use FULL JOIN to check
+# if the outputs are identical.
+step check1
+{
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+}
+teardown
+{
+	SELECT injection_points_detach('repack-concurrently-before-lock');
+}
+
+session s2
+# Change the existing data. UPDATE changes both key and non-key columns. Also
+# update one row twice to test whether tuple version generated by this session
+# can be found.
+step add_column
+{
+	alter table repack_deadlock add column noise text;
+}
+
+step wakeup_before_lock
+{
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+}
+
+# Test if data changes introduced while one session is performing REPACK
+# CONCURRENTLY find their way into the table.
+permutation
+	wait_before_lock
+	add_column
+	wakeup_before_lock
+	check1
-- 
2.47.3


--2fxmamo6mu2qbxgv
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v47-0009-Teach-snapshot-builder-to-skip-transactions-runn.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread

* [PATCH v45 7/8] Error out any process that would block at REPACK
@ 2026-03-25 19:35 Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Álvaro Herrera @ 2026-03-25 19:35 UTC (permalink / raw)

Any process waiting on REPACK to release its lock would actually cause
it to deadlock when it tries to upgrade its lock to AEL, losing all work
done to that point.  We avoid this by teaching the deadlock detector to
raise an error when this condition is detected.
---
 src/backend/commands/repack.c                 | 53 ++++++++----
 src/backend/storage/lmgr/deadlock.c           | 15 ++++
 src/include/storage/proc.h                    |  6 +-
 src/test/modules/injection_points/Makefile    |  1 +
 .../expected/repack_deadlock.out              | 63 ++++++++++++++
 src/test/modules/injection_points/meson.build |  1 +
 .../specs/repack_deadlock.spec                | 83 +++++++++++++++++++
 7 files changed, 202 insertions(+), 20 deletions(-)
 create mode 100644 src/test/modules/injection_points/expected/repack_deadlock.out
 create mode 100644 src/test/modules/injection_points/specs/repack_deadlock.spec

diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c
index d25ab43da7b..7291b4180ab 100644
--- a/src/backend/commands/repack.c
+++ b/src/backend/commands/repack.c
@@ -276,6 +276,16 @@ ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
 	/* Determine the lock mode to use. */
 	lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
 
+	/*
+	 * If in concurrent mode, set the PROC_IN_CONCURRENT_REPACK flag.  This
+	 * makes the deadlock checker cause anyone that would conflict with us
+	 * to error out.  It's important to set this flag ahead of actually locking
+	 * the relation; it won't of course affect anyone until we do have a lock
+	 * that others can conflict with.
+	 */
+	if ((params.options & CLUOPT_CONCURRENT) != 0)
+		MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK;
+
 	/*
 	 * If a single relation is specified, process it and we're done ... unless
 	 * the relation is a partitioned table, in which case we fall through.
@@ -476,11 +486,8 @@ RepackLockLevel(bool concurrent)
  * If indexOid is InvalidOid, the table will be rewritten in physical order
  * instead of index order.
  *
- * Note that, in the concurrent case, the function releases the lock at some
- * point, in order to get AccessExclusiveLock for the final steps (i.e. to
- * swap the relation files). To make things simpler, the caller should expect
- * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
- * AccessExclusiveLock is kept till the end of the transaction.)
+ * On return, OldHeap is closed but locked with AccessExclusiveLock - the lock
+ * will be released at end of the transaction.
  *
  * 'cmd' indicates which command is being executed, to be used for error
  * messages.
@@ -512,10 +519,12 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
 		/*
 		 * Make sure we're not in a transaction block.
 		 *
-		 * The reason is that repack_setup_logical_decoding() could deadlock
-		 * if there's an XID already assigned.  It would be possible to run in
-		 * a transaction block if we had no XID, but this restriction is
-		 * simpler for users to understand and we don't lose anything.
+		 * The reason is that repack_setup_logical_decoding() could wait
+		 * indefinitely for our XID to complete. (The deadlock detector would
+		 * not recognize it because we'd be waiting for ourselves, i.e. no
+		 * real lock conflict.) It would be possible to run in a transaction
+		 * block if we had no XID, but this restriction is simpler for users
+		 * to understand and we don't lose anything.
 		 */
 		PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
 
@@ -998,10 +1007,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose,
 		 * Note that the worker has to wait for all transactions with XID
 		 * already assigned to finish. If some of those transactions is
 		 * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
-		 * table (e.g.  it runs CREATE INDEX), we can end up in a deadlock.
-		 * Not sure this risk is worth unlocking/locking the table (and its
-		 * clustering index) and checking again if it's still eligible for
-		 * REPACK CONCURRENTLY.
+		 * table (e.g. it runs CREATE INDEX), it should encounter ERROR in the
+		 * deadlock checking code.
 		 */
 		start_repack_decoding_worker(tableOid);
 
@@ -3119,10 +3126,18 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
 		LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
 
 	/*
-	 * Tuples and pages of the old heap will be gone, but the heap will stay.
+	 * Now that we have all access-exclusive locks on all relations, we no
+	 * longer want other processes to error out when trying to acquire a
+	 * conflicting lock.  Therefore, unset our flag.
+	 */
+	MyProc->statusFlags &= ~PROC_IN_CONCURRENT_REPACK;
+
+	/*
+	 * Tuples and pages of the old heap will be gone, but the heap itself will
+	 * stay.  In order for predicate locks to continue to work, convert them
+	 * to relation-level locks.  We do this both for table and indexes.
 	 */
 	TransferPredicateLocksToHeapRelation(OldHeap);
-	/* The same for indexes. */
 	for (int i = 0; i < nind; i++)
 	{
 		Relation	index = ind_refs[i];
@@ -3342,9 +3357,11 @@ start_repack_decoding_worker(Oid relid)
 
 	/*
 	 * The decoding setup must be done before the caller can have XID assigned
-	 * for any reason, otherwise the worker might end up in a deadlock,
-	 * waiting for the caller's transaction to end. Therefore wait here until
-	 * the worker indicates that it has the logical decoding initialized.
+	 * for any reason, otherwise the worker might end up waiting for the
+	 * caller's transaction to end. (Deadlock detector does not consider this
+	 * a conflict because the worker is in the same locking group as the
+	 * backend that launched it.) Therefore wait here until the worker
+	 * indicates that it has the logical decoding initialized.
 	 */
 	ConditionVariablePrepareToSleep(&shared->cv);
 	for (;;)
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index b8962d875b6..3d3ec0da111 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -620,6 +620,21 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 						proc->statusFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
+					/*
+					 * Similarly, if we note that we're blocked by some process
+					 * running REPACK (CONCURRENTLY), just fail.  That process
+					 * is going to upgrade its lock at some point, and it would
+					 * be inappropriate for any other process to cause that
+					 * to fail.
+					 */
+					if (checkProc == MyProc &&
+						proc->statusFlags & PROC_IN_CONCURRENT_REPACK)
+						ereport(ERROR,
+								errcode(ERRCODE_OBJECT_IN_USE),
+								errmsg("could not wait for concurrent REPACK"),
+								errdetail("Process %d waits for REPACK running on process %d",
+									   MyProc->pid, proc->pid));
+
 					/* We're done looking at this proclock */
 					break;
 				}
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1dad125706e..8ad9718f3d6 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -69,10 +69,12 @@ struct XidCache
 #define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be
 												 * included in vacuum horizons
 												 * in all databases */
+#define		PROC_IN_CONCURRENT_REPACK	0x40	/* REPACK (CONCURRENTLY) */
 
-/* flags reset at EOXact */
+/* flags reset at EOXact.  A bit of a misnomer ... */
 #define		PROC_VACUUM_STATE_MASK \
-	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)
+	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | \
+	 PROC_IN_CONCURRENT_REPACK)
 
 /*
  * Xmin-related flags. Make sure any flags that affect how the process' Xmin
diff --git a/src/test/modules/injection_points/Makefile b/src/test/modules/injection_points/Makefile
index 2cd7d87c533..f7663859fe2 100644
--- a/src/test/modules/injection_points/Makefile
+++ b/src/test/modules/injection_points/Makefile
@@ -15,6 +15,7 @@ REGRESS_OPTS = --dlpath=$(top_builddir)/src/test/regress
 ISOLATION = basic \
 	    inplace \
 	    repack \
+	    repack_deadlock \
 	    repack_toast \
 	    syscache-update-pruned \
 	    heap_lock_update
diff --git a/src/test/modules/injection_points/expected/repack_deadlock.out b/src/test/modules/injection_points/expected/repack_deadlock.out
new file mode 100644
index 00000000000..a86e4767536
--- /dev/null
+++ b/src/test/modules/injection_points/expected/repack_deadlock.out
@@ -0,0 +1,63 @@
+Parsed test spec with 2 sessions
+
+starting permutation: wait_before_lock add_column wakeup_before_lock check1
+injection_points_attach
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: 
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+ <waiting ...>
+step add_column: 
+	alter table repack_deadlock add column noise text;
+ <waiting ...>
+step add_column: <... completed>
+ERROR:  could not wait for concurrent REPACK
+step wakeup_before_lock: 
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+
+injection_points_wakeup
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: <... completed>
+step check1: 
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+
+count
+-----
+    1
+(1 row)
+
+i|j
+-+-
+1|1
+2|2
+3|3
+4|4
+(4 rows)
+
+count
+-----
+    4
+(1 row)
+
+injection_points_detach
+-----------------------
+                       
+(1 row)
+
diff --git a/src/test/modules/injection_points/meson.build b/src/test/modules/injection_points/meson.build
index a414abb924b..1cd88d6db65 100644
--- a/src/test/modules/injection_points/meson.build
+++ b/src/test/modules/injection_points/meson.build
@@ -46,6 +46,7 @@ tests += {
       'basic',
       'inplace',
       'repack',
+      'repack_deadlock',
       'repack_toast',
       'syscache-update-pruned',
       'heap_lock_update',
diff --git a/src/test/modules/injection_points/specs/repack_deadlock.spec b/src/test/modules/injection_points/specs/repack_deadlock.spec
new file mode 100644
index 00000000000..9d23a6588c2
--- /dev/null
+++ b/src/test/modules/injection_points/specs/repack_deadlock.spec
@@ -0,0 +1,83 @@
+# Test REPACK with a concurrent transaction that would cause a deadlock
+setup
+{
+	CREATE EXTENSION injection_points;
+
+	CREATE TABLE repack_deadlock(i int PRIMARY KEY, j int);
+	INSERT INTO repack_deadlock(i, j) VALUES (1, 1), (2, 2), (3, 3), (4, 4);
+
+	CREATE TABLE relfilenodes(node oid);
+
+	CREATE TABLE data_s1(i int, j int);
+	CREATE TABLE data_s2(i int, j int);
+}
+
+teardown
+{
+	DROP TABLE repack_deadlock;
+	DROP EXTENSION injection_points;
+
+	DROP TABLE relfilenodes;
+	DROP TABLE data_s1;
+	DROP TABLE data_s2;
+}
+
+session s1
+setup
+{
+	SELECT injection_points_set_local();
+	SELECT injection_points_attach('repack-concurrently-before-lock', 'wait');
+}
+# Perform the initial load and wait for s2 to do some data changes.
+step wait_before_lock
+{
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+}
+# Check the table from the perspective of s1.
+#
+# Besides the contents, we also check that relfilenode has changed.
+
+# Have each session write the contents into a table and use FULL JOIN to check
+# if the outputs are identical.
+step check1
+{
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+}
+teardown
+{
+	SELECT injection_points_detach('repack-concurrently-before-lock');
+}
+
+session s2
+# Change the existing data. UPDATE changes both key and non-key columns. Also
+# update one row twice to test whether tuple version generated by this session
+# can be found.
+step add_column
+{
+	alter table repack_deadlock add column noise text;
+}
+
+step wakeup_before_lock
+{
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+}
+
+# Test if data changes introduced while one session is performing REPACK
+# CONCURRENTLY find their way into the table.
+permutation
+	wait_before_lock
+	add_column
+	wakeup_before_lock
+	check1
-- 
2.47.3


--cldir6umqisr673d
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v45-0008-Teach-snapshot-builder-to-skip-transactions-runn.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread

* [PATCH v48 6/7] Error out any process that would block at REPACK
@ 2026-03-25 19:35 Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Álvaro Herrera @ 2026-03-25 19:35 UTC (permalink / raw)

Any process waiting on REPACK to release its lock would actually cause
it to deadlock when it tries to upgrade its lock to AEL, losing all work
done to that point.  We avoid this by teaching the deadlock detector to
raise an error when this condition is detected.
---
 src/backend/commands/repack.c                 | 52 ++++++++----
 src/backend/storage/lmgr/deadlock.c           | 15 ++++
 src/include/storage/proc.h                    |  6 +-
 src/test/modules/injection_points/Makefile    |  1 +
 .../expected/repack_deadlock.out              | 63 ++++++++++++++
 src/test/modules/injection_points/meson.build |  1 +
 .../specs/repack_deadlock.spec                | 83 +++++++++++++++++++
 7 files changed, 202 insertions(+), 19 deletions(-)
 create mode 100644 src/test/modules/injection_points/expected/repack_deadlock.out
 create mode 100644 src/test/modules/injection_points/specs/repack_deadlock.spec

diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c
index a514053b777..781a471fc1d 100644
--- a/src/backend/commands/repack.c
+++ b/src/backend/commands/repack.c
@@ -276,6 +276,16 @@ ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
 	/* Determine the lock mode to use. */
 	lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
 
+	/*
+	 * If in concurrent mode, set the PROC_IN_CONCURRENT_REPACK flag.  This
+	 * makes the deadlock checker cause anyone that would conflict with us to
+	 * error out.  It's important to set this flag ahead of actually locking
+	 * the relation; it won't of course affect anyone until we do have a lock
+	 * that others can conflict with.
+	 */
+	if ((params.options & CLUOPT_CONCURRENT) != 0)
+		MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK;
+
 	/*
 	 * If a single relation is specified, process it and we're done ... unless
 	 * the relation is a partitioned table, in which case we fall through.
@@ -476,11 +486,8 @@ RepackLockLevel(bool concurrent)
  * If indexOid is InvalidOid, the table will be rewritten in physical order
  * instead of index order.
  *
- * Note that, in the concurrent case, the function releases the lock at some
- * point, in order to get AccessExclusiveLock for the final steps (i.e. to
- * swap the relation files). To make things simpler, the caller should expect
- * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
- * AccessExclusiveLock is kept till the end of the transaction.)
+ * On return, OldHeap is closed but locked with AccessExclusiveLock - the lock
+ * will be released at end of the transaction.
  *
  * 'cmd' indicates which command is being executed, to be used for error
  * messages.
@@ -512,10 +519,12 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
 		/*
 		 * Make sure we're not in a transaction block.
 		 *
-		 * The reason is that repack_setup_logical_decoding() could deadlock
-		 * if there's an XID already assigned.  It would be possible to run in
-		 * a transaction block if we had no XID, but this restriction is
-		 * simpler for users to understand and we don't lose anything.
+		 * The reason is that repack_setup_logical_decoding() could wait
+		 * indefinitely for our XID to complete. (The deadlock detector would
+		 * not recognize it because we'd be waiting for ourselves, i.e. no
+		 * real lock conflict.) It would be possible to run in a transaction
+		 * block if we had no XID, but this restriction is simpler for users
+		 * to understand and we don't lose anything.
 		 */
 		PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
 
@@ -998,10 +1007,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose,
 		 * Note that the worker has to wait for all transactions with XID
 		 * already assigned to finish. If some of those transactions is
 		 * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
-		 * table (e.g.  it runs CREATE INDEX), we can end up in a deadlock.
-		 * Not sure this risk is worth unlocking/locking the table (and its
-		 * clustering index) and checking again if it's still eligible for
-		 * REPACK CONCURRENTLY.
+		 * table (e.g. it runs CREATE INDEX), it should encounter ERROR in the
+		 * deadlock checking code.
 		 */
 		start_repack_decoding_worker(tableOid);
 
@@ -3090,7 +3097,16 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
 		LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
 
 	/*
-	 * Tuples and pages of the old heap will be gone, but the heap will stay.
+	 * Now that we have all access-exclusive locks on all relations, we no
+	 * longer want other processes to error out when trying to acquire a
+	 * conflicting lock.  Therefore, unset our flag.
+	 */
+	MyProc->statusFlags &= ~PROC_IN_CONCURRENT_REPACK;
+
+	/*
+	 * Tuples and pages of the old heap will be gone, but the heap itself will
+	 * stay.  In order for predicate locks to continue to work, convert them
+	 * to relation-level locks.  We do this both for table and indexes.
 	 */
 	TransferPredicateLocksToHeapRelation(OldHeap);
 	foreach_ptr(RelationData, index, indexrels)
@@ -3288,9 +3304,11 @@ start_repack_decoding_worker(Oid relid)
 
 	/*
 	 * The decoding setup must be done before the caller can have XID assigned
-	 * for any reason, otherwise the worker might end up in a deadlock,
-	 * waiting for the caller's transaction to end. Therefore wait here until
-	 * the worker indicates that it has the logical decoding initialized.
+	 * for any reason, otherwise the worker might end up waiting for the
+	 * caller's transaction to end. (Deadlock detector does not consider this
+	 * a conflict because the worker is in the same locking group as the
+	 * backend that launched it.) Therefore wait here until the worker
+	 * indicates that it has the logical decoding initialized.
 	 */
 	ConditionVariablePrepareToSleep(&shared->cv);
 	for (;;)
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index b8962d875b6..c20ac682b0d 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -620,6 +620,21 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 						proc->statusFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
+					/*
+					 * Similarly, if we note that we're blocked by some
+					 * process running REPACK (CONCURRENTLY), just fail.  That
+					 * process is going to upgrade its lock at some point, and
+					 * it would be inappropriate for any other process to
+					 * cause that to fail.
+					 */
+					if (checkProc == MyProc &&
+						proc->statusFlags & PROC_IN_CONCURRENT_REPACK)
+						ereport(ERROR,
+								errcode(ERRCODE_OBJECT_IN_USE),
+								errmsg("could not wait for concurrent REPACK"),
+								errdetail("Process %d waits for REPACK running on process %d",
+										  MyProc->pid, proc->pid));
+
 					/* We're done looking at this proclock */
 					break;
 				}
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1dad125706e..8ad9718f3d6 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -69,10 +69,12 @@ struct XidCache
 #define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be
 												 * included in vacuum horizons
 												 * in all databases */
+#define		PROC_IN_CONCURRENT_REPACK	0x40	/* REPACK (CONCURRENTLY) */
 
-/* flags reset at EOXact */
+/* flags reset at EOXact.  A bit of a misnomer ... */
 #define		PROC_VACUUM_STATE_MASK \
-	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)
+	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | \
+	 PROC_IN_CONCURRENT_REPACK)
 
 /*
  * Xmin-related flags. Make sure any flags that affect how the process' Xmin
diff --git a/src/test/modules/injection_points/Makefile b/src/test/modules/injection_points/Makefile
index 2cd7d87c533..f7663859fe2 100644
--- a/src/test/modules/injection_points/Makefile
+++ b/src/test/modules/injection_points/Makefile
@@ -15,6 +15,7 @@ REGRESS_OPTS = --dlpath=$(top_builddir)/src/test/regress
 ISOLATION = basic \
 	    inplace \
 	    repack \
+	    repack_deadlock \
 	    repack_toast \
 	    syscache-update-pruned \
 	    heap_lock_update
diff --git a/src/test/modules/injection_points/expected/repack_deadlock.out b/src/test/modules/injection_points/expected/repack_deadlock.out
new file mode 100644
index 00000000000..a86e4767536
--- /dev/null
+++ b/src/test/modules/injection_points/expected/repack_deadlock.out
@@ -0,0 +1,63 @@
+Parsed test spec with 2 sessions
+
+starting permutation: wait_before_lock add_column wakeup_before_lock check1
+injection_points_attach
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: 
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+ <waiting ...>
+step add_column: 
+	alter table repack_deadlock add column noise text;
+ <waiting ...>
+step add_column: <... completed>
+ERROR:  could not wait for concurrent REPACK
+step wakeup_before_lock: 
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+
+injection_points_wakeup
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: <... completed>
+step check1: 
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+
+count
+-----
+    1
+(1 row)
+
+i|j
+-+-
+1|1
+2|2
+3|3
+4|4
+(4 rows)
+
+count
+-----
+    4
+(1 row)
+
+injection_points_detach
+-----------------------
+                       
+(1 row)
+
diff --git a/src/test/modules/injection_points/meson.build b/src/test/modules/injection_points/meson.build
index a414abb924b..1cd88d6db65 100644
--- a/src/test/modules/injection_points/meson.build
+++ b/src/test/modules/injection_points/meson.build
@@ -46,6 +46,7 @@ tests += {
       'basic',
       'inplace',
       'repack',
+      'repack_deadlock',
       'repack_toast',
       'syscache-update-pruned',
       'heap_lock_update',
diff --git a/src/test/modules/injection_points/specs/repack_deadlock.spec b/src/test/modules/injection_points/specs/repack_deadlock.spec
new file mode 100644
index 00000000000..9d23a6588c2
--- /dev/null
+++ b/src/test/modules/injection_points/specs/repack_deadlock.spec
@@ -0,0 +1,83 @@
+# Test REPACK with a concurrent transaction that would cause a deadlock
+setup
+{
+	CREATE EXTENSION injection_points;
+
+	CREATE TABLE repack_deadlock(i int PRIMARY KEY, j int);
+	INSERT INTO repack_deadlock(i, j) VALUES (1, 1), (2, 2), (3, 3), (4, 4);
+
+	CREATE TABLE relfilenodes(node oid);
+
+	CREATE TABLE data_s1(i int, j int);
+	CREATE TABLE data_s2(i int, j int);
+}
+
+teardown
+{
+	DROP TABLE repack_deadlock;
+	DROP EXTENSION injection_points;
+
+	DROP TABLE relfilenodes;
+	DROP TABLE data_s1;
+	DROP TABLE data_s2;
+}
+
+session s1
+setup
+{
+	SELECT injection_points_set_local();
+	SELECT injection_points_attach('repack-concurrently-before-lock', 'wait');
+}
+# Perform the initial load and wait for s2 to do some data changes.
+step wait_before_lock
+{
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+}
+# Check the table from the perspective of s1.
+#
+# Besides the contents, we also check that relfilenode has changed.
+
+# Have each session write the contents into a table and use FULL JOIN to check
+# if the outputs are identical.
+step check1
+{
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+}
+teardown
+{
+	SELECT injection_points_detach('repack-concurrently-before-lock');
+}
+
+session s2
+# Change the existing data. UPDATE changes both key and non-key columns. Also
+# update one row twice to test whether tuple version generated by this session
+# can be found.
+step add_column
+{
+	alter table repack_deadlock add column noise text;
+}
+
+step wakeup_before_lock
+{
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+}
+
+# Test if data changes introduced while one session is performing REPACK
+# CONCURRENTLY find their way into the table.
+permutation
+	wait_before_lock
+	add_column
+	wakeup_before_lock
+	check1
-- 
2.47.3


--qfkt2ktdpcfeypib
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v48-0007-Teach-snapshot-builder-to-skip-transactions-runn.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread

* [PATCH v48 6/7] Error out any process that would block at REPACK
@ 2026-03-25 19:35 Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Álvaro Herrera @ 2026-03-25 19:35 UTC (permalink / raw)

Any process waiting on REPACK to release its lock would actually cause
it to deadlock when it tries to upgrade its lock to AEL, losing all work
done to that point.  We avoid this by teaching the deadlock detector to
raise an error when this condition is detected.
---
 src/backend/commands/repack.c                 | 52 ++++++++----
 src/backend/storage/lmgr/deadlock.c           | 15 ++++
 src/include/storage/proc.h                    |  6 +-
 src/test/modules/injection_points/Makefile    |  1 +
 .../expected/repack_deadlock.out              | 63 ++++++++++++++
 src/test/modules/injection_points/meson.build |  1 +
 .../specs/repack_deadlock.spec                | 83 +++++++++++++++++++
 7 files changed, 202 insertions(+), 19 deletions(-)
 create mode 100644 src/test/modules/injection_points/expected/repack_deadlock.out
 create mode 100644 src/test/modules/injection_points/specs/repack_deadlock.spec

diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c
index a514053b777..781a471fc1d 100644
--- a/src/backend/commands/repack.c
+++ b/src/backend/commands/repack.c
@@ -276,6 +276,16 @@ ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
 	/* Determine the lock mode to use. */
 	lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
 
+	/*
+	 * If in concurrent mode, set the PROC_IN_CONCURRENT_REPACK flag.  This
+	 * makes the deadlock checker cause anyone that would conflict with us to
+	 * error out.  It's important to set this flag ahead of actually locking
+	 * the relation; it won't of course affect anyone until we do have a lock
+	 * that others can conflict with.
+	 */
+	if ((params.options & CLUOPT_CONCURRENT) != 0)
+		MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK;
+
 	/*
 	 * If a single relation is specified, process it and we're done ... unless
 	 * the relation is a partitioned table, in which case we fall through.
@@ -476,11 +486,8 @@ RepackLockLevel(bool concurrent)
  * If indexOid is InvalidOid, the table will be rewritten in physical order
  * instead of index order.
  *
- * Note that, in the concurrent case, the function releases the lock at some
- * point, in order to get AccessExclusiveLock for the final steps (i.e. to
- * swap the relation files). To make things simpler, the caller should expect
- * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
- * AccessExclusiveLock is kept till the end of the transaction.)
+ * On return, OldHeap is closed but locked with AccessExclusiveLock - the lock
+ * will be released at end of the transaction.
  *
  * 'cmd' indicates which command is being executed, to be used for error
  * messages.
@@ -512,10 +519,12 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
 		/*
 		 * Make sure we're not in a transaction block.
 		 *
-		 * The reason is that repack_setup_logical_decoding() could deadlock
-		 * if there's an XID already assigned.  It would be possible to run in
-		 * a transaction block if we had no XID, but this restriction is
-		 * simpler for users to understand and we don't lose anything.
+		 * The reason is that repack_setup_logical_decoding() could wait
+		 * indefinitely for our XID to complete. (The deadlock detector would
+		 * not recognize it because we'd be waiting for ourselves, i.e. no
+		 * real lock conflict.) It would be possible to run in a transaction
+		 * block if we had no XID, but this restriction is simpler for users
+		 * to understand and we don't lose anything.
 		 */
 		PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
 
@@ -998,10 +1007,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose,
 		 * Note that the worker has to wait for all transactions with XID
 		 * already assigned to finish. If some of those transactions is
 		 * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
-		 * table (e.g.  it runs CREATE INDEX), we can end up in a deadlock.
-		 * Not sure this risk is worth unlocking/locking the table (and its
-		 * clustering index) and checking again if it's still eligible for
-		 * REPACK CONCURRENTLY.
+		 * table (e.g. it runs CREATE INDEX), it should encounter ERROR in the
+		 * deadlock checking code.
 		 */
 		start_repack_decoding_worker(tableOid);
 
@@ -3090,7 +3097,16 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
 		LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
 
 	/*
-	 * Tuples and pages of the old heap will be gone, but the heap will stay.
+	 * Now that we have all access-exclusive locks on all relations, we no
+	 * longer want other processes to error out when trying to acquire a
+	 * conflicting lock.  Therefore, unset our flag.
+	 */
+	MyProc->statusFlags &= ~PROC_IN_CONCURRENT_REPACK;
+
+	/*
+	 * Tuples and pages of the old heap will be gone, but the heap itself will
+	 * stay.  In order for predicate locks to continue to work, convert them
+	 * to relation-level locks.  We do this both for table and indexes.
 	 */
 	TransferPredicateLocksToHeapRelation(OldHeap);
 	foreach_ptr(RelationData, index, indexrels)
@@ -3288,9 +3304,11 @@ start_repack_decoding_worker(Oid relid)
 
 	/*
 	 * The decoding setup must be done before the caller can have XID assigned
-	 * for any reason, otherwise the worker might end up in a deadlock,
-	 * waiting for the caller's transaction to end. Therefore wait here until
-	 * the worker indicates that it has the logical decoding initialized.
+	 * for any reason, otherwise the worker might end up waiting for the
+	 * caller's transaction to end. (Deadlock detector does not consider this
+	 * a conflict because the worker is in the same locking group as the
+	 * backend that launched it.) Therefore wait here until the worker
+	 * indicates that it has the logical decoding initialized.
 	 */
 	ConditionVariablePrepareToSleep(&shared->cv);
 	for (;;)
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index b8962d875b6..c20ac682b0d 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -620,6 +620,21 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 						proc->statusFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
+					/*
+					 * Similarly, if we note that we're blocked by some
+					 * process running REPACK (CONCURRENTLY), just fail.  That
+					 * process is going to upgrade its lock at some point, and
+					 * it would be inappropriate for any other process to
+					 * cause that to fail.
+					 */
+					if (checkProc == MyProc &&
+						proc->statusFlags & PROC_IN_CONCURRENT_REPACK)
+						ereport(ERROR,
+								errcode(ERRCODE_OBJECT_IN_USE),
+								errmsg("could not wait for concurrent REPACK"),
+								errdetail("Process %d waits for REPACK running on process %d",
+										  MyProc->pid, proc->pid));
+
 					/* We're done looking at this proclock */
 					break;
 				}
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1dad125706e..8ad9718f3d6 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -69,10 +69,12 @@ struct XidCache
 #define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be
 												 * included in vacuum horizons
 												 * in all databases */
+#define		PROC_IN_CONCURRENT_REPACK	0x40	/* REPACK (CONCURRENTLY) */
 
-/* flags reset at EOXact */
+/* flags reset at EOXact.  A bit of a misnomer ... */
 #define		PROC_VACUUM_STATE_MASK \
-	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)
+	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | \
+	 PROC_IN_CONCURRENT_REPACK)
 
 /*
  * Xmin-related flags. Make sure any flags that affect how the process' Xmin
diff --git a/src/test/modules/injection_points/Makefile b/src/test/modules/injection_points/Makefile
index 2cd7d87c533..f7663859fe2 100644
--- a/src/test/modules/injection_points/Makefile
+++ b/src/test/modules/injection_points/Makefile
@@ -15,6 +15,7 @@ REGRESS_OPTS = --dlpath=$(top_builddir)/src/test/regress
 ISOLATION = basic \
 	    inplace \
 	    repack \
+	    repack_deadlock \
 	    repack_toast \
 	    syscache-update-pruned \
 	    heap_lock_update
diff --git a/src/test/modules/injection_points/expected/repack_deadlock.out b/src/test/modules/injection_points/expected/repack_deadlock.out
new file mode 100644
index 00000000000..a86e4767536
--- /dev/null
+++ b/src/test/modules/injection_points/expected/repack_deadlock.out
@@ -0,0 +1,63 @@
+Parsed test spec with 2 sessions
+
+starting permutation: wait_before_lock add_column wakeup_before_lock check1
+injection_points_attach
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: 
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+ <waiting ...>
+step add_column: 
+	alter table repack_deadlock add column noise text;
+ <waiting ...>
+step add_column: <... completed>
+ERROR:  could not wait for concurrent REPACK
+step wakeup_before_lock: 
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+
+injection_points_wakeup
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: <... completed>
+step check1: 
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+
+count
+-----
+    1
+(1 row)
+
+i|j
+-+-
+1|1
+2|2
+3|3
+4|4
+(4 rows)
+
+count
+-----
+    4
+(1 row)
+
+injection_points_detach
+-----------------------
+                       
+(1 row)
+
diff --git a/src/test/modules/injection_points/meson.build b/src/test/modules/injection_points/meson.build
index a414abb924b..1cd88d6db65 100644
--- a/src/test/modules/injection_points/meson.build
+++ b/src/test/modules/injection_points/meson.build
@@ -46,6 +46,7 @@ tests += {
       'basic',
       'inplace',
       'repack',
+      'repack_deadlock',
       'repack_toast',
       'syscache-update-pruned',
       'heap_lock_update',
diff --git a/src/test/modules/injection_points/specs/repack_deadlock.spec b/src/test/modules/injection_points/specs/repack_deadlock.spec
new file mode 100644
index 00000000000..9d23a6588c2
--- /dev/null
+++ b/src/test/modules/injection_points/specs/repack_deadlock.spec
@@ -0,0 +1,83 @@
+# Test REPACK with a concurrent transaction that would cause a deadlock
+setup
+{
+	CREATE EXTENSION injection_points;
+
+	CREATE TABLE repack_deadlock(i int PRIMARY KEY, j int);
+	INSERT INTO repack_deadlock(i, j) VALUES (1, 1), (2, 2), (3, 3), (4, 4);
+
+	CREATE TABLE relfilenodes(node oid);
+
+	CREATE TABLE data_s1(i int, j int);
+	CREATE TABLE data_s2(i int, j int);
+}
+
+teardown
+{
+	DROP TABLE repack_deadlock;
+	DROP EXTENSION injection_points;
+
+	DROP TABLE relfilenodes;
+	DROP TABLE data_s1;
+	DROP TABLE data_s2;
+}
+
+session s1
+setup
+{
+	SELECT injection_points_set_local();
+	SELECT injection_points_attach('repack-concurrently-before-lock', 'wait');
+}
+# Perform the initial load and wait for s2 to do some data changes.
+step wait_before_lock
+{
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+}
+# Check the table from the perspective of s1.
+#
+# Besides the contents, we also check that relfilenode has changed.
+
+# Have each session write the contents into a table and use FULL JOIN to check
+# if the outputs are identical.
+step check1
+{
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+}
+teardown
+{
+	SELECT injection_points_detach('repack-concurrently-before-lock');
+}
+
+session s2
+# Change the existing data. UPDATE changes both key and non-key columns. Also
+# update one row twice to test whether tuple version generated by this session
+# can be found.
+step add_column
+{
+	alter table repack_deadlock add column noise text;
+}
+
+step wakeup_before_lock
+{
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+}
+
+# Test if data changes introduced while one session is performing REPACK
+# CONCURRENTLY find their way into the table.
+permutation
+	wait_before_lock
+	add_column
+	wakeup_before_lock
+	check1
-- 
2.47.3


--qfkt2ktdpcfeypib
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v48-0007-Teach-snapshot-builder-to-skip-transactions-runn.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread

* [PATCH v45 7/8] Error out any process that would block at REPACK
@ 2026-03-25 19:35 Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Álvaro Herrera @ 2026-03-25 19:35 UTC (permalink / raw)

Any process waiting on REPACK to release its lock would actually cause
it to deadlock when it tries to upgrade its lock to AEL, losing all work
done to that point.  We avoid this by teaching the deadlock detector to
raise an error when this condition is detected.
---
 src/backend/commands/repack.c                 | 53 ++++++++----
 src/backend/storage/lmgr/deadlock.c           | 15 ++++
 src/include/storage/proc.h                    |  6 +-
 src/test/modules/injection_points/Makefile    |  1 +
 .../expected/repack_deadlock.out              | 63 ++++++++++++++
 src/test/modules/injection_points/meson.build |  1 +
 .../specs/repack_deadlock.spec                | 83 +++++++++++++++++++
 7 files changed, 202 insertions(+), 20 deletions(-)
 create mode 100644 src/test/modules/injection_points/expected/repack_deadlock.out
 create mode 100644 src/test/modules/injection_points/specs/repack_deadlock.spec

diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c
index d25ab43da7b..7291b4180ab 100644
--- a/src/backend/commands/repack.c
+++ b/src/backend/commands/repack.c
@@ -276,6 +276,16 @@ ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
 	/* Determine the lock mode to use. */
 	lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
 
+	/*
+	 * If in concurrent mode, set the PROC_IN_CONCURRENT_REPACK flag.  This
+	 * makes the deadlock checker cause anyone that would conflict with us
+	 * to error out.  It's important to set this flag ahead of actually locking
+	 * the relation; it won't of course affect anyone until we do have a lock
+	 * that others can conflict with.
+	 */
+	if ((params.options & CLUOPT_CONCURRENT) != 0)
+		MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK;
+
 	/*
 	 * If a single relation is specified, process it and we're done ... unless
 	 * the relation is a partitioned table, in which case we fall through.
@@ -476,11 +486,8 @@ RepackLockLevel(bool concurrent)
  * If indexOid is InvalidOid, the table will be rewritten in physical order
  * instead of index order.
  *
- * Note that, in the concurrent case, the function releases the lock at some
- * point, in order to get AccessExclusiveLock for the final steps (i.e. to
- * swap the relation files). To make things simpler, the caller should expect
- * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
- * AccessExclusiveLock is kept till the end of the transaction.)
+ * On return, OldHeap is closed but locked with AccessExclusiveLock - the lock
+ * will be released at end of the transaction.
  *
  * 'cmd' indicates which command is being executed, to be used for error
  * messages.
@@ -512,10 +519,12 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
 		/*
 		 * Make sure we're not in a transaction block.
 		 *
-		 * The reason is that repack_setup_logical_decoding() could deadlock
-		 * if there's an XID already assigned.  It would be possible to run in
-		 * a transaction block if we had no XID, but this restriction is
-		 * simpler for users to understand and we don't lose anything.
+		 * The reason is that repack_setup_logical_decoding() could wait
+		 * indefinitely for our XID to complete. (The deadlock detector would
+		 * not recognize it because we'd be waiting for ourselves, i.e. no
+		 * real lock conflict.) It would be possible to run in a transaction
+		 * block if we had no XID, but this restriction is simpler for users
+		 * to understand and we don't lose anything.
 		 */
 		PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
 
@@ -998,10 +1007,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose,
 		 * Note that the worker has to wait for all transactions with XID
 		 * already assigned to finish. If some of those transactions is
 		 * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
-		 * table (e.g.  it runs CREATE INDEX), we can end up in a deadlock.
-		 * Not sure this risk is worth unlocking/locking the table (and its
-		 * clustering index) and checking again if it's still eligible for
-		 * REPACK CONCURRENTLY.
+		 * table (e.g. it runs CREATE INDEX), it should encounter ERROR in the
+		 * deadlock checking code.
 		 */
 		start_repack_decoding_worker(tableOid);
 
@@ -3119,10 +3126,18 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
 		LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
 
 	/*
-	 * Tuples and pages of the old heap will be gone, but the heap will stay.
+	 * Now that we have all access-exclusive locks on all relations, we no
+	 * longer want other processes to error out when trying to acquire a
+	 * conflicting lock.  Therefore, unset our flag.
+	 */
+	MyProc->statusFlags &= ~PROC_IN_CONCURRENT_REPACK;
+
+	/*
+	 * Tuples and pages of the old heap will be gone, but the heap itself will
+	 * stay.  In order for predicate locks to continue to work, convert them
+	 * to relation-level locks.  We do this both for table and indexes.
 	 */
 	TransferPredicateLocksToHeapRelation(OldHeap);
-	/* The same for indexes. */
 	for (int i = 0; i < nind; i++)
 	{
 		Relation	index = ind_refs[i];
@@ -3342,9 +3357,11 @@ start_repack_decoding_worker(Oid relid)
 
 	/*
 	 * The decoding setup must be done before the caller can have XID assigned
-	 * for any reason, otherwise the worker might end up in a deadlock,
-	 * waiting for the caller's transaction to end. Therefore wait here until
-	 * the worker indicates that it has the logical decoding initialized.
+	 * for any reason, otherwise the worker might end up waiting for the
+	 * caller's transaction to end. (Deadlock detector does not consider this
+	 * a conflict because the worker is in the same locking group as the
+	 * backend that launched it.) Therefore wait here until the worker
+	 * indicates that it has the logical decoding initialized.
 	 */
 	ConditionVariablePrepareToSleep(&shared->cv);
 	for (;;)
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index b8962d875b6..3d3ec0da111 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -620,6 +620,21 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 						proc->statusFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
+					/*
+					 * Similarly, if we note that we're blocked by some process
+					 * running REPACK (CONCURRENTLY), just fail.  That process
+					 * is going to upgrade its lock at some point, and it would
+					 * be inappropriate for any other process to cause that
+					 * to fail.
+					 */
+					if (checkProc == MyProc &&
+						proc->statusFlags & PROC_IN_CONCURRENT_REPACK)
+						ereport(ERROR,
+								errcode(ERRCODE_OBJECT_IN_USE),
+								errmsg("could not wait for concurrent REPACK"),
+								errdetail("Process %d waits for REPACK running on process %d",
+									   MyProc->pid, proc->pid));
+
 					/* We're done looking at this proclock */
 					break;
 				}
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1dad125706e..8ad9718f3d6 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -69,10 +69,12 @@ struct XidCache
 #define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be
 												 * included in vacuum horizons
 												 * in all databases */
+#define		PROC_IN_CONCURRENT_REPACK	0x40	/* REPACK (CONCURRENTLY) */
 
-/* flags reset at EOXact */
+/* flags reset at EOXact.  A bit of a misnomer ... */
 #define		PROC_VACUUM_STATE_MASK \
-	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)
+	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | \
+	 PROC_IN_CONCURRENT_REPACK)
 
 /*
  * Xmin-related flags. Make sure any flags that affect how the process' Xmin
diff --git a/src/test/modules/injection_points/Makefile b/src/test/modules/injection_points/Makefile
index 2cd7d87c533..f7663859fe2 100644
--- a/src/test/modules/injection_points/Makefile
+++ b/src/test/modules/injection_points/Makefile
@@ -15,6 +15,7 @@ REGRESS_OPTS = --dlpath=$(top_builddir)/src/test/regress
 ISOLATION = basic \
 	    inplace \
 	    repack \
+	    repack_deadlock \
 	    repack_toast \
 	    syscache-update-pruned \
 	    heap_lock_update
diff --git a/src/test/modules/injection_points/expected/repack_deadlock.out b/src/test/modules/injection_points/expected/repack_deadlock.out
new file mode 100644
index 00000000000..a86e4767536
--- /dev/null
+++ b/src/test/modules/injection_points/expected/repack_deadlock.out
@@ -0,0 +1,63 @@
+Parsed test spec with 2 sessions
+
+starting permutation: wait_before_lock add_column wakeup_before_lock check1
+injection_points_attach
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: 
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+ <waiting ...>
+step add_column: 
+	alter table repack_deadlock add column noise text;
+ <waiting ...>
+step add_column: <... completed>
+ERROR:  could not wait for concurrent REPACK
+step wakeup_before_lock: 
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+
+injection_points_wakeup
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: <... completed>
+step check1: 
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+
+count
+-----
+    1
+(1 row)
+
+i|j
+-+-
+1|1
+2|2
+3|3
+4|4
+(4 rows)
+
+count
+-----
+    4
+(1 row)
+
+injection_points_detach
+-----------------------
+                       
+(1 row)
+
diff --git a/src/test/modules/injection_points/meson.build b/src/test/modules/injection_points/meson.build
index a414abb924b..1cd88d6db65 100644
--- a/src/test/modules/injection_points/meson.build
+++ b/src/test/modules/injection_points/meson.build
@@ -46,6 +46,7 @@ tests += {
       'basic',
       'inplace',
       'repack',
+      'repack_deadlock',
       'repack_toast',
       'syscache-update-pruned',
       'heap_lock_update',
diff --git a/src/test/modules/injection_points/specs/repack_deadlock.spec b/src/test/modules/injection_points/specs/repack_deadlock.spec
new file mode 100644
index 00000000000..9d23a6588c2
--- /dev/null
+++ b/src/test/modules/injection_points/specs/repack_deadlock.spec
@@ -0,0 +1,83 @@
+# Test REPACK with a concurrent transaction that would cause a deadlock
+setup
+{
+	CREATE EXTENSION injection_points;
+
+	CREATE TABLE repack_deadlock(i int PRIMARY KEY, j int);
+	INSERT INTO repack_deadlock(i, j) VALUES (1, 1), (2, 2), (3, 3), (4, 4);
+
+	CREATE TABLE relfilenodes(node oid);
+
+	CREATE TABLE data_s1(i int, j int);
+	CREATE TABLE data_s2(i int, j int);
+}
+
+teardown
+{
+	DROP TABLE repack_deadlock;
+	DROP EXTENSION injection_points;
+
+	DROP TABLE relfilenodes;
+	DROP TABLE data_s1;
+	DROP TABLE data_s2;
+}
+
+session s1
+setup
+{
+	SELECT injection_points_set_local();
+	SELECT injection_points_attach('repack-concurrently-before-lock', 'wait');
+}
+# Perform the initial load and wait for s2 to do some data changes.
+step wait_before_lock
+{
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+}
+# Check the table from the perspective of s1.
+#
+# Besides the contents, we also check that relfilenode has changed.
+
+# Have each session write the contents into a table and use FULL JOIN to check
+# if the outputs are identical.
+step check1
+{
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+}
+teardown
+{
+	SELECT injection_points_detach('repack-concurrently-before-lock');
+}
+
+session s2
+# Change the existing data. UPDATE changes both key and non-key columns. Also
+# update one row twice to test whether tuple version generated by this session
+# can be found.
+step add_column
+{
+	alter table repack_deadlock add column noise text;
+}
+
+step wakeup_before_lock
+{
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+}
+
+# Test if data changes introduced while one session is performing REPACK
+# CONCURRENTLY find their way into the table.
+permutation
+	wait_before_lock
+	add_column
+	wakeup_before_lock
+	check1
-- 
2.47.3


--cldir6umqisr673d
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v45-0008-Teach-snapshot-builder-to-skip-transactions-runn.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread

* [PATCH v46 6/7] Error out any process that would block at REPACK
@ 2026-03-25 19:35 Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Álvaro Herrera @ 2026-03-25 19:35 UTC (permalink / raw)

Any process waiting on REPACK to release its lock would actually cause
it to deadlock when it tries to upgrade its lock to AEL, losing all work
done to that point.  We avoid this by teaching the deadlock detector to
raise an error when this condition is detected.
---
 src/backend/commands/repack.c                 | 53 ++++++++----
 src/backend/storage/lmgr/deadlock.c           | 15 ++++
 src/include/storage/proc.h                    |  6 +-
 src/test/modules/injection_points/Makefile    |  1 +
 .../expected/repack_deadlock.out              | 63 ++++++++++++++
 src/test/modules/injection_points/meson.build |  1 +
 .../specs/repack_deadlock.spec                | 83 +++++++++++++++++++
 7 files changed, 202 insertions(+), 20 deletions(-)
 create mode 100644 src/test/modules/injection_points/expected/repack_deadlock.out
 create mode 100644 src/test/modules/injection_points/specs/repack_deadlock.spec

diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c
index a58277bb6c1..21ab81c6f7f 100644
--- a/src/backend/commands/repack.c
+++ b/src/backend/commands/repack.c
@@ -276,6 +276,16 @@ ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
 	/* Determine the lock mode to use. */
 	lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
 
+	/*
+	 * If in concurrent mode, set the PROC_IN_CONCURRENT_REPACK flag.  This
+	 * makes the deadlock checker cause anyone that would conflict with us
+	 * to error out.  It's important to set this flag ahead of actually locking
+	 * the relation; it won't of course affect anyone until we do have a lock
+	 * that others can conflict with.
+	 */
+	if ((params.options & CLUOPT_CONCURRENT) != 0)
+		MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK;
+
 	/*
 	 * If a single relation is specified, process it and we're done ... unless
 	 * the relation is a partitioned table, in which case we fall through.
@@ -476,11 +486,8 @@ RepackLockLevel(bool concurrent)
  * If indexOid is InvalidOid, the table will be rewritten in physical order
  * instead of index order.
  *
- * Note that, in the concurrent case, the function releases the lock at some
- * point, in order to get AccessExclusiveLock for the final steps (i.e. to
- * swap the relation files). To make things simpler, the caller should expect
- * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
- * AccessExclusiveLock is kept till the end of the transaction.)
+ * On return, OldHeap is closed but locked with AccessExclusiveLock - the lock
+ * will be released at end of the transaction.
  *
  * 'cmd' indicates which command is being executed, to be used for error
  * messages.
@@ -512,10 +519,12 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
 		/*
 		 * Make sure we're not in a transaction block.
 		 *
-		 * The reason is that repack_setup_logical_decoding() could deadlock
-		 * if there's an XID already assigned.  It would be possible to run in
-		 * a transaction block if we had no XID, but this restriction is
-		 * simpler for users to understand and we don't lose anything.
+		 * The reason is that repack_setup_logical_decoding() could wait
+		 * indefinitely for our XID to complete. (The deadlock detector would
+		 * not recognize it because we'd be waiting for ourselves, i.e. no
+		 * real lock conflict.) It would be possible to run in a transaction
+		 * block if we had no XID, but this restriction is simpler for users
+		 * to understand and we don't lose anything.
 		 */
 		PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
 
@@ -998,10 +1007,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose,
 		 * Note that the worker has to wait for all transactions with XID
 		 * already assigned to finish. If some of those transactions is
 		 * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
-		 * table (e.g.  it runs CREATE INDEX), we can end up in a deadlock.
-		 * Not sure this risk is worth unlocking/locking the table (and its
-		 * clustering index) and checking again if it's still eligible for
-		 * REPACK CONCURRENTLY.
+		 * table (e.g. it runs CREATE INDEX), it should encounter ERROR in the
+		 * deadlock checking code.
 		 */
 		start_repack_decoding_worker(tableOid);
 
@@ -3123,10 +3130,18 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
 		LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
 
 	/*
-	 * Tuples and pages of the old heap will be gone, but the heap will stay.
+	 * Now that we have all access-exclusive locks on all relations, we no
+	 * longer want other processes to error out when trying to acquire a
+	 * conflicting lock.  Therefore, unset our flag.
+	 */
+	MyProc->statusFlags &= ~PROC_IN_CONCURRENT_REPACK;
+
+	/*
+	 * Tuples and pages of the old heap will be gone, but the heap itself will
+	 * stay.  In order for predicate locks to continue to work, convert them
+	 * to relation-level locks.  We do this both for table and indexes.
 	 */
 	TransferPredicateLocksToHeapRelation(OldHeap);
-	/* The same for indexes. */
 	for (int i = 0; i < nind; i++)
 	{
 		Relation	index = ind_refs[i];
@@ -3346,9 +3361,11 @@ start_repack_decoding_worker(Oid relid)
 
 	/*
 	 * The decoding setup must be done before the caller can have XID assigned
-	 * for any reason, otherwise the worker might end up in a deadlock,
-	 * waiting for the caller's transaction to end. Therefore wait here until
-	 * the worker indicates that it has the logical decoding initialized.
+	 * for any reason, otherwise the worker might end up waiting for the
+	 * caller's transaction to end. (Deadlock detector does not consider this
+	 * a conflict because the worker is in the same locking group as the
+	 * backend that launched it.) Therefore wait here until the worker
+	 * indicates that it has the logical decoding initialized.
 	 */
 	ConditionVariablePrepareToSleep(&shared->cv);
 	for (;;)
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index b8962d875b6..3d3ec0da111 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -620,6 +620,21 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 						proc->statusFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
+					/*
+					 * Similarly, if we note that we're blocked by some process
+					 * running REPACK (CONCURRENTLY), just fail.  That process
+					 * is going to upgrade its lock at some point, and it would
+					 * be inappropriate for any other process to cause that
+					 * to fail.
+					 */
+					if (checkProc == MyProc &&
+						proc->statusFlags & PROC_IN_CONCURRENT_REPACK)
+						ereport(ERROR,
+								errcode(ERRCODE_OBJECT_IN_USE),
+								errmsg("could not wait for concurrent REPACK"),
+								errdetail("Process %d waits for REPACK running on process %d",
+									   MyProc->pid, proc->pid));
+
 					/* We're done looking at this proclock */
 					break;
 				}
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1dad125706e..8ad9718f3d6 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -69,10 +69,12 @@ struct XidCache
 #define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be
 												 * included in vacuum horizons
 												 * in all databases */
+#define		PROC_IN_CONCURRENT_REPACK	0x40	/* REPACK (CONCURRENTLY) */
 
-/* flags reset at EOXact */
+/* flags reset at EOXact.  A bit of a misnomer ... */
 #define		PROC_VACUUM_STATE_MASK \
-	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)
+	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | \
+	 PROC_IN_CONCURRENT_REPACK)
 
 /*
  * Xmin-related flags. Make sure any flags that affect how the process' Xmin
diff --git a/src/test/modules/injection_points/Makefile b/src/test/modules/injection_points/Makefile
index 2cd7d87c533..f7663859fe2 100644
--- a/src/test/modules/injection_points/Makefile
+++ b/src/test/modules/injection_points/Makefile
@@ -15,6 +15,7 @@ REGRESS_OPTS = --dlpath=$(top_builddir)/src/test/regress
 ISOLATION = basic \
 	    inplace \
 	    repack \
+	    repack_deadlock \
 	    repack_toast \
 	    syscache-update-pruned \
 	    heap_lock_update
diff --git a/src/test/modules/injection_points/expected/repack_deadlock.out b/src/test/modules/injection_points/expected/repack_deadlock.out
new file mode 100644
index 00000000000..a86e4767536
--- /dev/null
+++ b/src/test/modules/injection_points/expected/repack_deadlock.out
@@ -0,0 +1,63 @@
+Parsed test spec with 2 sessions
+
+starting permutation: wait_before_lock add_column wakeup_before_lock check1
+injection_points_attach
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: 
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+ <waiting ...>
+step add_column: 
+	alter table repack_deadlock add column noise text;
+ <waiting ...>
+step add_column: <... completed>
+ERROR:  could not wait for concurrent REPACK
+step wakeup_before_lock: 
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+
+injection_points_wakeup
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: <... completed>
+step check1: 
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+
+count
+-----
+    1
+(1 row)
+
+i|j
+-+-
+1|1
+2|2
+3|3
+4|4
+(4 rows)
+
+count
+-----
+    4
+(1 row)
+
+injection_points_detach
+-----------------------
+                       
+(1 row)
+
diff --git a/src/test/modules/injection_points/meson.build b/src/test/modules/injection_points/meson.build
index a414abb924b..1cd88d6db65 100644
--- a/src/test/modules/injection_points/meson.build
+++ b/src/test/modules/injection_points/meson.build
@@ -46,6 +46,7 @@ tests += {
       'basic',
       'inplace',
       'repack',
+      'repack_deadlock',
       'repack_toast',
       'syscache-update-pruned',
       'heap_lock_update',
diff --git a/src/test/modules/injection_points/specs/repack_deadlock.spec b/src/test/modules/injection_points/specs/repack_deadlock.spec
new file mode 100644
index 00000000000..9d23a6588c2
--- /dev/null
+++ b/src/test/modules/injection_points/specs/repack_deadlock.spec
@@ -0,0 +1,83 @@
+# Test REPACK with a concurrent transaction that would cause a deadlock
+setup
+{
+	CREATE EXTENSION injection_points;
+
+	CREATE TABLE repack_deadlock(i int PRIMARY KEY, j int);
+	INSERT INTO repack_deadlock(i, j) VALUES (1, 1), (2, 2), (3, 3), (4, 4);
+
+	CREATE TABLE relfilenodes(node oid);
+
+	CREATE TABLE data_s1(i int, j int);
+	CREATE TABLE data_s2(i int, j int);
+}
+
+teardown
+{
+	DROP TABLE repack_deadlock;
+	DROP EXTENSION injection_points;
+
+	DROP TABLE relfilenodes;
+	DROP TABLE data_s1;
+	DROP TABLE data_s2;
+}
+
+session s1
+setup
+{
+	SELECT injection_points_set_local();
+	SELECT injection_points_attach('repack-concurrently-before-lock', 'wait');
+}
+# Perform the initial load and wait for s2 to do some data changes.
+step wait_before_lock
+{
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+}
+# Check the table from the perspective of s1.
+#
+# Besides the contents, we also check that relfilenode has changed.
+
+# Have each session write the contents into a table and use FULL JOIN to check
+# if the outputs are identical.
+step check1
+{
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+}
+teardown
+{
+	SELECT injection_points_detach('repack-concurrently-before-lock');
+}
+
+session s2
+# Change the existing data. UPDATE changes both key and non-key columns. Also
+# update one row twice to test whether tuple version generated by this session
+# can be found.
+step add_column
+{
+	alter table repack_deadlock add column noise text;
+}
+
+step wakeup_before_lock
+{
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+}
+
+# Test if data changes introduced while one session is performing REPACK
+# CONCURRENTLY find their way into the table.
+permutation
+	wait_before_lock
+	add_column
+	wakeup_before_lock
+	check1
-- 
2.47.3


--mdsv5bdgfjjj6d77
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v46-0007-Teach-snapshot-builder-to-skip-transactions-runn.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread

* [PATCH v46 6/7] Error out any process that would block at REPACK
@ 2026-03-25 19:35 Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Álvaro Herrera @ 2026-03-25 19:35 UTC (permalink / raw)

Any process waiting on REPACK to release its lock would actually cause
it to deadlock when it tries to upgrade its lock to AEL, losing all work
done to that point.  We avoid this by teaching the deadlock detector to
raise an error when this condition is detected.
---
 src/backend/commands/repack.c                 | 53 ++++++++----
 src/backend/storage/lmgr/deadlock.c           | 15 ++++
 src/include/storage/proc.h                    |  6 +-
 src/test/modules/injection_points/Makefile    |  1 +
 .../expected/repack_deadlock.out              | 63 ++++++++++++++
 src/test/modules/injection_points/meson.build |  1 +
 .../specs/repack_deadlock.spec                | 83 +++++++++++++++++++
 7 files changed, 202 insertions(+), 20 deletions(-)
 create mode 100644 src/test/modules/injection_points/expected/repack_deadlock.out
 create mode 100644 src/test/modules/injection_points/specs/repack_deadlock.spec

diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c
index a58277bb6c1..21ab81c6f7f 100644
--- a/src/backend/commands/repack.c
+++ b/src/backend/commands/repack.c
@@ -276,6 +276,16 @@ ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
 	/* Determine the lock mode to use. */
 	lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
 
+	/*
+	 * If in concurrent mode, set the PROC_IN_CONCURRENT_REPACK flag.  This
+	 * makes the deadlock checker cause anyone that would conflict with us
+	 * to error out.  It's important to set this flag ahead of actually locking
+	 * the relation; it won't of course affect anyone until we do have a lock
+	 * that others can conflict with.
+	 */
+	if ((params.options & CLUOPT_CONCURRENT) != 0)
+		MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK;
+
 	/*
 	 * If a single relation is specified, process it and we're done ... unless
 	 * the relation is a partitioned table, in which case we fall through.
@@ -476,11 +486,8 @@ RepackLockLevel(bool concurrent)
  * If indexOid is InvalidOid, the table will be rewritten in physical order
  * instead of index order.
  *
- * Note that, in the concurrent case, the function releases the lock at some
- * point, in order to get AccessExclusiveLock for the final steps (i.e. to
- * swap the relation files). To make things simpler, the caller should expect
- * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
- * AccessExclusiveLock is kept till the end of the transaction.)
+ * On return, OldHeap is closed but locked with AccessExclusiveLock - the lock
+ * will be released at end of the transaction.
  *
  * 'cmd' indicates which command is being executed, to be used for error
  * messages.
@@ -512,10 +519,12 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
 		/*
 		 * Make sure we're not in a transaction block.
 		 *
-		 * The reason is that repack_setup_logical_decoding() could deadlock
-		 * if there's an XID already assigned.  It would be possible to run in
-		 * a transaction block if we had no XID, but this restriction is
-		 * simpler for users to understand and we don't lose anything.
+		 * The reason is that repack_setup_logical_decoding() could wait
+		 * indefinitely for our XID to complete. (The deadlock detector would
+		 * not recognize it because we'd be waiting for ourselves, i.e. no
+		 * real lock conflict.) It would be possible to run in a transaction
+		 * block if we had no XID, but this restriction is simpler for users
+		 * to understand and we don't lose anything.
 		 */
 		PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
 
@@ -998,10 +1007,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose,
 		 * Note that the worker has to wait for all transactions with XID
 		 * already assigned to finish. If some of those transactions is
 		 * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
-		 * table (e.g.  it runs CREATE INDEX), we can end up in a deadlock.
-		 * Not sure this risk is worth unlocking/locking the table (and its
-		 * clustering index) and checking again if it's still eligible for
-		 * REPACK CONCURRENTLY.
+		 * table (e.g. it runs CREATE INDEX), it should encounter ERROR in the
+		 * deadlock checking code.
 		 */
 		start_repack_decoding_worker(tableOid);
 
@@ -3123,10 +3130,18 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
 		LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
 
 	/*
-	 * Tuples and pages of the old heap will be gone, but the heap will stay.
+	 * Now that we have all access-exclusive locks on all relations, we no
+	 * longer want other processes to error out when trying to acquire a
+	 * conflicting lock.  Therefore, unset our flag.
+	 */
+	MyProc->statusFlags &= ~PROC_IN_CONCURRENT_REPACK;
+
+	/*
+	 * Tuples and pages of the old heap will be gone, but the heap itself will
+	 * stay.  In order for predicate locks to continue to work, convert them
+	 * to relation-level locks.  We do this both for table and indexes.
 	 */
 	TransferPredicateLocksToHeapRelation(OldHeap);
-	/* The same for indexes. */
 	for (int i = 0; i < nind; i++)
 	{
 		Relation	index = ind_refs[i];
@@ -3346,9 +3361,11 @@ start_repack_decoding_worker(Oid relid)
 
 	/*
 	 * The decoding setup must be done before the caller can have XID assigned
-	 * for any reason, otherwise the worker might end up in a deadlock,
-	 * waiting for the caller's transaction to end. Therefore wait here until
-	 * the worker indicates that it has the logical decoding initialized.
+	 * for any reason, otherwise the worker might end up waiting for the
+	 * caller's transaction to end. (Deadlock detector does not consider this
+	 * a conflict because the worker is in the same locking group as the
+	 * backend that launched it.) Therefore wait here until the worker
+	 * indicates that it has the logical decoding initialized.
 	 */
 	ConditionVariablePrepareToSleep(&shared->cv);
 	for (;;)
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index b8962d875b6..3d3ec0da111 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -620,6 +620,21 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 						proc->statusFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
+					/*
+					 * Similarly, if we note that we're blocked by some process
+					 * running REPACK (CONCURRENTLY), just fail.  That process
+					 * is going to upgrade its lock at some point, and it would
+					 * be inappropriate for any other process to cause that
+					 * to fail.
+					 */
+					if (checkProc == MyProc &&
+						proc->statusFlags & PROC_IN_CONCURRENT_REPACK)
+						ereport(ERROR,
+								errcode(ERRCODE_OBJECT_IN_USE),
+								errmsg("could not wait for concurrent REPACK"),
+								errdetail("Process %d waits for REPACK running on process %d",
+									   MyProc->pid, proc->pid));
+
 					/* We're done looking at this proclock */
 					break;
 				}
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1dad125706e..8ad9718f3d6 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -69,10 +69,12 @@ struct XidCache
 #define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be
 												 * included in vacuum horizons
 												 * in all databases */
+#define		PROC_IN_CONCURRENT_REPACK	0x40	/* REPACK (CONCURRENTLY) */
 
-/* flags reset at EOXact */
+/* flags reset at EOXact.  A bit of a misnomer ... */
 #define		PROC_VACUUM_STATE_MASK \
-	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)
+	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | \
+	 PROC_IN_CONCURRENT_REPACK)
 
 /*
  * Xmin-related flags. Make sure any flags that affect how the process' Xmin
diff --git a/src/test/modules/injection_points/Makefile b/src/test/modules/injection_points/Makefile
index 2cd7d87c533..f7663859fe2 100644
--- a/src/test/modules/injection_points/Makefile
+++ b/src/test/modules/injection_points/Makefile
@@ -15,6 +15,7 @@ REGRESS_OPTS = --dlpath=$(top_builddir)/src/test/regress
 ISOLATION = basic \
 	    inplace \
 	    repack \
+	    repack_deadlock \
 	    repack_toast \
 	    syscache-update-pruned \
 	    heap_lock_update
diff --git a/src/test/modules/injection_points/expected/repack_deadlock.out b/src/test/modules/injection_points/expected/repack_deadlock.out
new file mode 100644
index 00000000000..a86e4767536
--- /dev/null
+++ b/src/test/modules/injection_points/expected/repack_deadlock.out
@@ -0,0 +1,63 @@
+Parsed test spec with 2 sessions
+
+starting permutation: wait_before_lock add_column wakeup_before_lock check1
+injection_points_attach
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: 
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+ <waiting ...>
+step add_column: 
+	alter table repack_deadlock add column noise text;
+ <waiting ...>
+step add_column: <... completed>
+ERROR:  could not wait for concurrent REPACK
+step wakeup_before_lock: 
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+
+injection_points_wakeup
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: <... completed>
+step check1: 
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+
+count
+-----
+    1
+(1 row)
+
+i|j
+-+-
+1|1
+2|2
+3|3
+4|4
+(4 rows)
+
+count
+-----
+    4
+(1 row)
+
+injection_points_detach
+-----------------------
+                       
+(1 row)
+
diff --git a/src/test/modules/injection_points/meson.build b/src/test/modules/injection_points/meson.build
index a414abb924b..1cd88d6db65 100644
--- a/src/test/modules/injection_points/meson.build
+++ b/src/test/modules/injection_points/meson.build
@@ -46,6 +46,7 @@ tests += {
       'basic',
       'inplace',
       'repack',
+      'repack_deadlock',
       'repack_toast',
       'syscache-update-pruned',
       'heap_lock_update',
diff --git a/src/test/modules/injection_points/specs/repack_deadlock.spec b/src/test/modules/injection_points/specs/repack_deadlock.spec
new file mode 100644
index 00000000000..9d23a6588c2
--- /dev/null
+++ b/src/test/modules/injection_points/specs/repack_deadlock.spec
@@ -0,0 +1,83 @@
+# Test REPACK with a concurrent transaction that would cause a deadlock
+setup
+{
+	CREATE EXTENSION injection_points;
+
+	CREATE TABLE repack_deadlock(i int PRIMARY KEY, j int);
+	INSERT INTO repack_deadlock(i, j) VALUES (1, 1), (2, 2), (3, 3), (4, 4);
+
+	CREATE TABLE relfilenodes(node oid);
+
+	CREATE TABLE data_s1(i int, j int);
+	CREATE TABLE data_s2(i int, j int);
+}
+
+teardown
+{
+	DROP TABLE repack_deadlock;
+	DROP EXTENSION injection_points;
+
+	DROP TABLE relfilenodes;
+	DROP TABLE data_s1;
+	DROP TABLE data_s2;
+}
+
+session s1
+setup
+{
+	SELECT injection_points_set_local();
+	SELECT injection_points_attach('repack-concurrently-before-lock', 'wait');
+}
+# Perform the initial load and wait for s2 to do some data changes.
+step wait_before_lock
+{
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+}
+# Check the table from the perspective of s1.
+#
+# Besides the contents, we also check that relfilenode has changed.
+
+# Have each session write the contents into a table and use FULL JOIN to check
+# if the outputs are identical.
+step check1
+{
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+}
+teardown
+{
+	SELECT injection_points_detach('repack-concurrently-before-lock');
+}
+
+session s2
+# Change the existing data. UPDATE changes both key and non-key columns. Also
+# update one row twice to test whether tuple version generated by this session
+# can be found.
+step add_column
+{
+	alter table repack_deadlock add column noise text;
+}
+
+step wakeup_before_lock
+{
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+}
+
+# Test if data changes introduced while one session is performing REPACK
+# CONCURRENTLY find their way into the table.
+permutation
+	wait_before_lock
+	add_column
+	wakeup_before_lock
+	check1
-- 
2.47.3


--mdsv5bdgfjjj6d77
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v46-0007-Teach-snapshot-builder-to-skip-transactions-runn.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread

* [PATCH v47 8/9] Error out any process that would block at REPACK
@ 2026-03-25 19:35 Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Álvaro Herrera @ 2026-03-25 19:35 UTC (permalink / raw)

Any process waiting on REPACK to release its lock would actually cause
it to deadlock when it tries to upgrade its lock to AEL, losing all work
done to that point.  We avoid this by teaching the deadlock detector to
raise an error when this condition is detected.
---
 src/backend/commands/repack.c                 | 52 ++++++++----
 src/backend/storage/lmgr/deadlock.c           | 15 ++++
 src/include/storage/proc.h                    |  6 +-
 src/test/modules/injection_points/Makefile    |  1 +
 .../expected/repack_deadlock.out              | 63 ++++++++++++++
 src/test/modules/injection_points/meson.build |  1 +
 .../specs/repack_deadlock.spec                | 83 +++++++++++++++++++
 7 files changed, 202 insertions(+), 19 deletions(-)
 create mode 100644 src/test/modules/injection_points/expected/repack_deadlock.out
 create mode 100644 src/test/modules/injection_points/specs/repack_deadlock.spec

diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c
index 9c807f75d71..3acb2e1d2ba 100644
--- a/src/backend/commands/repack.c
+++ b/src/backend/commands/repack.c
@@ -276,6 +276,16 @@ ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
 	/* Determine the lock mode to use. */
 	lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
 
+	/*
+	 * If in concurrent mode, set the PROC_IN_CONCURRENT_REPACK flag.  This
+	 * makes the deadlock checker cause anyone that would conflict with us
+	 * to error out.  It's important to set this flag ahead of actually locking
+	 * the relation; it won't of course affect anyone until we do have a lock
+	 * that others can conflict with.
+	 */
+	if ((params.options & CLUOPT_CONCURRENT) != 0)
+		MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK;
+
 	/*
 	 * If a single relation is specified, process it and we're done ... unless
 	 * the relation is a partitioned table, in which case we fall through.
@@ -476,11 +486,8 @@ RepackLockLevel(bool concurrent)
  * If indexOid is InvalidOid, the table will be rewritten in physical order
  * instead of index order.
  *
- * Note that, in the concurrent case, the function releases the lock at some
- * point, in order to get AccessExclusiveLock for the final steps (i.e. to
- * swap the relation files). To make things simpler, the caller should expect
- * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
- * AccessExclusiveLock is kept till the end of the transaction.)
+ * On return, OldHeap is closed but locked with AccessExclusiveLock - the lock
+ * will be released at end of the transaction.
  *
  * 'cmd' indicates which command is being executed, to be used for error
  * messages.
@@ -512,10 +519,12 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
 		/*
 		 * Make sure we're not in a transaction block.
 		 *
-		 * The reason is that repack_setup_logical_decoding() could deadlock
-		 * if there's an XID already assigned.  It would be possible to run in
-		 * a transaction block if we had no XID, but this restriction is
-		 * simpler for users to understand and we don't lose anything.
+		 * The reason is that repack_setup_logical_decoding() could wait
+		 * indefinitely for our XID to complete. (The deadlock detector would
+		 * not recognize it because we'd be waiting for ourselves, i.e. no
+		 * real lock conflict.) It would be possible to run in a transaction
+		 * block if we had no XID, but this restriction is simpler for users
+		 * to understand and we don't lose anything.
 		 */
 		PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
 
@@ -998,10 +1007,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose,
 		 * Note that the worker has to wait for all transactions with XID
 		 * already assigned to finish. If some of those transactions is
 		 * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
-		 * table (e.g.  it runs CREATE INDEX), we can end up in a deadlock.
-		 * Not sure this risk is worth unlocking/locking the table (and its
-		 * clustering index) and checking again if it's still eligible for
-		 * REPACK CONCURRENTLY.
+		 * table (e.g. it runs CREATE INDEX), it should encounter ERROR in the
+		 * deadlock checking code.
 		 */
 		start_repack_decoding_worker(tableOid);
 
@@ -3090,7 +3097,16 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
 		LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
 
 	/*
-	 * Tuples and pages of the old heap will be gone, but the heap will stay.
+	 * Now that we have all access-exclusive locks on all relations, we no
+	 * longer want other processes to error out when trying to acquire a
+	 * conflicting lock.  Therefore, unset our flag.
+	 */
+	MyProc->statusFlags &= ~PROC_IN_CONCURRENT_REPACK;
+
+	/*
+	 * Tuples and pages of the old heap will be gone, but the heap itself will
+	 * stay.  In order for predicate locks to continue to work, convert them
+	 * to relation-level locks.  We do this both for table and indexes.
 	 */
 	TransferPredicateLocksToHeapRelation(OldHeap);
 	foreach_ptr(RelationData, index, indexrels)
@@ -3287,9 +3303,11 @@ start_repack_decoding_worker(Oid relid)
 
 	/*
 	 * The decoding setup must be done before the caller can have XID assigned
-	 * for any reason, otherwise the worker might end up in a deadlock,
-	 * waiting for the caller's transaction to end. Therefore wait here until
-	 * the worker indicates that it has the logical decoding initialized.
+	 * for any reason, otherwise the worker might end up waiting for the
+	 * caller's transaction to end. (Deadlock detector does not consider this
+	 * a conflict because the worker is in the same locking group as the
+	 * backend that launched it.) Therefore wait here until the worker
+	 * indicates that it has the logical decoding initialized.
 	 */
 	ConditionVariablePrepareToSleep(&shared->cv);
 	for (;;)
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index b8962d875b6..3d3ec0da111 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -620,6 +620,21 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 						proc->statusFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
+					/*
+					 * Similarly, if we note that we're blocked by some process
+					 * running REPACK (CONCURRENTLY), just fail.  That process
+					 * is going to upgrade its lock at some point, and it would
+					 * be inappropriate for any other process to cause that
+					 * to fail.
+					 */
+					if (checkProc == MyProc &&
+						proc->statusFlags & PROC_IN_CONCURRENT_REPACK)
+						ereport(ERROR,
+								errcode(ERRCODE_OBJECT_IN_USE),
+								errmsg("could not wait for concurrent REPACK"),
+								errdetail("Process %d waits for REPACK running on process %d",
+									   MyProc->pid, proc->pid));
+
 					/* We're done looking at this proclock */
 					break;
 				}
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1dad125706e..8ad9718f3d6 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -69,10 +69,12 @@ struct XidCache
 #define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be
 												 * included in vacuum horizons
 												 * in all databases */
+#define		PROC_IN_CONCURRENT_REPACK	0x40	/* REPACK (CONCURRENTLY) */
 
-/* flags reset at EOXact */
+/* flags reset at EOXact.  A bit of a misnomer ... */
 #define		PROC_VACUUM_STATE_MASK \
-	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)
+	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | \
+	 PROC_IN_CONCURRENT_REPACK)
 
 /*
  * Xmin-related flags. Make sure any flags that affect how the process' Xmin
diff --git a/src/test/modules/injection_points/Makefile b/src/test/modules/injection_points/Makefile
index 2cd7d87c533..f7663859fe2 100644
--- a/src/test/modules/injection_points/Makefile
+++ b/src/test/modules/injection_points/Makefile
@@ -15,6 +15,7 @@ REGRESS_OPTS = --dlpath=$(top_builddir)/src/test/regress
 ISOLATION = basic \
 	    inplace \
 	    repack \
+	    repack_deadlock \
 	    repack_toast \
 	    syscache-update-pruned \
 	    heap_lock_update
diff --git a/src/test/modules/injection_points/expected/repack_deadlock.out b/src/test/modules/injection_points/expected/repack_deadlock.out
new file mode 100644
index 00000000000..a86e4767536
--- /dev/null
+++ b/src/test/modules/injection_points/expected/repack_deadlock.out
@@ -0,0 +1,63 @@
+Parsed test spec with 2 sessions
+
+starting permutation: wait_before_lock add_column wakeup_before_lock check1
+injection_points_attach
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: 
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+ <waiting ...>
+step add_column: 
+	alter table repack_deadlock add column noise text;
+ <waiting ...>
+step add_column: <... completed>
+ERROR:  could not wait for concurrent REPACK
+step wakeup_before_lock: 
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+
+injection_points_wakeup
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: <... completed>
+step check1: 
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+
+count
+-----
+    1
+(1 row)
+
+i|j
+-+-
+1|1
+2|2
+3|3
+4|4
+(4 rows)
+
+count
+-----
+    4
+(1 row)
+
+injection_points_detach
+-----------------------
+                       
+(1 row)
+
diff --git a/src/test/modules/injection_points/meson.build b/src/test/modules/injection_points/meson.build
index a414abb924b..1cd88d6db65 100644
--- a/src/test/modules/injection_points/meson.build
+++ b/src/test/modules/injection_points/meson.build
@@ -46,6 +46,7 @@ tests += {
       'basic',
       'inplace',
       'repack',
+      'repack_deadlock',
       'repack_toast',
       'syscache-update-pruned',
       'heap_lock_update',
diff --git a/src/test/modules/injection_points/specs/repack_deadlock.spec b/src/test/modules/injection_points/specs/repack_deadlock.spec
new file mode 100644
index 00000000000..9d23a6588c2
--- /dev/null
+++ b/src/test/modules/injection_points/specs/repack_deadlock.spec
@@ -0,0 +1,83 @@
+# Test REPACK with a concurrent transaction that would cause a deadlock
+setup
+{
+	CREATE EXTENSION injection_points;
+
+	CREATE TABLE repack_deadlock(i int PRIMARY KEY, j int);
+	INSERT INTO repack_deadlock(i, j) VALUES (1, 1), (2, 2), (3, 3), (4, 4);
+
+	CREATE TABLE relfilenodes(node oid);
+
+	CREATE TABLE data_s1(i int, j int);
+	CREATE TABLE data_s2(i int, j int);
+}
+
+teardown
+{
+	DROP TABLE repack_deadlock;
+	DROP EXTENSION injection_points;
+
+	DROP TABLE relfilenodes;
+	DROP TABLE data_s1;
+	DROP TABLE data_s2;
+}
+
+session s1
+setup
+{
+	SELECT injection_points_set_local();
+	SELECT injection_points_attach('repack-concurrently-before-lock', 'wait');
+}
+# Perform the initial load and wait for s2 to do some data changes.
+step wait_before_lock
+{
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+}
+# Check the table from the perspective of s1.
+#
+# Besides the contents, we also check that relfilenode has changed.
+
+# Have each session write the contents into a table and use FULL JOIN to check
+# if the outputs are identical.
+step check1
+{
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+}
+teardown
+{
+	SELECT injection_points_detach('repack-concurrently-before-lock');
+}
+
+session s2
+# Change the existing data. UPDATE changes both key and non-key columns. Also
+# update one row twice to test whether tuple version generated by this session
+# can be found.
+step add_column
+{
+	alter table repack_deadlock add column noise text;
+}
+
+step wakeup_before_lock
+{
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+}
+
+# Test if data changes introduced while one session is performing REPACK
+# CONCURRENTLY find their way into the table.
+permutation
+	wait_before_lock
+	add_column
+	wakeup_before_lock
+	check1
-- 
2.47.3


--2fxmamo6mu2qbxgv
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v47-0009-Teach-snapshot-builder-to-skip-transactions-runn.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread

* [PATCH v48 6/7] Error out any process that would block at REPACK
@ 2026-03-25 19:35 Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Álvaro Herrera @ 2026-03-25 19:35 UTC (permalink / raw)

Any process waiting on REPACK to release its lock would actually cause
it to deadlock when it tries to upgrade its lock to AEL, losing all work
done to that point.  We avoid this by teaching the deadlock detector to
raise an error when this condition is detected.
---
 src/backend/commands/repack.c                 | 52 ++++++++----
 src/backend/storage/lmgr/deadlock.c           | 15 ++++
 src/include/storage/proc.h                    |  6 +-
 src/test/modules/injection_points/Makefile    |  1 +
 .../expected/repack_deadlock.out              | 63 ++++++++++++++
 src/test/modules/injection_points/meson.build |  1 +
 .../specs/repack_deadlock.spec                | 83 +++++++++++++++++++
 7 files changed, 202 insertions(+), 19 deletions(-)
 create mode 100644 src/test/modules/injection_points/expected/repack_deadlock.out
 create mode 100644 src/test/modules/injection_points/specs/repack_deadlock.spec

diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c
index a514053b777..781a471fc1d 100644
--- a/src/backend/commands/repack.c
+++ b/src/backend/commands/repack.c
@@ -276,6 +276,16 @@ ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
 	/* Determine the lock mode to use. */
 	lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
 
+	/*
+	 * If in concurrent mode, set the PROC_IN_CONCURRENT_REPACK flag.  This
+	 * makes the deadlock checker cause anyone that would conflict with us to
+	 * error out.  It's important to set this flag ahead of actually locking
+	 * the relation; it won't of course affect anyone until we do have a lock
+	 * that others can conflict with.
+	 */
+	if ((params.options & CLUOPT_CONCURRENT) != 0)
+		MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK;
+
 	/*
 	 * If a single relation is specified, process it and we're done ... unless
 	 * the relation is a partitioned table, in which case we fall through.
@@ -476,11 +486,8 @@ RepackLockLevel(bool concurrent)
  * If indexOid is InvalidOid, the table will be rewritten in physical order
  * instead of index order.
  *
- * Note that, in the concurrent case, the function releases the lock at some
- * point, in order to get AccessExclusiveLock for the final steps (i.e. to
- * swap the relation files). To make things simpler, the caller should expect
- * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
- * AccessExclusiveLock is kept till the end of the transaction.)
+ * On return, OldHeap is closed but locked with AccessExclusiveLock - the lock
+ * will be released at end of the transaction.
  *
  * 'cmd' indicates which command is being executed, to be used for error
  * messages.
@@ -512,10 +519,12 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
 		/*
 		 * Make sure we're not in a transaction block.
 		 *
-		 * The reason is that repack_setup_logical_decoding() could deadlock
-		 * if there's an XID already assigned.  It would be possible to run in
-		 * a transaction block if we had no XID, but this restriction is
-		 * simpler for users to understand and we don't lose anything.
+		 * The reason is that repack_setup_logical_decoding() could wait
+		 * indefinitely for our XID to complete. (The deadlock detector would
+		 * not recognize it because we'd be waiting for ourselves, i.e. no
+		 * real lock conflict.) It would be possible to run in a transaction
+		 * block if we had no XID, but this restriction is simpler for users
+		 * to understand and we don't lose anything.
 		 */
 		PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
 
@@ -998,10 +1007,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose,
 		 * Note that the worker has to wait for all transactions with XID
 		 * already assigned to finish. If some of those transactions is
 		 * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
-		 * table (e.g.  it runs CREATE INDEX), we can end up in a deadlock.
-		 * Not sure this risk is worth unlocking/locking the table (and its
-		 * clustering index) and checking again if it's still eligible for
-		 * REPACK CONCURRENTLY.
+		 * table (e.g. it runs CREATE INDEX), it should encounter ERROR in the
+		 * deadlock checking code.
 		 */
 		start_repack_decoding_worker(tableOid);
 
@@ -3090,7 +3097,16 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
 		LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
 
 	/*
-	 * Tuples and pages of the old heap will be gone, but the heap will stay.
+	 * Now that we have all access-exclusive locks on all relations, we no
+	 * longer want other processes to error out when trying to acquire a
+	 * conflicting lock.  Therefore, unset our flag.
+	 */
+	MyProc->statusFlags &= ~PROC_IN_CONCURRENT_REPACK;
+
+	/*
+	 * Tuples and pages of the old heap will be gone, but the heap itself will
+	 * stay.  In order for predicate locks to continue to work, convert them
+	 * to relation-level locks.  We do this both for table and indexes.
 	 */
 	TransferPredicateLocksToHeapRelation(OldHeap);
 	foreach_ptr(RelationData, index, indexrels)
@@ -3288,9 +3304,11 @@ start_repack_decoding_worker(Oid relid)
 
 	/*
 	 * The decoding setup must be done before the caller can have XID assigned
-	 * for any reason, otherwise the worker might end up in a deadlock,
-	 * waiting for the caller's transaction to end. Therefore wait here until
-	 * the worker indicates that it has the logical decoding initialized.
+	 * for any reason, otherwise the worker might end up waiting for the
+	 * caller's transaction to end. (Deadlock detector does not consider this
+	 * a conflict because the worker is in the same locking group as the
+	 * backend that launched it.) Therefore wait here until the worker
+	 * indicates that it has the logical decoding initialized.
 	 */
 	ConditionVariablePrepareToSleep(&shared->cv);
 	for (;;)
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index b8962d875b6..c20ac682b0d 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -620,6 +620,21 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 						proc->statusFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
+					/*
+					 * Similarly, if we note that we're blocked by some
+					 * process running REPACK (CONCURRENTLY), just fail.  That
+					 * process is going to upgrade its lock at some point, and
+					 * it would be inappropriate for any other process to
+					 * cause that to fail.
+					 */
+					if (checkProc == MyProc &&
+						proc->statusFlags & PROC_IN_CONCURRENT_REPACK)
+						ereport(ERROR,
+								errcode(ERRCODE_OBJECT_IN_USE),
+								errmsg("could not wait for concurrent REPACK"),
+								errdetail("Process %d waits for REPACK running on process %d",
+										  MyProc->pid, proc->pid));
+
 					/* We're done looking at this proclock */
 					break;
 				}
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1dad125706e..8ad9718f3d6 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -69,10 +69,12 @@ struct XidCache
 #define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be
 												 * included in vacuum horizons
 												 * in all databases */
+#define		PROC_IN_CONCURRENT_REPACK	0x40	/* REPACK (CONCURRENTLY) */
 
-/* flags reset at EOXact */
+/* flags reset at EOXact.  A bit of a misnomer ... */
 #define		PROC_VACUUM_STATE_MASK \
-	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)
+	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | \
+	 PROC_IN_CONCURRENT_REPACK)
 
 /*
  * Xmin-related flags. Make sure any flags that affect how the process' Xmin
diff --git a/src/test/modules/injection_points/Makefile b/src/test/modules/injection_points/Makefile
index 2cd7d87c533..f7663859fe2 100644
--- a/src/test/modules/injection_points/Makefile
+++ b/src/test/modules/injection_points/Makefile
@@ -15,6 +15,7 @@ REGRESS_OPTS = --dlpath=$(top_builddir)/src/test/regress
 ISOLATION = basic \
 	    inplace \
 	    repack \
+	    repack_deadlock \
 	    repack_toast \
 	    syscache-update-pruned \
 	    heap_lock_update
diff --git a/src/test/modules/injection_points/expected/repack_deadlock.out b/src/test/modules/injection_points/expected/repack_deadlock.out
new file mode 100644
index 00000000000..a86e4767536
--- /dev/null
+++ b/src/test/modules/injection_points/expected/repack_deadlock.out
@@ -0,0 +1,63 @@
+Parsed test spec with 2 sessions
+
+starting permutation: wait_before_lock add_column wakeup_before_lock check1
+injection_points_attach
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: 
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+ <waiting ...>
+step add_column: 
+	alter table repack_deadlock add column noise text;
+ <waiting ...>
+step add_column: <... completed>
+ERROR:  could not wait for concurrent REPACK
+step wakeup_before_lock: 
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+
+injection_points_wakeup
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: <... completed>
+step check1: 
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+
+count
+-----
+    1
+(1 row)
+
+i|j
+-+-
+1|1
+2|2
+3|3
+4|4
+(4 rows)
+
+count
+-----
+    4
+(1 row)
+
+injection_points_detach
+-----------------------
+                       
+(1 row)
+
diff --git a/src/test/modules/injection_points/meson.build b/src/test/modules/injection_points/meson.build
index a414abb924b..1cd88d6db65 100644
--- a/src/test/modules/injection_points/meson.build
+++ b/src/test/modules/injection_points/meson.build
@@ -46,6 +46,7 @@ tests += {
       'basic',
       'inplace',
       'repack',
+      'repack_deadlock',
       'repack_toast',
       'syscache-update-pruned',
       'heap_lock_update',
diff --git a/src/test/modules/injection_points/specs/repack_deadlock.spec b/src/test/modules/injection_points/specs/repack_deadlock.spec
new file mode 100644
index 00000000000..9d23a6588c2
--- /dev/null
+++ b/src/test/modules/injection_points/specs/repack_deadlock.spec
@@ -0,0 +1,83 @@
+# Test REPACK with a concurrent transaction that would cause a deadlock
+setup
+{
+	CREATE EXTENSION injection_points;
+
+	CREATE TABLE repack_deadlock(i int PRIMARY KEY, j int);
+	INSERT INTO repack_deadlock(i, j) VALUES (1, 1), (2, 2), (3, 3), (4, 4);
+
+	CREATE TABLE relfilenodes(node oid);
+
+	CREATE TABLE data_s1(i int, j int);
+	CREATE TABLE data_s2(i int, j int);
+}
+
+teardown
+{
+	DROP TABLE repack_deadlock;
+	DROP EXTENSION injection_points;
+
+	DROP TABLE relfilenodes;
+	DROP TABLE data_s1;
+	DROP TABLE data_s2;
+}
+
+session s1
+setup
+{
+	SELECT injection_points_set_local();
+	SELECT injection_points_attach('repack-concurrently-before-lock', 'wait');
+}
+# Perform the initial load and wait for s2 to do some data changes.
+step wait_before_lock
+{
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+}
+# Check the table from the perspective of s1.
+#
+# Besides the contents, we also check that relfilenode has changed.
+
+# Have each session write the contents into a table and use FULL JOIN to check
+# if the outputs are identical.
+step check1
+{
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+}
+teardown
+{
+	SELECT injection_points_detach('repack-concurrently-before-lock');
+}
+
+session s2
+# Change the existing data. UPDATE changes both key and non-key columns. Also
+# update one row twice to test whether tuple version generated by this session
+# can be found.
+step add_column
+{
+	alter table repack_deadlock add column noise text;
+}
+
+step wakeup_before_lock
+{
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+}
+
+# Test if data changes introduced while one session is performing REPACK
+# CONCURRENTLY find their way into the table.
+permutation
+	wait_before_lock
+	add_column
+	wakeup_before_lock
+	check1
-- 
2.47.3


--qfkt2ktdpcfeypib
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v48-0007-Teach-snapshot-builder-to-skip-transactions-runn.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread

* [PATCH v45 7/8] Error out any process that would block at REPACK
@ 2026-03-25 19:35 Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Álvaro Herrera @ 2026-03-25 19:35 UTC (permalink / raw)

Any process waiting on REPACK to release its lock would actually cause
it to deadlock when it tries to upgrade its lock to AEL, losing all work
done to that point.  We avoid this by teaching the deadlock detector to
raise an error when this condition is detected.
---
 src/backend/commands/repack.c                 | 53 ++++++++----
 src/backend/storage/lmgr/deadlock.c           | 15 ++++
 src/include/storage/proc.h                    |  6 +-
 src/test/modules/injection_points/Makefile    |  1 +
 .../expected/repack_deadlock.out              | 63 ++++++++++++++
 src/test/modules/injection_points/meson.build |  1 +
 .../specs/repack_deadlock.spec                | 83 +++++++++++++++++++
 7 files changed, 202 insertions(+), 20 deletions(-)
 create mode 100644 src/test/modules/injection_points/expected/repack_deadlock.out
 create mode 100644 src/test/modules/injection_points/specs/repack_deadlock.spec

diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c
index d25ab43da7b..7291b4180ab 100644
--- a/src/backend/commands/repack.c
+++ b/src/backend/commands/repack.c
@@ -276,6 +276,16 @@ ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
 	/* Determine the lock mode to use. */
 	lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
 
+	/*
+	 * If in concurrent mode, set the PROC_IN_CONCURRENT_REPACK flag.  This
+	 * makes the deadlock checker cause anyone that would conflict with us
+	 * to error out.  It's important to set this flag ahead of actually locking
+	 * the relation; it won't of course affect anyone until we do have a lock
+	 * that others can conflict with.
+	 */
+	if ((params.options & CLUOPT_CONCURRENT) != 0)
+		MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK;
+
 	/*
 	 * If a single relation is specified, process it and we're done ... unless
 	 * the relation is a partitioned table, in which case we fall through.
@@ -476,11 +486,8 @@ RepackLockLevel(bool concurrent)
  * If indexOid is InvalidOid, the table will be rewritten in physical order
  * instead of index order.
  *
- * Note that, in the concurrent case, the function releases the lock at some
- * point, in order to get AccessExclusiveLock for the final steps (i.e. to
- * swap the relation files). To make things simpler, the caller should expect
- * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
- * AccessExclusiveLock is kept till the end of the transaction.)
+ * On return, OldHeap is closed but locked with AccessExclusiveLock - the lock
+ * will be released at end of the transaction.
  *
  * 'cmd' indicates which command is being executed, to be used for error
  * messages.
@@ -512,10 +519,12 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
 		/*
 		 * Make sure we're not in a transaction block.
 		 *
-		 * The reason is that repack_setup_logical_decoding() could deadlock
-		 * if there's an XID already assigned.  It would be possible to run in
-		 * a transaction block if we had no XID, but this restriction is
-		 * simpler for users to understand and we don't lose anything.
+		 * The reason is that repack_setup_logical_decoding() could wait
+		 * indefinitely for our XID to complete. (The deadlock detector would
+		 * not recognize it because we'd be waiting for ourselves, i.e. no
+		 * real lock conflict.) It would be possible to run in a transaction
+		 * block if we had no XID, but this restriction is simpler for users
+		 * to understand and we don't lose anything.
 		 */
 		PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
 
@@ -998,10 +1007,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose,
 		 * Note that the worker has to wait for all transactions with XID
 		 * already assigned to finish. If some of those transactions is
 		 * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
-		 * table (e.g.  it runs CREATE INDEX), we can end up in a deadlock.
-		 * Not sure this risk is worth unlocking/locking the table (and its
-		 * clustering index) and checking again if it's still eligible for
-		 * REPACK CONCURRENTLY.
+		 * table (e.g. it runs CREATE INDEX), it should encounter ERROR in the
+		 * deadlock checking code.
 		 */
 		start_repack_decoding_worker(tableOid);
 
@@ -3119,10 +3126,18 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
 		LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
 
 	/*
-	 * Tuples and pages of the old heap will be gone, but the heap will stay.
+	 * Now that we have all access-exclusive locks on all relations, we no
+	 * longer want other processes to error out when trying to acquire a
+	 * conflicting lock.  Therefore, unset our flag.
+	 */
+	MyProc->statusFlags &= ~PROC_IN_CONCURRENT_REPACK;
+
+	/*
+	 * Tuples and pages of the old heap will be gone, but the heap itself will
+	 * stay.  In order for predicate locks to continue to work, convert them
+	 * to relation-level locks.  We do this both for table and indexes.
 	 */
 	TransferPredicateLocksToHeapRelation(OldHeap);
-	/* The same for indexes. */
 	for (int i = 0; i < nind; i++)
 	{
 		Relation	index = ind_refs[i];
@@ -3342,9 +3357,11 @@ start_repack_decoding_worker(Oid relid)
 
 	/*
 	 * The decoding setup must be done before the caller can have XID assigned
-	 * for any reason, otherwise the worker might end up in a deadlock,
-	 * waiting for the caller's transaction to end. Therefore wait here until
-	 * the worker indicates that it has the logical decoding initialized.
+	 * for any reason, otherwise the worker might end up waiting for the
+	 * caller's transaction to end. (Deadlock detector does not consider this
+	 * a conflict because the worker is in the same locking group as the
+	 * backend that launched it.) Therefore wait here until the worker
+	 * indicates that it has the logical decoding initialized.
 	 */
 	ConditionVariablePrepareToSleep(&shared->cv);
 	for (;;)
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index b8962d875b6..3d3ec0da111 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -620,6 +620,21 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 						proc->statusFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
+					/*
+					 * Similarly, if we note that we're blocked by some process
+					 * running REPACK (CONCURRENTLY), just fail.  That process
+					 * is going to upgrade its lock at some point, and it would
+					 * be inappropriate for any other process to cause that
+					 * to fail.
+					 */
+					if (checkProc == MyProc &&
+						proc->statusFlags & PROC_IN_CONCURRENT_REPACK)
+						ereport(ERROR,
+								errcode(ERRCODE_OBJECT_IN_USE),
+								errmsg("could not wait for concurrent REPACK"),
+								errdetail("Process %d waits for REPACK running on process %d",
+									   MyProc->pid, proc->pid));
+
 					/* We're done looking at this proclock */
 					break;
 				}
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1dad125706e..8ad9718f3d6 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -69,10 +69,12 @@ struct XidCache
 #define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be
 												 * included in vacuum horizons
 												 * in all databases */
+#define		PROC_IN_CONCURRENT_REPACK	0x40	/* REPACK (CONCURRENTLY) */
 
-/* flags reset at EOXact */
+/* flags reset at EOXact.  A bit of a misnomer ... */
 #define		PROC_VACUUM_STATE_MASK \
-	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)
+	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | \
+	 PROC_IN_CONCURRENT_REPACK)
 
 /*
  * Xmin-related flags. Make sure any flags that affect how the process' Xmin
diff --git a/src/test/modules/injection_points/Makefile b/src/test/modules/injection_points/Makefile
index 2cd7d87c533..f7663859fe2 100644
--- a/src/test/modules/injection_points/Makefile
+++ b/src/test/modules/injection_points/Makefile
@@ -15,6 +15,7 @@ REGRESS_OPTS = --dlpath=$(top_builddir)/src/test/regress
 ISOLATION = basic \
 	    inplace \
 	    repack \
+	    repack_deadlock \
 	    repack_toast \
 	    syscache-update-pruned \
 	    heap_lock_update
diff --git a/src/test/modules/injection_points/expected/repack_deadlock.out b/src/test/modules/injection_points/expected/repack_deadlock.out
new file mode 100644
index 00000000000..a86e4767536
--- /dev/null
+++ b/src/test/modules/injection_points/expected/repack_deadlock.out
@@ -0,0 +1,63 @@
+Parsed test spec with 2 sessions
+
+starting permutation: wait_before_lock add_column wakeup_before_lock check1
+injection_points_attach
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: 
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+ <waiting ...>
+step add_column: 
+	alter table repack_deadlock add column noise text;
+ <waiting ...>
+step add_column: <... completed>
+ERROR:  could not wait for concurrent REPACK
+step wakeup_before_lock: 
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+
+injection_points_wakeup
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: <... completed>
+step check1: 
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+
+count
+-----
+    1
+(1 row)
+
+i|j
+-+-
+1|1
+2|2
+3|3
+4|4
+(4 rows)
+
+count
+-----
+    4
+(1 row)
+
+injection_points_detach
+-----------------------
+                       
+(1 row)
+
diff --git a/src/test/modules/injection_points/meson.build b/src/test/modules/injection_points/meson.build
index a414abb924b..1cd88d6db65 100644
--- a/src/test/modules/injection_points/meson.build
+++ b/src/test/modules/injection_points/meson.build
@@ -46,6 +46,7 @@ tests += {
       'basic',
       'inplace',
       'repack',
+      'repack_deadlock',
       'repack_toast',
       'syscache-update-pruned',
       'heap_lock_update',
diff --git a/src/test/modules/injection_points/specs/repack_deadlock.spec b/src/test/modules/injection_points/specs/repack_deadlock.spec
new file mode 100644
index 00000000000..9d23a6588c2
--- /dev/null
+++ b/src/test/modules/injection_points/specs/repack_deadlock.spec
@@ -0,0 +1,83 @@
+# Test REPACK with a concurrent transaction that would cause a deadlock
+setup
+{
+	CREATE EXTENSION injection_points;
+
+	CREATE TABLE repack_deadlock(i int PRIMARY KEY, j int);
+	INSERT INTO repack_deadlock(i, j) VALUES (1, 1), (2, 2), (3, 3), (4, 4);
+
+	CREATE TABLE relfilenodes(node oid);
+
+	CREATE TABLE data_s1(i int, j int);
+	CREATE TABLE data_s2(i int, j int);
+}
+
+teardown
+{
+	DROP TABLE repack_deadlock;
+	DROP EXTENSION injection_points;
+
+	DROP TABLE relfilenodes;
+	DROP TABLE data_s1;
+	DROP TABLE data_s2;
+}
+
+session s1
+setup
+{
+	SELECT injection_points_set_local();
+	SELECT injection_points_attach('repack-concurrently-before-lock', 'wait');
+}
+# Perform the initial load and wait for s2 to do some data changes.
+step wait_before_lock
+{
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+}
+# Check the table from the perspective of s1.
+#
+# Besides the contents, we also check that relfilenode has changed.
+
+# Have each session write the contents into a table and use FULL JOIN to check
+# if the outputs are identical.
+step check1
+{
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+}
+teardown
+{
+	SELECT injection_points_detach('repack-concurrently-before-lock');
+}
+
+session s2
+# Change the existing data. UPDATE changes both key and non-key columns. Also
+# update one row twice to test whether tuple version generated by this session
+# can be found.
+step add_column
+{
+	alter table repack_deadlock add column noise text;
+}
+
+step wakeup_before_lock
+{
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+}
+
+# Test if data changes introduced while one session is performing REPACK
+# CONCURRENTLY find their way into the table.
+permutation
+	wait_before_lock
+	add_column
+	wakeup_before_lock
+	check1
-- 
2.47.3


--cldir6umqisr673d
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v45-0008-Teach-snapshot-builder-to-skip-transactions-runn.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread

* [PATCH v46 6/7] Error out any process that would block at REPACK
@ 2026-03-25 19:35 Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Álvaro Herrera @ 2026-03-25 19:35 UTC (permalink / raw)

Any process waiting on REPACK to release its lock would actually cause
it to deadlock when it tries to upgrade its lock to AEL, losing all work
done to that point.  We avoid this by teaching the deadlock detector to
raise an error when this condition is detected.
---
 src/backend/commands/repack.c                 | 53 ++++++++----
 src/backend/storage/lmgr/deadlock.c           | 15 ++++
 src/include/storage/proc.h                    |  6 +-
 src/test/modules/injection_points/Makefile    |  1 +
 .../expected/repack_deadlock.out              | 63 ++++++++++++++
 src/test/modules/injection_points/meson.build |  1 +
 .../specs/repack_deadlock.spec                | 83 +++++++++++++++++++
 7 files changed, 202 insertions(+), 20 deletions(-)
 create mode 100644 src/test/modules/injection_points/expected/repack_deadlock.out
 create mode 100644 src/test/modules/injection_points/specs/repack_deadlock.spec

diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c
index a58277bb6c1..21ab81c6f7f 100644
--- a/src/backend/commands/repack.c
+++ b/src/backend/commands/repack.c
@@ -276,6 +276,16 @@ ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
 	/* Determine the lock mode to use. */
 	lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
 
+	/*
+	 * If in concurrent mode, set the PROC_IN_CONCURRENT_REPACK flag.  This
+	 * makes the deadlock checker cause anyone that would conflict with us
+	 * to error out.  It's important to set this flag ahead of actually locking
+	 * the relation; it won't of course affect anyone until we do have a lock
+	 * that others can conflict with.
+	 */
+	if ((params.options & CLUOPT_CONCURRENT) != 0)
+		MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK;
+
 	/*
 	 * If a single relation is specified, process it and we're done ... unless
 	 * the relation is a partitioned table, in which case we fall through.
@@ -476,11 +486,8 @@ RepackLockLevel(bool concurrent)
  * If indexOid is InvalidOid, the table will be rewritten in physical order
  * instead of index order.
  *
- * Note that, in the concurrent case, the function releases the lock at some
- * point, in order to get AccessExclusiveLock for the final steps (i.e. to
- * swap the relation files). To make things simpler, the caller should expect
- * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
- * AccessExclusiveLock is kept till the end of the transaction.)
+ * On return, OldHeap is closed but locked with AccessExclusiveLock - the lock
+ * will be released at end of the transaction.
  *
  * 'cmd' indicates which command is being executed, to be used for error
  * messages.
@@ -512,10 +519,12 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
 		/*
 		 * Make sure we're not in a transaction block.
 		 *
-		 * The reason is that repack_setup_logical_decoding() could deadlock
-		 * if there's an XID already assigned.  It would be possible to run in
-		 * a transaction block if we had no XID, but this restriction is
-		 * simpler for users to understand and we don't lose anything.
+		 * The reason is that repack_setup_logical_decoding() could wait
+		 * indefinitely for our XID to complete. (The deadlock detector would
+		 * not recognize it because we'd be waiting for ourselves, i.e. no
+		 * real lock conflict.) It would be possible to run in a transaction
+		 * block if we had no XID, but this restriction is simpler for users
+		 * to understand and we don't lose anything.
 		 */
 		PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
 
@@ -998,10 +1007,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose,
 		 * Note that the worker has to wait for all transactions with XID
 		 * already assigned to finish. If some of those transactions is
 		 * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
-		 * table (e.g.  it runs CREATE INDEX), we can end up in a deadlock.
-		 * Not sure this risk is worth unlocking/locking the table (and its
-		 * clustering index) and checking again if it's still eligible for
-		 * REPACK CONCURRENTLY.
+		 * table (e.g. it runs CREATE INDEX), it should encounter ERROR in the
+		 * deadlock checking code.
 		 */
 		start_repack_decoding_worker(tableOid);
 
@@ -3123,10 +3130,18 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
 		LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
 
 	/*
-	 * Tuples and pages of the old heap will be gone, but the heap will stay.
+	 * Now that we have all access-exclusive locks on all relations, we no
+	 * longer want other processes to error out when trying to acquire a
+	 * conflicting lock.  Therefore, unset our flag.
+	 */
+	MyProc->statusFlags &= ~PROC_IN_CONCURRENT_REPACK;
+
+	/*
+	 * Tuples and pages of the old heap will be gone, but the heap itself will
+	 * stay.  In order for predicate locks to continue to work, convert them
+	 * to relation-level locks.  We do this both for table and indexes.
 	 */
 	TransferPredicateLocksToHeapRelation(OldHeap);
-	/* The same for indexes. */
 	for (int i = 0; i < nind; i++)
 	{
 		Relation	index = ind_refs[i];
@@ -3346,9 +3361,11 @@ start_repack_decoding_worker(Oid relid)
 
 	/*
 	 * The decoding setup must be done before the caller can have XID assigned
-	 * for any reason, otherwise the worker might end up in a deadlock,
-	 * waiting for the caller's transaction to end. Therefore wait here until
-	 * the worker indicates that it has the logical decoding initialized.
+	 * for any reason, otherwise the worker might end up waiting for the
+	 * caller's transaction to end. (Deadlock detector does not consider this
+	 * a conflict because the worker is in the same locking group as the
+	 * backend that launched it.) Therefore wait here until the worker
+	 * indicates that it has the logical decoding initialized.
 	 */
 	ConditionVariablePrepareToSleep(&shared->cv);
 	for (;;)
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index b8962d875b6..3d3ec0da111 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -620,6 +620,21 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 						proc->statusFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
+					/*
+					 * Similarly, if we note that we're blocked by some process
+					 * running REPACK (CONCURRENTLY), just fail.  That process
+					 * is going to upgrade its lock at some point, and it would
+					 * be inappropriate for any other process to cause that
+					 * to fail.
+					 */
+					if (checkProc == MyProc &&
+						proc->statusFlags & PROC_IN_CONCURRENT_REPACK)
+						ereport(ERROR,
+								errcode(ERRCODE_OBJECT_IN_USE),
+								errmsg("could not wait for concurrent REPACK"),
+								errdetail("Process %d waits for REPACK running on process %d",
+									   MyProc->pid, proc->pid));
+
 					/* We're done looking at this proclock */
 					break;
 				}
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1dad125706e..8ad9718f3d6 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -69,10 +69,12 @@ struct XidCache
 #define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be
 												 * included in vacuum horizons
 												 * in all databases */
+#define		PROC_IN_CONCURRENT_REPACK	0x40	/* REPACK (CONCURRENTLY) */
 
-/* flags reset at EOXact */
+/* flags reset at EOXact.  A bit of a misnomer ... */
 #define		PROC_VACUUM_STATE_MASK \
-	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)
+	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | \
+	 PROC_IN_CONCURRENT_REPACK)
 
 /*
  * Xmin-related flags. Make sure any flags that affect how the process' Xmin
diff --git a/src/test/modules/injection_points/Makefile b/src/test/modules/injection_points/Makefile
index 2cd7d87c533..f7663859fe2 100644
--- a/src/test/modules/injection_points/Makefile
+++ b/src/test/modules/injection_points/Makefile
@@ -15,6 +15,7 @@ REGRESS_OPTS = --dlpath=$(top_builddir)/src/test/regress
 ISOLATION = basic \
 	    inplace \
 	    repack \
+	    repack_deadlock \
 	    repack_toast \
 	    syscache-update-pruned \
 	    heap_lock_update
diff --git a/src/test/modules/injection_points/expected/repack_deadlock.out b/src/test/modules/injection_points/expected/repack_deadlock.out
new file mode 100644
index 00000000000..a86e4767536
--- /dev/null
+++ b/src/test/modules/injection_points/expected/repack_deadlock.out
@@ -0,0 +1,63 @@
+Parsed test spec with 2 sessions
+
+starting permutation: wait_before_lock add_column wakeup_before_lock check1
+injection_points_attach
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: 
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+ <waiting ...>
+step add_column: 
+	alter table repack_deadlock add column noise text;
+ <waiting ...>
+step add_column: <... completed>
+ERROR:  could not wait for concurrent REPACK
+step wakeup_before_lock: 
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+
+injection_points_wakeup
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: <... completed>
+step check1: 
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+
+count
+-----
+    1
+(1 row)
+
+i|j
+-+-
+1|1
+2|2
+3|3
+4|4
+(4 rows)
+
+count
+-----
+    4
+(1 row)
+
+injection_points_detach
+-----------------------
+                       
+(1 row)
+
diff --git a/src/test/modules/injection_points/meson.build b/src/test/modules/injection_points/meson.build
index a414abb924b..1cd88d6db65 100644
--- a/src/test/modules/injection_points/meson.build
+++ b/src/test/modules/injection_points/meson.build
@@ -46,6 +46,7 @@ tests += {
       'basic',
       'inplace',
       'repack',
+      'repack_deadlock',
       'repack_toast',
       'syscache-update-pruned',
       'heap_lock_update',
diff --git a/src/test/modules/injection_points/specs/repack_deadlock.spec b/src/test/modules/injection_points/specs/repack_deadlock.spec
new file mode 100644
index 00000000000..9d23a6588c2
--- /dev/null
+++ b/src/test/modules/injection_points/specs/repack_deadlock.spec
@@ -0,0 +1,83 @@
+# Test REPACK with a concurrent transaction that would cause a deadlock
+setup
+{
+	CREATE EXTENSION injection_points;
+
+	CREATE TABLE repack_deadlock(i int PRIMARY KEY, j int);
+	INSERT INTO repack_deadlock(i, j) VALUES (1, 1), (2, 2), (3, 3), (4, 4);
+
+	CREATE TABLE relfilenodes(node oid);
+
+	CREATE TABLE data_s1(i int, j int);
+	CREATE TABLE data_s2(i int, j int);
+}
+
+teardown
+{
+	DROP TABLE repack_deadlock;
+	DROP EXTENSION injection_points;
+
+	DROP TABLE relfilenodes;
+	DROP TABLE data_s1;
+	DROP TABLE data_s2;
+}
+
+session s1
+setup
+{
+	SELECT injection_points_set_local();
+	SELECT injection_points_attach('repack-concurrently-before-lock', 'wait');
+}
+# Perform the initial load and wait for s2 to do some data changes.
+step wait_before_lock
+{
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+}
+# Check the table from the perspective of s1.
+#
+# Besides the contents, we also check that relfilenode has changed.
+
+# Have each session write the contents into a table and use FULL JOIN to check
+# if the outputs are identical.
+step check1
+{
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+}
+teardown
+{
+	SELECT injection_points_detach('repack-concurrently-before-lock');
+}
+
+session s2
+# Change the existing data. UPDATE changes both key and non-key columns. Also
+# update one row twice to test whether tuple version generated by this session
+# can be found.
+step add_column
+{
+	alter table repack_deadlock add column noise text;
+}
+
+step wakeup_before_lock
+{
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+}
+
+# Test if data changes introduced while one session is performing REPACK
+# CONCURRENTLY find their way into the table.
+permutation
+	wait_before_lock
+	add_column
+	wakeup_before_lock
+	check1
-- 
2.47.3


--mdsv5bdgfjjj6d77
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v46-0007-Teach-snapshot-builder-to-skip-transactions-runn.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread

* [PATCH v45 7/8] Error out any process that would block at REPACK
@ 2026-03-25 19:35 Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Álvaro Herrera @ 2026-03-25 19:35 UTC (permalink / raw)

Any process waiting on REPACK to release its lock would actually cause
it to deadlock when it tries to upgrade its lock to AEL, losing all work
done to that point.  We avoid this by teaching the deadlock detector to
raise an error when this condition is detected.
---
 src/backend/commands/repack.c                 | 53 ++++++++----
 src/backend/storage/lmgr/deadlock.c           | 15 ++++
 src/include/storage/proc.h                    |  6 +-
 src/test/modules/injection_points/Makefile    |  1 +
 .../expected/repack_deadlock.out              | 63 ++++++++++++++
 src/test/modules/injection_points/meson.build |  1 +
 .../specs/repack_deadlock.spec                | 83 +++++++++++++++++++
 7 files changed, 202 insertions(+), 20 deletions(-)
 create mode 100644 src/test/modules/injection_points/expected/repack_deadlock.out
 create mode 100644 src/test/modules/injection_points/specs/repack_deadlock.spec

diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c
index d25ab43da7b..7291b4180ab 100644
--- a/src/backend/commands/repack.c
+++ b/src/backend/commands/repack.c
@@ -276,6 +276,16 @@ ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
 	/* Determine the lock mode to use. */
 	lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
 
+	/*
+	 * If in concurrent mode, set the PROC_IN_CONCURRENT_REPACK flag.  This
+	 * makes the deadlock checker cause anyone that would conflict with us
+	 * to error out.  It's important to set this flag ahead of actually locking
+	 * the relation; it won't of course affect anyone until we do have a lock
+	 * that others can conflict with.
+	 */
+	if ((params.options & CLUOPT_CONCURRENT) != 0)
+		MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK;
+
 	/*
 	 * If a single relation is specified, process it and we're done ... unless
 	 * the relation is a partitioned table, in which case we fall through.
@@ -476,11 +486,8 @@ RepackLockLevel(bool concurrent)
  * If indexOid is InvalidOid, the table will be rewritten in physical order
  * instead of index order.
  *
- * Note that, in the concurrent case, the function releases the lock at some
- * point, in order to get AccessExclusiveLock for the final steps (i.e. to
- * swap the relation files). To make things simpler, the caller should expect
- * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
- * AccessExclusiveLock is kept till the end of the transaction.)
+ * On return, OldHeap is closed but locked with AccessExclusiveLock - the lock
+ * will be released at end of the transaction.
  *
  * 'cmd' indicates which command is being executed, to be used for error
  * messages.
@@ -512,10 +519,12 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
 		/*
 		 * Make sure we're not in a transaction block.
 		 *
-		 * The reason is that repack_setup_logical_decoding() could deadlock
-		 * if there's an XID already assigned.  It would be possible to run in
-		 * a transaction block if we had no XID, but this restriction is
-		 * simpler for users to understand and we don't lose anything.
+		 * The reason is that repack_setup_logical_decoding() could wait
+		 * indefinitely for our XID to complete. (The deadlock detector would
+		 * not recognize it because we'd be waiting for ourselves, i.e. no
+		 * real lock conflict.) It would be possible to run in a transaction
+		 * block if we had no XID, but this restriction is simpler for users
+		 * to understand and we don't lose anything.
 		 */
 		PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
 
@@ -998,10 +1007,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose,
 		 * Note that the worker has to wait for all transactions with XID
 		 * already assigned to finish. If some of those transactions is
 		 * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
-		 * table (e.g.  it runs CREATE INDEX), we can end up in a deadlock.
-		 * Not sure this risk is worth unlocking/locking the table (and its
-		 * clustering index) and checking again if it's still eligible for
-		 * REPACK CONCURRENTLY.
+		 * table (e.g. it runs CREATE INDEX), it should encounter ERROR in the
+		 * deadlock checking code.
 		 */
 		start_repack_decoding_worker(tableOid);
 
@@ -3119,10 +3126,18 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
 		LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
 
 	/*
-	 * Tuples and pages of the old heap will be gone, but the heap will stay.
+	 * Now that we have all access-exclusive locks on all relations, we no
+	 * longer want other processes to error out when trying to acquire a
+	 * conflicting lock.  Therefore, unset our flag.
+	 */
+	MyProc->statusFlags &= ~PROC_IN_CONCURRENT_REPACK;
+
+	/*
+	 * Tuples and pages of the old heap will be gone, but the heap itself will
+	 * stay.  In order for predicate locks to continue to work, convert them
+	 * to relation-level locks.  We do this both for table and indexes.
 	 */
 	TransferPredicateLocksToHeapRelation(OldHeap);
-	/* The same for indexes. */
 	for (int i = 0; i < nind; i++)
 	{
 		Relation	index = ind_refs[i];
@@ -3342,9 +3357,11 @@ start_repack_decoding_worker(Oid relid)
 
 	/*
 	 * The decoding setup must be done before the caller can have XID assigned
-	 * for any reason, otherwise the worker might end up in a deadlock,
-	 * waiting for the caller's transaction to end. Therefore wait here until
-	 * the worker indicates that it has the logical decoding initialized.
+	 * for any reason, otherwise the worker might end up waiting for the
+	 * caller's transaction to end. (Deadlock detector does not consider this
+	 * a conflict because the worker is in the same locking group as the
+	 * backend that launched it.) Therefore wait here until the worker
+	 * indicates that it has the logical decoding initialized.
 	 */
 	ConditionVariablePrepareToSleep(&shared->cv);
 	for (;;)
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index b8962d875b6..3d3ec0da111 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -620,6 +620,21 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 						proc->statusFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
+					/*
+					 * Similarly, if we note that we're blocked by some process
+					 * running REPACK (CONCURRENTLY), just fail.  That process
+					 * is going to upgrade its lock at some point, and it would
+					 * be inappropriate for any other process to cause that
+					 * to fail.
+					 */
+					if (checkProc == MyProc &&
+						proc->statusFlags & PROC_IN_CONCURRENT_REPACK)
+						ereport(ERROR,
+								errcode(ERRCODE_OBJECT_IN_USE),
+								errmsg("could not wait for concurrent REPACK"),
+								errdetail("Process %d waits for REPACK running on process %d",
+									   MyProc->pid, proc->pid));
+
 					/* We're done looking at this proclock */
 					break;
 				}
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1dad125706e..8ad9718f3d6 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -69,10 +69,12 @@ struct XidCache
 #define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be
 												 * included in vacuum horizons
 												 * in all databases */
+#define		PROC_IN_CONCURRENT_REPACK	0x40	/* REPACK (CONCURRENTLY) */
 
-/* flags reset at EOXact */
+/* flags reset at EOXact.  A bit of a misnomer ... */
 #define		PROC_VACUUM_STATE_MASK \
-	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)
+	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | \
+	 PROC_IN_CONCURRENT_REPACK)
 
 /*
  * Xmin-related flags. Make sure any flags that affect how the process' Xmin
diff --git a/src/test/modules/injection_points/Makefile b/src/test/modules/injection_points/Makefile
index 2cd7d87c533..f7663859fe2 100644
--- a/src/test/modules/injection_points/Makefile
+++ b/src/test/modules/injection_points/Makefile
@@ -15,6 +15,7 @@ REGRESS_OPTS = --dlpath=$(top_builddir)/src/test/regress
 ISOLATION = basic \
 	    inplace \
 	    repack \
+	    repack_deadlock \
 	    repack_toast \
 	    syscache-update-pruned \
 	    heap_lock_update
diff --git a/src/test/modules/injection_points/expected/repack_deadlock.out b/src/test/modules/injection_points/expected/repack_deadlock.out
new file mode 100644
index 00000000000..a86e4767536
--- /dev/null
+++ b/src/test/modules/injection_points/expected/repack_deadlock.out
@@ -0,0 +1,63 @@
+Parsed test spec with 2 sessions
+
+starting permutation: wait_before_lock add_column wakeup_before_lock check1
+injection_points_attach
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: 
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+ <waiting ...>
+step add_column: 
+	alter table repack_deadlock add column noise text;
+ <waiting ...>
+step add_column: <... completed>
+ERROR:  could not wait for concurrent REPACK
+step wakeup_before_lock: 
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+
+injection_points_wakeup
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: <... completed>
+step check1: 
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+
+count
+-----
+    1
+(1 row)
+
+i|j
+-+-
+1|1
+2|2
+3|3
+4|4
+(4 rows)
+
+count
+-----
+    4
+(1 row)
+
+injection_points_detach
+-----------------------
+                       
+(1 row)
+
diff --git a/src/test/modules/injection_points/meson.build b/src/test/modules/injection_points/meson.build
index a414abb924b..1cd88d6db65 100644
--- a/src/test/modules/injection_points/meson.build
+++ b/src/test/modules/injection_points/meson.build
@@ -46,6 +46,7 @@ tests += {
       'basic',
       'inplace',
       'repack',
+      'repack_deadlock',
       'repack_toast',
       'syscache-update-pruned',
       'heap_lock_update',
diff --git a/src/test/modules/injection_points/specs/repack_deadlock.spec b/src/test/modules/injection_points/specs/repack_deadlock.spec
new file mode 100644
index 00000000000..9d23a6588c2
--- /dev/null
+++ b/src/test/modules/injection_points/specs/repack_deadlock.spec
@@ -0,0 +1,83 @@
+# Test REPACK with a concurrent transaction that would cause a deadlock
+setup
+{
+	CREATE EXTENSION injection_points;
+
+	CREATE TABLE repack_deadlock(i int PRIMARY KEY, j int);
+	INSERT INTO repack_deadlock(i, j) VALUES (1, 1), (2, 2), (3, 3), (4, 4);
+
+	CREATE TABLE relfilenodes(node oid);
+
+	CREATE TABLE data_s1(i int, j int);
+	CREATE TABLE data_s2(i int, j int);
+}
+
+teardown
+{
+	DROP TABLE repack_deadlock;
+	DROP EXTENSION injection_points;
+
+	DROP TABLE relfilenodes;
+	DROP TABLE data_s1;
+	DROP TABLE data_s2;
+}
+
+session s1
+setup
+{
+	SELECT injection_points_set_local();
+	SELECT injection_points_attach('repack-concurrently-before-lock', 'wait');
+}
+# Perform the initial load and wait for s2 to do some data changes.
+step wait_before_lock
+{
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+}
+# Check the table from the perspective of s1.
+#
+# Besides the contents, we also check that relfilenode has changed.
+
+# Have each session write the contents into a table and use FULL JOIN to check
+# if the outputs are identical.
+step check1
+{
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+}
+teardown
+{
+	SELECT injection_points_detach('repack-concurrently-before-lock');
+}
+
+session s2
+# Change the existing data. UPDATE changes both key and non-key columns. Also
+# update one row twice to test whether tuple version generated by this session
+# can be found.
+step add_column
+{
+	alter table repack_deadlock add column noise text;
+}
+
+step wakeup_before_lock
+{
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+}
+
+# Test if data changes introduced while one session is performing REPACK
+# CONCURRENTLY find their way into the table.
+permutation
+	wait_before_lock
+	add_column
+	wakeup_before_lock
+	check1
-- 
2.47.3


--cldir6umqisr673d
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v45-0008-Teach-snapshot-builder-to-skip-transactions-runn.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread

* [PATCH v48 6/7] Error out any process that would block at REPACK
@ 2026-03-25 19:35 Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Álvaro Herrera @ 2026-03-25 19:35 UTC (permalink / raw)

Any process waiting on REPACK to release its lock would actually cause
it to deadlock when it tries to upgrade its lock to AEL, losing all work
done to that point.  We avoid this by teaching the deadlock detector to
raise an error when this condition is detected.
---
 src/backend/commands/repack.c                 | 52 ++++++++----
 src/backend/storage/lmgr/deadlock.c           | 15 ++++
 src/include/storage/proc.h                    |  6 +-
 src/test/modules/injection_points/Makefile    |  1 +
 .../expected/repack_deadlock.out              | 63 ++++++++++++++
 src/test/modules/injection_points/meson.build |  1 +
 .../specs/repack_deadlock.spec                | 83 +++++++++++++++++++
 7 files changed, 202 insertions(+), 19 deletions(-)
 create mode 100644 src/test/modules/injection_points/expected/repack_deadlock.out
 create mode 100644 src/test/modules/injection_points/specs/repack_deadlock.spec

diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c
index a514053b777..781a471fc1d 100644
--- a/src/backend/commands/repack.c
+++ b/src/backend/commands/repack.c
@@ -276,6 +276,16 @@ ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
 	/* Determine the lock mode to use. */
 	lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
 
+	/*
+	 * If in concurrent mode, set the PROC_IN_CONCURRENT_REPACK flag.  This
+	 * makes the deadlock checker cause anyone that would conflict with us to
+	 * error out.  It's important to set this flag ahead of actually locking
+	 * the relation; it won't of course affect anyone until we do have a lock
+	 * that others can conflict with.
+	 */
+	if ((params.options & CLUOPT_CONCURRENT) != 0)
+		MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK;
+
 	/*
 	 * If a single relation is specified, process it and we're done ... unless
 	 * the relation is a partitioned table, in which case we fall through.
@@ -476,11 +486,8 @@ RepackLockLevel(bool concurrent)
  * If indexOid is InvalidOid, the table will be rewritten in physical order
  * instead of index order.
  *
- * Note that, in the concurrent case, the function releases the lock at some
- * point, in order to get AccessExclusiveLock for the final steps (i.e. to
- * swap the relation files). To make things simpler, the caller should expect
- * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
- * AccessExclusiveLock is kept till the end of the transaction.)
+ * On return, OldHeap is closed but locked with AccessExclusiveLock - the lock
+ * will be released at end of the transaction.
  *
  * 'cmd' indicates which command is being executed, to be used for error
  * messages.
@@ -512,10 +519,12 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
 		/*
 		 * Make sure we're not in a transaction block.
 		 *
-		 * The reason is that repack_setup_logical_decoding() could deadlock
-		 * if there's an XID already assigned.  It would be possible to run in
-		 * a transaction block if we had no XID, but this restriction is
-		 * simpler for users to understand and we don't lose anything.
+		 * The reason is that repack_setup_logical_decoding() could wait
+		 * indefinitely for our XID to complete. (The deadlock detector would
+		 * not recognize it because we'd be waiting for ourselves, i.e. no
+		 * real lock conflict.) It would be possible to run in a transaction
+		 * block if we had no XID, but this restriction is simpler for users
+		 * to understand and we don't lose anything.
 		 */
 		PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
 
@@ -998,10 +1007,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose,
 		 * Note that the worker has to wait for all transactions with XID
 		 * already assigned to finish. If some of those transactions is
 		 * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
-		 * table (e.g.  it runs CREATE INDEX), we can end up in a deadlock.
-		 * Not sure this risk is worth unlocking/locking the table (and its
-		 * clustering index) and checking again if it's still eligible for
-		 * REPACK CONCURRENTLY.
+		 * table (e.g. it runs CREATE INDEX), it should encounter ERROR in the
+		 * deadlock checking code.
 		 */
 		start_repack_decoding_worker(tableOid);
 
@@ -3090,7 +3097,16 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
 		LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
 
 	/*
-	 * Tuples and pages of the old heap will be gone, but the heap will stay.
+	 * Now that we have all access-exclusive locks on all relations, we no
+	 * longer want other processes to error out when trying to acquire a
+	 * conflicting lock.  Therefore, unset our flag.
+	 */
+	MyProc->statusFlags &= ~PROC_IN_CONCURRENT_REPACK;
+
+	/*
+	 * Tuples and pages of the old heap will be gone, but the heap itself will
+	 * stay.  In order for predicate locks to continue to work, convert them
+	 * to relation-level locks.  We do this both for table and indexes.
 	 */
 	TransferPredicateLocksToHeapRelation(OldHeap);
 	foreach_ptr(RelationData, index, indexrels)
@@ -3288,9 +3304,11 @@ start_repack_decoding_worker(Oid relid)
 
 	/*
 	 * The decoding setup must be done before the caller can have XID assigned
-	 * for any reason, otherwise the worker might end up in a deadlock,
-	 * waiting for the caller's transaction to end. Therefore wait here until
-	 * the worker indicates that it has the logical decoding initialized.
+	 * for any reason, otherwise the worker might end up waiting for the
+	 * caller's transaction to end. (Deadlock detector does not consider this
+	 * a conflict because the worker is in the same locking group as the
+	 * backend that launched it.) Therefore wait here until the worker
+	 * indicates that it has the logical decoding initialized.
 	 */
 	ConditionVariablePrepareToSleep(&shared->cv);
 	for (;;)
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index b8962d875b6..c20ac682b0d 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -620,6 +620,21 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 						proc->statusFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
+					/*
+					 * Similarly, if we note that we're blocked by some
+					 * process running REPACK (CONCURRENTLY), just fail.  That
+					 * process is going to upgrade its lock at some point, and
+					 * it would be inappropriate for any other process to
+					 * cause that to fail.
+					 */
+					if (checkProc == MyProc &&
+						proc->statusFlags & PROC_IN_CONCURRENT_REPACK)
+						ereport(ERROR,
+								errcode(ERRCODE_OBJECT_IN_USE),
+								errmsg("could not wait for concurrent REPACK"),
+								errdetail("Process %d waits for REPACK running on process %d",
+										  MyProc->pid, proc->pid));
+
 					/* We're done looking at this proclock */
 					break;
 				}
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1dad125706e..8ad9718f3d6 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -69,10 +69,12 @@ struct XidCache
 #define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be
 												 * included in vacuum horizons
 												 * in all databases */
+#define		PROC_IN_CONCURRENT_REPACK	0x40	/* REPACK (CONCURRENTLY) */
 
-/* flags reset at EOXact */
+/* flags reset at EOXact.  A bit of a misnomer ... */
 #define		PROC_VACUUM_STATE_MASK \
-	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)
+	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | \
+	 PROC_IN_CONCURRENT_REPACK)
 
 /*
  * Xmin-related flags. Make sure any flags that affect how the process' Xmin
diff --git a/src/test/modules/injection_points/Makefile b/src/test/modules/injection_points/Makefile
index 2cd7d87c533..f7663859fe2 100644
--- a/src/test/modules/injection_points/Makefile
+++ b/src/test/modules/injection_points/Makefile
@@ -15,6 +15,7 @@ REGRESS_OPTS = --dlpath=$(top_builddir)/src/test/regress
 ISOLATION = basic \
 	    inplace \
 	    repack \
+	    repack_deadlock \
 	    repack_toast \
 	    syscache-update-pruned \
 	    heap_lock_update
diff --git a/src/test/modules/injection_points/expected/repack_deadlock.out b/src/test/modules/injection_points/expected/repack_deadlock.out
new file mode 100644
index 00000000000..a86e4767536
--- /dev/null
+++ b/src/test/modules/injection_points/expected/repack_deadlock.out
@@ -0,0 +1,63 @@
+Parsed test spec with 2 sessions
+
+starting permutation: wait_before_lock add_column wakeup_before_lock check1
+injection_points_attach
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: 
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+ <waiting ...>
+step add_column: 
+	alter table repack_deadlock add column noise text;
+ <waiting ...>
+step add_column: <... completed>
+ERROR:  could not wait for concurrent REPACK
+step wakeup_before_lock: 
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+
+injection_points_wakeup
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: <... completed>
+step check1: 
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+
+count
+-----
+    1
+(1 row)
+
+i|j
+-+-
+1|1
+2|2
+3|3
+4|4
+(4 rows)
+
+count
+-----
+    4
+(1 row)
+
+injection_points_detach
+-----------------------
+                       
+(1 row)
+
diff --git a/src/test/modules/injection_points/meson.build b/src/test/modules/injection_points/meson.build
index a414abb924b..1cd88d6db65 100644
--- a/src/test/modules/injection_points/meson.build
+++ b/src/test/modules/injection_points/meson.build
@@ -46,6 +46,7 @@ tests += {
       'basic',
       'inplace',
       'repack',
+      'repack_deadlock',
       'repack_toast',
       'syscache-update-pruned',
       'heap_lock_update',
diff --git a/src/test/modules/injection_points/specs/repack_deadlock.spec b/src/test/modules/injection_points/specs/repack_deadlock.spec
new file mode 100644
index 00000000000..9d23a6588c2
--- /dev/null
+++ b/src/test/modules/injection_points/specs/repack_deadlock.spec
@@ -0,0 +1,83 @@
+# Test REPACK with a concurrent transaction that would cause a deadlock
+setup
+{
+	CREATE EXTENSION injection_points;
+
+	CREATE TABLE repack_deadlock(i int PRIMARY KEY, j int);
+	INSERT INTO repack_deadlock(i, j) VALUES (1, 1), (2, 2), (3, 3), (4, 4);
+
+	CREATE TABLE relfilenodes(node oid);
+
+	CREATE TABLE data_s1(i int, j int);
+	CREATE TABLE data_s2(i int, j int);
+}
+
+teardown
+{
+	DROP TABLE repack_deadlock;
+	DROP EXTENSION injection_points;
+
+	DROP TABLE relfilenodes;
+	DROP TABLE data_s1;
+	DROP TABLE data_s2;
+}
+
+session s1
+setup
+{
+	SELECT injection_points_set_local();
+	SELECT injection_points_attach('repack-concurrently-before-lock', 'wait');
+}
+# Perform the initial load and wait for s2 to do some data changes.
+step wait_before_lock
+{
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+}
+# Check the table from the perspective of s1.
+#
+# Besides the contents, we also check that relfilenode has changed.
+
+# Have each session write the contents into a table and use FULL JOIN to check
+# if the outputs are identical.
+step check1
+{
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+}
+teardown
+{
+	SELECT injection_points_detach('repack-concurrently-before-lock');
+}
+
+session s2
+# Change the existing data. UPDATE changes both key and non-key columns. Also
+# update one row twice to test whether tuple version generated by this session
+# can be found.
+step add_column
+{
+	alter table repack_deadlock add column noise text;
+}
+
+step wakeup_before_lock
+{
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+}
+
+# Test if data changes introduced while one session is performing REPACK
+# CONCURRENTLY find their way into the table.
+permutation
+	wait_before_lock
+	add_column
+	wakeup_before_lock
+	check1
-- 
2.47.3


--qfkt2ktdpcfeypib
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v48-0007-Teach-snapshot-builder-to-skip-transactions-runn.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread

* [PATCH v46 6/7] Error out any process that would block at REPACK
@ 2026-03-25 19:35 Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Álvaro Herrera @ 2026-03-25 19:35 UTC (permalink / raw)

Any process waiting on REPACK to release its lock would actually cause
it to deadlock when it tries to upgrade its lock to AEL, losing all work
done to that point.  We avoid this by teaching the deadlock detector to
raise an error when this condition is detected.
---
 src/backend/commands/repack.c                 | 53 ++++++++----
 src/backend/storage/lmgr/deadlock.c           | 15 ++++
 src/include/storage/proc.h                    |  6 +-
 src/test/modules/injection_points/Makefile    |  1 +
 .../expected/repack_deadlock.out              | 63 ++++++++++++++
 src/test/modules/injection_points/meson.build |  1 +
 .../specs/repack_deadlock.spec                | 83 +++++++++++++++++++
 7 files changed, 202 insertions(+), 20 deletions(-)
 create mode 100644 src/test/modules/injection_points/expected/repack_deadlock.out
 create mode 100644 src/test/modules/injection_points/specs/repack_deadlock.spec

diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c
index a58277bb6c1..21ab81c6f7f 100644
--- a/src/backend/commands/repack.c
+++ b/src/backend/commands/repack.c
@@ -276,6 +276,16 @@ ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
 	/* Determine the lock mode to use. */
 	lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
 
+	/*
+	 * If in concurrent mode, set the PROC_IN_CONCURRENT_REPACK flag.  This
+	 * makes the deadlock checker cause anyone that would conflict with us
+	 * to error out.  It's important to set this flag ahead of actually locking
+	 * the relation; it won't of course affect anyone until we do have a lock
+	 * that others can conflict with.
+	 */
+	if ((params.options & CLUOPT_CONCURRENT) != 0)
+		MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK;
+
 	/*
 	 * If a single relation is specified, process it and we're done ... unless
 	 * the relation is a partitioned table, in which case we fall through.
@@ -476,11 +486,8 @@ RepackLockLevel(bool concurrent)
  * If indexOid is InvalidOid, the table will be rewritten in physical order
  * instead of index order.
  *
- * Note that, in the concurrent case, the function releases the lock at some
- * point, in order to get AccessExclusiveLock for the final steps (i.e. to
- * swap the relation files). To make things simpler, the caller should expect
- * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
- * AccessExclusiveLock is kept till the end of the transaction.)
+ * On return, OldHeap is closed but locked with AccessExclusiveLock - the lock
+ * will be released at end of the transaction.
  *
  * 'cmd' indicates which command is being executed, to be used for error
  * messages.
@@ -512,10 +519,12 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
 		/*
 		 * Make sure we're not in a transaction block.
 		 *
-		 * The reason is that repack_setup_logical_decoding() could deadlock
-		 * if there's an XID already assigned.  It would be possible to run in
-		 * a transaction block if we had no XID, but this restriction is
-		 * simpler for users to understand and we don't lose anything.
+		 * The reason is that repack_setup_logical_decoding() could wait
+		 * indefinitely for our XID to complete. (The deadlock detector would
+		 * not recognize it because we'd be waiting for ourselves, i.e. no
+		 * real lock conflict.) It would be possible to run in a transaction
+		 * block if we had no XID, but this restriction is simpler for users
+		 * to understand and we don't lose anything.
 		 */
 		PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
 
@@ -998,10 +1007,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose,
 		 * Note that the worker has to wait for all transactions with XID
 		 * already assigned to finish. If some of those transactions is
 		 * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
-		 * table (e.g.  it runs CREATE INDEX), we can end up in a deadlock.
-		 * Not sure this risk is worth unlocking/locking the table (and its
-		 * clustering index) and checking again if it's still eligible for
-		 * REPACK CONCURRENTLY.
+		 * table (e.g. it runs CREATE INDEX), it should encounter ERROR in the
+		 * deadlock checking code.
 		 */
 		start_repack_decoding_worker(tableOid);
 
@@ -3123,10 +3130,18 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
 		LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
 
 	/*
-	 * Tuples and pages of the old heap will be gone, but the heap will stay.
+	 * Now that we have all access-exclusive locks on all relations, we no
+	 * longer want other processes to error out when trying to acquire a
+	 * conflicting lock.  Therefore, unset our flag.
+	 */
+	MyProc->statusFlags &= ~PROC_IN_CONCURRENT_REPACK;
+
+	/*
+	 * Tuples and pages of the old heap will be gone, but the heap itself will
+	 * stay.  In order for predicate locks to continue to work, convert them
+	 * to relation-level locks.  We do this both for table and indexes.
 	 */
 	TransferPredicateLocksToHeapRelation(OldHeap);
-	/* The same for indexes. */
 	for (int i = 0; i < nind; i++)
 	{
 		Relation	index = ind_refs[i];
@@ -3346,9 +3361,11 @@ start_repack_decoding_worker(Oid relid)
 
 	/*
 	 * The decoding setup must be done before the caller can have XID assigned
-	 * for any reason, otherwise the worker might end up in a deadlock,
-	 * waiting for the caller's transaction to end. Therefore wait here until
-	 * the worker indicates that it has the logical decoding initialized.
+	 * for any reason, otherwise the worker might end up waiting for the
+	 * caller's transaction to end. (Deadlock detector does not consider this
+	 * a conflict because the worker is in the same locking group as the
+	 * backend that launched it.) Therefore wait here until the worker
+	 * indicates that it has the logical decoding initialized.
 	 */
 	ConditionVariablePrepareToSleep(&shared->cv);
 	for (;;)
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index b8962d875b6..3d3ec0da111 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -620,6 +620,21 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 						proc->statusFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
+					/*
+					 * Similarly, if we note that we're blocked by some process
+					 * running REPACK (CONCURRENTLY), just fail.  That process
+					 * is going to upgrade its lock at some point, and it would
+					 * be inappropriate for any other process to cause that
+					 * to fail.
+					 */
+					if (checkProc == MyProc &&
+						proc->statusFlags & PROC_IN_CONCURRENT_REPACK)
+						ereport(ERROR,
+								errcode(ERRCODE_OBJECT_IN_USE),
+								errmsg("could not wait for concurrent REPACK"),
+								errdetail("Process %d waits for REPACK running on process %d",
+									   MyProc->pid, proc->pid));
+
 					/* We're done looking at this proclock */
 					break;
 				}
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1dad125706e..8ad9718f3d6 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -69,10 +69,12 @@ struct XidCache
 #define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be
 												 * included in vacuum horizons
 												 * in all databases */
+#define		PROC_IN_CONCURRENT_REPACK	0x40	/* REPACK (CONCURRENTLY) */
 
-/* flags reset at EOXact */
+/* flags reset at EOXact.  A bit of a misnomer ... */
 #define		PROC_VACUUM_STATE_MASK \
-	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)
+	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | \
+	 PROC_IN_CONCURRENT_REPACK)
 
 /*
  * Xmin-related flags. Make sure any flags that affect how the process' Xmin
diff --git a/src/test/modules/injection_points/Makefile b/src/test/modules/injection_points/Makefile
index 2cd7d87c533..f7663859fe2 100644
--- a/src/test/modules/injection_points/Makefile
+++ b/src/test/modules/injection_points/Makefile
@@ -15,6 +15,7 @@ REGRESS_OPTS = --dlpath=$(top_builddir)/src/test/regress
 ISOLATION = basic \
 	    inplace \
 	    repack \
+	    repack_deadlock \
 	    repack_toast \
 	    syscache-update-pruned \
 	    heap_lock_update
diff --git a/src/test/modules/injection_points/expected/repack_deadlock.out b/src/test/modules/injection_points/expected/repack_deadlock.out
new file mode 100644
index 00000000000..a86e4767536
--- /dev/null
+++ b/src/test/modules/injection_points/expected/repack_deadlock.out
@@ -0,0 +1,63 @@
+Parsed test spec with 2 sessions
+
+starting permutation: wait_before_lock add_column wakeup_before_lock check1
+injection_points_attach
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: 
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+ <waiting ...>
+step add_column: 
+	alter table repack_deadlock add column noise text;
+ <waiting ...>
+step add_column: <... completed>
+ERROR:  could not wait for concurrent REPACK
+step wakeup_before_lock: 
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+
+injection_points_wakeup
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: <... completed>
+step check1: 
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+
+count
+-----
+    1
+(1 row)
+
+i|j
+-+-
+1|1
+2|2
+3|3
+4|4
+(4 rows)
+
+count
+-----
+    4
+(1 row)
+
+injection_points_detach
+-----------------------
+                       
+(1 row)
+
diff --git a/src/test/modules/injection_points/meson.build b/src/test/modules/injection_points/meson.build
index a414abb924b..1cd88d6db65 100644
--- a/src/test/modules/injection_points/meson.build
+++ b/src/test/modules/injection_points/meson.build
@@ -46,6 +46,7 @@ tests += {
       'basic',
       'inplace',
       'repack',
+      'repack_deadlock',
       'repack_toast',
       'syscache-update-pruned',
       'heap_lock_update',
diff --git a/src/test/modules/injection_points/specs/repack_deadlock.spec b/src/test/modules/injection_points/specs/repack_deadlock.spec
new file mode 100644
index 00000000000..9d23a6588c2
--- /dev/null
+++ b/src/test/modules/injection_points/specs/repack_deadlock.spec
@@ -0,0 +1,83 @@
+# Test REPACK with a concurrent transaction that would cause a deadlock
+setup
+{
+	CREATE EXTENSION injection_points;
+
+	CREATE TABLE repack_deadlock(i int PRIMARY KEY, j int);
+	INSERT INTO repack_deadlock(i, j) VALUES (1, 1), (2, 2), (3, 3), (4, 4);
+
+	CREATE TABLE relfilenodes(node oid);
+
+	CREATE TABLE data_s1(i int, j int);
+	CREATE TABLE data_s2(i int, j int);
+}
+
+teardown
+{
+	DROP TABLE repack_deadlock;
+	DROP EXTENSION injection_points;
+
+	DROP TABLE relfilenodes;
+	DROP TABLE data_s1;
+	DROP TABLE data_s2;
+}
+
+session s1
+setup
+{
+	SELECT injection_points_set_local();
+	SELECT injection_points_attach('repack-concurrently-before-lock', 'wait');
+}
+# Perform the initial load and wait for s2 to do some data changes.
+step wait_before_lock
+{
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+}
+# Check the table from the perspective of s1.
+#
+# Besides the contents, we also check that relfilenode has changed.
+
+# Have each session write the contents into a table and use FULL JOIN to check
+# if the outputs are identical.
+step check1
+{
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+}
+teardown
+{
+	SELECT injection_points_detach('repack-concurrently-before-lock');
+}
+
+session s2
+# Change the existing data. UPDATE changes both key and non-key columns. Also
+# update one row twice to test whether tuple version generated by this session
+# can be found.
+step add_column
+{
+	alter table repack_deadlock add column noise text;
+}
+
+step wakeup_before_lock
+{
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+}
+
+# Test if data changes introduced while one session is performing REPACK
+# CONCURRENTLY find their way into the table.
+permutation
+	wait_before_lock
+	add_column
+	wakeup_before_lock
+	check1
-- 
2.47.3


--mdsv5bdgfjjj6d77
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v46-0007-Teach-snapshot-builder-to-skip-transactions-runn.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread

* [PATCH v46 6/7] Error out any process that would block at REPACK
@ 2026-03-25 19:35 Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Álvaro Herrera @ 2026-03-25 19:35 UTC (permalink / raw)

Any process waiting on REPACK to release its lock would actually cause
it to deadlock when it tries to upgrade its lock to AEL, losing all work
done to that point.  We avoid this by teaching the deadlock detector to
raise an error when this condition is detected.
---
 src/backend/commands/repack.c                 | 53 ++++++++----
 src/backend/storage/lmgr/deadlock.c           | 15 ++++
 src/include/storage/proc.h                    |  6 +-
 src/test/modules/injection_points/Makefile    |  1 +
 .../expected/repack_deadlock.out              | 63 ++++++++++++++
 src/test/modules/injection_points/meson.build |  1 +
 .../specs/repack_deadlock.spec                | 83 +++++++++++++++++++
 7 files changed, 202 insertions(+), 20 deletions(-)
 create mode 100644 src/test/modules/injection_points/expected/repack_deadlock.out
 create mode 100644 src/test/modules/injection_points/specs/repack_deadlock.spec

diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c
index a58277bb6c1..21ab81c6f7f 100644
--- a/src/backend/commands/repack.c
+++ b/src/backend/commands/repack.c
@@ -276,6 +276,16 @@ ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
 	/* Determine the lock mode to use. */
 	lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
 
+	/*
+	 * If in concurrent mode, set the PROC_IN_CONCURRENT_REPACK flag.  This
+	 * makes the deadlock checker cause anyone that would conflict with us
+	 * to error out.  It's important to set this flag ahead of actually locking
+	 * the relation; it won't of course affect anyone until we do have a lock
+	 * that others can conflict with.
+	 */
+	if ((params.options & CLUOPT_CONCURRENT) != 0)
+		MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK;
+
 	/*
 	 * If a single relation is specified, process it and we're done ... unless
 	 * the relation is a partitioned table, in which case we fall through.
@@ -476,11 +486,8 @@ RepackLockLevel(bool concurrent)
  * If indexOid is InvalidOid, the table will be rewritten in physical order
  * instead of index order.
  *
- * Note that, in the concurrent case, the function releases the lock at some
- * point, in order to get AccessExclusiveLock for the final steps (i.e. to
- * swap the relation files). To make things simpler, the caller should expect
- * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
- * AccessExclusiveLock is kept till the end of the transaction.)
+ * On return, OldHeap is closed but locked with AccessExclusiveLock - the lock
+ * will be released at end of the transaction.
  *
  * 'cmd' indicates which command is being executed, to be used for error
  * messages.
@@ -512,10 +519,12 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
 		/*
 		 * Make sure we're not in a transaction block.
 		 *
-		 * The reason is that repack_setup_logical_decoding() could deadlock
-		 * if there's an XID already assigned.  It would be possible to run in
-		 * a transaction block if we had no XID, but this restriction is
-		 * simpler for users to understand and we don't lose anything.
+		 * The reason is that repack_setup_logical_decoding() could wait
+		 * indefinitely for our XID to complete. (The deadlock detector would
+		 * not recognize it because we'd be waiting for ourselves, i.e. no
+		 * real lock conflict.) It would be possible to run in a transaction
+		 * block if we had no XID, but this restriction is simpler for users
+		 * to understand and we don't lose anything.
 		 */
 		PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
 
@@ -998,10 +1007,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose,
 		 * Note that the worker has to wait for all transactions with XID
 		 * already assigned to finish. If some of those transactions is
 		 * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
-		 * table (e.g.  it runs CREATE INDEX), we can end up in a deadlock.
-		 * Not sure this risk is worth unlocking/locking the table (and its
-		 * clustering index) and checking again if it's still eligible for
-		 * REPACK CONCURRENTLY.
+		 * table (e.g. it runs CREATE INDEX), it should encounter ERROR in the
+		 * deadlock checking code.
 		 */
 		start_repack_decoding_worker(tableOid);
 
@@ -3123,10 +3130,18 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
 		LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
 
 	/*
-	 * Tuples and pages of the old heap will be gone, but the heap will stay.
+	 * Now that we have all access-exclusive locks on all relations, we no
+	 * longer want other processes to error out when trying to acquire a
+	 * conflicting lock.  Therefore, unset our flag.
+	 */
+	MyProc->statusFlags &= ~PROC_IN_CONCURRENT_REPACK;
+
+	/*
+	 * Tuples and pages of the old heap will be gone, but the heap itself will
+	 * stay.  In order for predicate locks to continue to work, convert them
+	 * to relation-level locks.  We do this both for table and indexes.
 	 */
 	TransferPredicateLocksToHeapRelation(OldHeap);
-	/* The same for indexes. */
 	for (int i = 0; i < nind; i++)
 	{
 		Relation	index = ind_refs[i];
@@ -3346,9 +3361,11 @@ start_repack_decoding_worker(Oid relid)
 
 	/*
 	 * The decoding setup must be done before the caller can have XID assigned
-	 * for any reason, otherwise the worker might end up in a deadlock,
-	 * waiting for the caller's transaction to end. Therefore wait here until
-	 * the worker indicates that it has the logical decoding initialized.
+	 * for any reason, otherwise the worker might end up waiting for the
+	 * caller's transaction to end. (Deadlock detector does not consider this
+	 * a conflict because the worker is in the same locking group as the
+	 * backend that launched it.) Therefore wait here until the worker
+	 * indicates that it has the logical decoding initialized.
 	 */
 	ConditionVariablePrepareToSleep(&shared->cv);
 	for (;;)
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index b8962d875b6..3d3ec0da111 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -620,6 +620,21 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 						proc->statusFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
+					/*
+					 * Similarly, if we note that we're blocked by some process
+					 * running REPACK (CONCURRENTLY), just fail.  That process
+					 * is going to upgrade its lock at some point, and it would
+					 * be inappropriate for any other process to cause that
+					 * to fail.
+					 */
+					if (checkProc == MyProc &&
+						proc->statusFlags & PROC_IN_CONCURRENT_REPACK)
+						ereport(ERROR,
+								errcode(ERRCODE_OBJECT_IN_USE),
+								errmsg("could not wait for concurrent REPACK"),
+								errdetail("Process %d waits for REPACK running on process %d",
+									   MyProc->pid, proc->pid));
+
 					/* We're done looking at this proclock */
 					break;
 				}
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1dad125706e..8ad9718f3d6 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -69,10 +69,12 @@ struct XidCache
 #define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be
 												 * included in vacuum horizons
 												 * in all databases */
+#define		PROC_IN_CONCURRENT_REPACK	0x40	/* REPACK (CONCURRENTLY) */
 
-/* flags reset at EOXact */
+/* flags reset at EOXact.  A bit of a misnomer ... */
 #define		PROC_VACUUM_STATE_MASK \
-	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)
+	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | \
+	 PROC_IN_CONCURRENT_REPACK)
 
 /*
  * Xmin-related flags. Make sure any flags that affect how the process' Xmin
diff --git a/src/test/modules/injection_points/Makefile b/src/test/modules/injection_points/Makefile
index 2cd7d87c533..f7663859fe2 100644
--- a/src/test/modules/injection_points/Makefile
+++ b/src/test/modules/injection_points/Makefile
@@ -15,6 +15,7 @@ REGRESS_OPTS = --dlpath=$(top_builddir)/src/test/regress
 ISOLATION = basic \
 	    inplace \
 	    repack \
+	    repack_deadlock \
 	    repack_toast \
 	    syscache-update-pruned \
 	    heap_lock_update
diff --git a/src/test/modules/injection_points/expected/repack_deadlock.out b/src/test/modules/injection_points/expected/repack_deadlock.out
new file mode 100644
index 00000000000..a86e4767536
--- /dev/null
+++ b/src/test/modules/injection_points/expected/repack_deadlock.out
@@ -0,0 +1,63 @@
+Parsed test spec with 2 sessions
+
+starting permutation: wait_before_lock add_column wakeup_before_lock check1
+injection_points_attach
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: 
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+ <waiting ...>
+step add_column: 
+	alter table repack_deadlock add column noise text;
+ <waiting ...>
+step add_column: <... completed>
+ERROR:  could not wait for concurrent REPACK
+step wakeup_before_lock: 
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+
+injection_points_wakeup
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: <... completed>
+step check1: 
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+
+count
+-----
+    1
+(1 row)
+
+i|j
+-+-
+1|1
+2|2
+3|3
+4|4
+(4 rows)
+
+count
+-----
+    4
+(1 row)
+
+injection_points_detach
+-----------------------
+                       
+(1 row)
+
diff --git a/src/test/modules/injection_points/meson.build b/src/test/modules/injection_points/meson.build
index a414abb924b..1cd88d6db65 100644
--- a/src/test/modules/injection_points/meson.build
+++ b/src/test/modules/injection_points/meson.build
@@ -46,6 +46,7 @@ tests += {
       'basic',
       'inplace',
       'repack',
+      'repack_deadlock',
       'repack_toast',
       'syscache-update-pruned',
       'heap_lock_update',
diff --git a/src/test/modules/injection_points/specs/repack_deadlock.spec b/src/test/modules/injection_points/specs/repack_deadlock.spec
new file mode 100644
index 00000000000..9d23a6588c2
--- /dev/null
+++ b/src/test/modules/injection_points/specs/repack_deadlock.spec
@@ -0,0 +1,83 @@
+# Test REPACK with a concurrent transaction that would cause a deadlock
+setup
+{
+	CREATE EXTENSION injection_points;
+
+	CREATE TABLE repack_deadlock(i int PRIMARY KEY, j int);
+	INSERT INTO repack_deadlock(i, j) VALUES (1, 1), (2, 2), (3, 3), (4, 4);
+
+	CREATE TABLE relfilenodes(node oid);
+
+	CREATE TABLE data_s1(i int, j int);
+	CREATE TABLE data_s2(i int, j int);
+}
+
+teardown
+{
+	DROP TABLE repack_deadlock;
+	DROP EXTENSION injection_points;
+
+	DROP TABLE relfilenodes;
+	DROP TABLE data_s1;
+	DROP TABLE data_s2;
+}
+
+session s1
+setup
+{
+	SELECT injection_points_set_local();
+	SELECT injection_points_attach('repack-concurrently-before-lock', 'wait');
+}
+# Perform the initial load and wait for s2 to do some data changes.
+step wait_before_lock
+{
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+}
+# Check the table from the perspective of s1.
+#
+# Besides the contents, we also check that relfilenode has changed.
+
+# Have each session write the contents into a table and use FULL JOIN to check
+# if the outputs are identical.
+step check1
+{
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+}
+teardown
+{
+	SELECT injection_points_detach('repack-concurrently-before-lock');
+}
+
+session s2
+# Change the existing data. UPDATE changes both key and non-key columns. Also
+# update one row twice to test whether tuple version generated by this session
+# can be found.
+step add_column
+{
+	alter table repack_deadlock add column noise text;
+}
+
+step wakeup_before_lock
+{
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+}
+
+# Test if data changes introduced while one session is performing REPACK
+# CONCURRENTLY find their way into the table.
+permutation
+	wait_before_lock
+	add_column
+	wakeup_before_lock
+	check1
-- 
2.47.3


--mdsv5bdgfjjj6d77
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v46-0007-Teach-snapshot-builder-to-skip-transactions-runn.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread

* [PATCH v47 8/9] Error out any process that would block at REPACK
@ 2026-03-25 19:35 Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Álvaro Herrera @ 2026-03-25 19:35 UTC (permalink / raw)

Any process waiting on REPACK to release its lock would actually cause
it to deadlock when it tries to upgrade its lock to AEL, losing all work
done to that point.  We avoid this by teaching the deadlock detector to
raise an error when this condition is detected.
---
 src/backend/commands/repack.c                 | 52 ++++++++----
 src/backend/storage/lmgr/deadlock.c           | 15 ++++
 src/include/storage/proc.h                    |  6 +-
 src/test/modules/injection_points/Makefile    |  1 +
 .../expected/repack_deadlock.out              | 63 ++++++++++++++
 src/test/modules/injection_points/meson.build |  1 +
 .../specs/repack_deadlock.spec                | 83 +++++++++++++++++++
 7 files changed, 202 insertions(+), 19 deletions(-)
 create mode 100644 src/test/modules/injection_points/expected/repack_deadlock.out
 create mode 100644 src/test/modules/injection_points/specs/repack_deadlock.spec

diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c
index 9c807f75d71..3acb2e1d2ba 100644
--- a/src/backend/commands/repack.c
+++ b/src/backend/commands/repack.c
@@ -276,6 +276,16 @@ ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
 	/* Determine the lock mode to use. */
 	lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
 
+	/*
+	 * If in concurrent mode, set the PROC_IN_CONCURRENT_REPACK flag.  This
+	 * makes the deadlock checker cause anyone that would conflict with us
+	 * to error out.  It's important to set this flag ahead of actually locking
+	 * the relation; it won't of course affect anyone until we do have a lock
+	 * that others can conflict with.
+	 */
+	if ((params.options & CLUOPT_CONCURRENT) != 0)
+		MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK;
+
 	/*
 	 * If a single relation is specified, process it and we're done ... unless
 	 * the relation is a partitioned table, in which case we fall through.
@@ -476,11 +486,8 @@ RepackLockLevel(bool concurrent)
  * If indexOid is InvalidOid, the table will be rewritten in physical order
  * instead of index order.
  *
- * Note that, in the concurrent case, the function releases the lock at some
- * point, in order to get AccessExclusiveLock for the final steps (i.e. to
- * swap the relation files). To make things simpler, the caller should expect
- * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
- * AccessExclusiveLock is kept till the end of the transaction.)
+ * On return, OldHeap is closed but locked with AccessExclusiveLock - the lock
+ * will be released at end of the transaction.
  *
  * 'cmd' indicates which command is being executed, to be used for error
  * messages.
@@ -512,10 +519,12 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
 		/*
 		 * Make sure we're not in a transaction block.
 		 *
-		 * The reason is that repack_setup_logical_decoding() could deadlock
-		 * if there's an XID already assigned.  It would be possible to run in
-		 * a transaction block if we had no XID, but this restriction is
-		 * simpler for users to understand and we don't lose anything.
+		 * The reason is that repack_setup_logical_decoding() could wait
+		 * indefinitely for our XID to complete. (The deadlock detector would
+		 * not recognize it because we'd be waiting for ourselves, i.e. no
+		 * real lock conflict.) It would be possible to run in a transaction
+		 * block if we had no XID, but this restriction is simpler for users
+		 * to understand and we don't lose anything.
 		 */
 		PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
 
@@ -998,10 +1007,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose,
 		 * Note that the worker has to wait for all transactions with XID
 		 * already assigned to finish. If some of those transactions is
 		 * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
-		 * table (e.g.  it runs CREATE INDEX), we can end up in a deadlock.
-		 * Not sure this risk is worth unlocking/locking the table (and its
-		 * clustering index) and checking again if it's still eligible for
-		 * REPACK CONCURRENTLY.
+		 * table (e.g. it runs CREATE INDEX), it should encounter ERROR in the
+		 * deadlock checking code.
 		 */
 		start_repack_decoding_worker(tableOid);
 
@@ -3090,7 +3097,16 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
 		LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
 
 	/*
-	 * Tuples and pages of the old heap will be gone, but the heap will stay.
+	 * Now that we have all access-exclusive locks on all relations, we no
+	 * longer want other processes to error out when trying to acquire a
+	 * conflicting lock.  Therefore, unset our flag.
+	 */
+	MyProc->statusFlags &= ~PROC_IN_CONCURRENT_REPACK;
+
+	/*
+	 * Tuples and pages of the old heap will be gone, but the heap itself will
+	 * stay.  In order for predicate locks to continue to work, convert them
+	 * to relation-level locks.  We do this both for table and indexes.
 	 */
 	TransferPredicateLocksToHeapRelation(OldHeap);
 	foreach_ptr(RelationData, index, indexrels)
@@ -3287,9 +3303,11 @@ start_repack_decoding_worker(Oid relid)
 
 	/*
 	 * The decoding setup must be done before the caller can have XID assigned
-	 * for any reason, otherwise the worker might end up in a deadlock,
-	 * waiting for the caller's transaction to end. Therefore wait here until
-	 * the worker indicates that it has the logical decoding initialized.
+	 * for any reason, otherwise the worker might end up waiting for the
+	 * caller's transaction to end. (Deadlock detector does not consider this
+	 * a conflict because the worker is in the same locking group as the
+	 * backend that launched it.) Therefore wait here until the worker
+	 * indicates that it has the logical decoding initialized.
 	 */
 	ConditionVariablePrepareToSleep(&shared->cv);
 	for (;;)
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index b8962d875b6..3d3ec0da111 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -620,6 +620,21 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 						proc->statusFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
+					/*
+					 * Similarly, if we note that we're blocked by some process
+					 * running REPACK (CONCURRENTLY), just fail.  That process
+					 * is going to upgrade its lock at some point, and it would
+					 * be inappropriate for any other process to cause that
+					 * to fail.
+					 */
+					if (checkProc == MyProc &&
+						proc->statusFlags & PROC_IN_CONCURRENT_REPACK)
+						ereport(ERROR,
+								errcode(ERRCODE_OBJECT_IN_USE),
+								errmsg("could not wait for concurrent REPACK"),
+								errdetail("Process %d waits for REPACK running on process %d",
+									   MyProc->pid, proc->pid));
+
 					/* We're done looking at this proclock */
 					break;
 				}
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1dad125706e..8ad9718f3d6 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -69,10 +69,12 @@ struct XidCache
 #define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be
 												 * included in vacuum horizons
 												 * in all databases */
+#define		PROC_IN_CONCURRENT_REPACK	0x40	/* REPACK (CONCURRENTLY) */
 
-/* flags reset at EOXact */
+/* flags reset at EOXact.  A bit of a misnomer ... */
 #define		PROC_VACUUM_STATE_MASK \
-	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)
+	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | \
+	 PROC_IN_CONCURRENT_REPACK)
 
 /*
  * Xmin-related flags. Make sure any flags that affect how the process' Xmin
diff --git a/src/test/modules/injection_points/Makefile b/src/test/modules/injection_points/Makefile
index 2cd7d87c533..f7663859fe2 100644
--- a/src/test/modules/injection_points/Makefile
+++ b/src/test/modules/injection_points/Makefile
@@ -15,6 +15,7 @@ REGRESS_OPTS = --dlpath=$(top_builddir)/src/test/regress
 ISOLATION = basic \
 	    inplace \
 	    repack \
+	    repack_deadlock \
 	    repack_toast \
 	    syscache-update-pruned \
 	    heap_lock_update
diff --git a/src/test/modules/injection_points/expected/repack_deadlock.out b/src/test/modules/injection_points/expected/repack_deadlock.out
new file mode 100644
index 00000000000..a86e4767536
--- /dev/null
+++ b/src/test/modules/injection_points/expected/repack_deadlock.out
@@ -0,0 +1,63 @@
+Parsed test spec with 2 sessions
+
+starting permutation: wait_before_lock add_column wakeup_before_lock check1
+injection_points_attach
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: 
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+ <waiting ...>
+step add_column: 
+	alter table repack_deadlock add column noise text;
+ <waiting ...>
+step add_column: <... completed>
+ERROR:  could not wait for concurrent REPACK
+step wakeup_before_lock: 
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+
+injection_points_wakeup
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: <... completed>
+step check1: 
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+
+count
+-----
+    1
+(1 row)
+
+i|j
+-+-
+1|1
+2|2
+3|3
+4|4
+(4 rows)
+
+count
+-----
+    4
+(1 row)
+
+injection_points_detach
+-----------------------
+                       
+(1 row)
+
diff --git a/src/test/modules/injection_points/meson.build b/src/test/modules/injection_points/meson.build
index a414abb924b..1cd88d6db65 100644
--- a/src/test/modules/injection_points/meson.build
+++ b/src/test/modules/injection_points/meson.build
@@ -46,6 +46,7 @@ tests += {
       'basic',
       'inplace',
       'repack',
+      'repack_deadlock',
       'repack_toast',
       'syscache-update-pruned',
       'heap_lock_update',
diff --git a/src/test/modules/injection_points/specs/repack_deadlock.spec b/src/test/modules/injection_points/specs/repack_deadlock.spec
new file mode 100644
index 00000000000..9d23a6588c2
--- /dev/null
+++ b/src/test/modules/injection_points/specs/repack_deadlock.spec
@@ -0,0 +1,83 @@
+# Test REPACK with a concurrent transaction that would cause a deadlock
+setup
+{
+	CREATE EXTENSION injection_points;
+
+	CREATE TABLE repack_deadlock(i int PRIMARY KEY, j int);
+	INSERT INTO repack_deadlock(i, j) VALUES (1, 1), (2, 2), (3, 3), (4, 4);
+
+	CREATE TABLE relfilenodes(node oid);
+
+	CREATE TABLE data_s1(i int, j int);
+	CREATE TABLE data_s2(i int, j int);
+}
+
+teardown
+{
+	DROP TABLE repack_deadlock;
+	DROP EXTENSION injection_points;
+
+	DROP TABLE relfilenodes;
+	DROP TABLE data_s1;
+	DROP TABLE data_s2;
+}
+
+session s1
+setup
+{
+	SELECT injection_points_set_local();
+	SELECT injection_points_attach('repack-concurrently-before-lock', 'wait');
+}
+# Perform the initial load and wait for s2 to do some data changes.
+step wait_before_lock
+{
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+}
+# Check the table from the perspective of s1.
+#
+# Besides the contents, we also check that relfilenode has changed.
+
+# Have each session write the contents into a table and use FULL JOIN to check
+# if the outputs are identical.
+step check1
+{
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+}
+teardown
+{
+	SELECT injection_points_detach('repack-concurrently-before-lock');
+}
+
+session s2
+# Change the existing data. UPDATE changes both key and non-key columns. Also
+# update one row twice to test whether tuple version generated by this session
+# can be found.
+step add_column
+{
+	alter table repack_deadlock add column noise text;
+}
+
+step wakeup_before_lock
+{
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+}
+
+# Test if data changes introduced while one session is performing REPACK
+# CONCURRENTLY find their way into the table.
+permutation
+	wait_before_lock
+	add_column
+	wakeup_before_lock
+	check1
-- 
2.47.3


--2fxmamo6mu2qbxgv
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v47-0009-Teach-snapshot-builder-to-skip-transactions-runn.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread

* [PATCH v46 6/7] Error out any process that would block at REPACK
@ 2026-03-25 19:35 Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Álvaro Herrera @ 2026-03-25 19:35 UTC (permalink / raw)

Any process waiting on REPACK to release its lock would actually cause
it to deadlock when it tries to upgrade its lock to AEL, losing all work
done to that point.  We avoid this by teaching the deadlock detector to
raise an error when this condition is detected.
---
 src/backend/commands/repack.c                 | 53 ++++++++----
 src/backend/storage/lmgr/deadlock.c           | 15 ++++
 src/include/storage/proc.h                    |  6 +-
 src/test/modules/injection_points/Makefile    |  1 +
 .../expected/repack_deadlock.out              | 63 ++++++++++++++
 src/test/modules/injection_points/meson.build |  1 +
 .../specs/repack_deadlock.spec                | 83 +++++++++++++++++++
 7 files changed, 202 insertions(+), 20 deletions(-)
 create mode 100644 src/test/modules/injection_points/expected/repack_deadlock.out
 create mode 100644 src/test/modules/injection_points/specs/repack_deadlock.spec

diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c
index a58277bb6c1..21ab81c6f7f 100644
--- a/src/backend/commands/repack.c
+++ b/src/backend/commands/repack.c
@@ -276,6 +276,16 @@ ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
 	/* Determine the lock mode to use. */
 	lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
 
+	/*
+	 * If in concurrent mode, set the PROC_IN_CONCURRENT_REPACK flag.  This
+	 * makes the deadlock checker cause anyone that would conflict with us
+	 * to error out.  It's important to set this flag ahead of actually locking
+	 * the relation; it won't of course affect anyone until we do have a lock
+	 * that others can conflict with.
+	 */
+	if ((params.options & CLUOPT_CONCURRENT) != 0)
+		MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK;
+
 	/*
 	 * If a single relation is specified, process it and we're done ... unless
 	 * the relation is a partitioned table, in which case we fall through.
@@ -476,11 +486,8 @@ RepackLockLevel(bool concurrent)
  * If indexOid is InvalidOid, the table will be rewritten in physical order
  * instead of index order.
  *
- * Note that, in the concurrent case, the function releases the lock at some
- * point, in order to get AccessExclusiveLock for the final steps (i.e. to
- * swap the relation files). To make things simpler, the caller should expect
- * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
- * AccessExclusiveLock is kept till the end of the transaction.)
+ * On return, OldHeap is closed but locked with AccessExclusiveLock - the lock
+ * will be released at end of the transaction.
  *
  * 'cmd' indicates which command is being executed, to be used for error
  * messages.
@@ -512,10 +519,12 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
 		/*
 		 * Make sure we're not in a transaction block.
 		 *
-		 * The reason is that repack_setup_logical_decoding() could deadlock
-		 * if there's an XID already assigned.  It would be possible to run in
-		 * a transaction block if we had no XID, but this restriction is
-		 * simpler for users to understand and we don't lose anything.
+		 * The reason is that repack_setup_logical_decoding() could wait
+		 * indefinitely for our XID to complete. (The deadlock detector would
+		 * not recognize it because we'd be waiting for ourselves, i.e. no
+		 * real lock conflict.) It would be possible to run in a transaction
+		 * block if we had no XID, but this restriction is simpler for users
+		 * to understand and we don't lose anything.
 		 */
 		PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
 
@@ -998,10 +1007,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose,
 		 * Note that the worker has to wait for all transactions with XID
 		 * already assigned to finish. If some of those transactions is
 		 * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
-		 * table (e.g.  it runs CREATE INDEX), we can end up in a deadlock.
-		 * Not sure this risk is worth unlocking/locking the table (and its
-		 * clustering index) and checking again if it's still eligible for
-		 * REPACK CONCURRENTLY.
+		 * table (e.g. it runs CREATE INDEX), it should encounter ERROR in the
+		 * deadlock checking code.
 		 */
 		start_repack_decoding_worker(tableOid);
 
@@ -3123,10 +3130,18 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
 		LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
 
 	/*
-	 * Tuples and pages of the old heap will be gone, but the heap will stay.
+	 * Now that we have all access-exclusive locks on all relations, we no
+	 * longer want other processes to error out when trying to acquire a
+	 * conflicting lock.  Therefore, unset our flag.
+	 */
+	MyProc->statusFlags &= ~PROC_IN_CONCURRENT_REPACK;
+
+	/*
+	 * Tuples and pages of the old heap will be gone, but the heap itself will
+	 * stay.  In order for predicate locks to continue to work, convert them
+	 * to relation-level locks.  We do this both for table and indexes.
 	 */
 	TransferPredicateLocksToHeapRelation(OldHeap);
-	/* The same for indexes. */
 	for (int i = 0; i < nind; i++)
 	{
 		Relation	index = ind_refs[i];
@@ -3346,9 +3361,11 @@ start_repack_decoding_worker(Oid relid)
 
 	/*
 	 * The decoding setup must be done before the caller can have XID assigned
-	 * for any reason, otherwise the worker might end up in a deadlock,
-	 * waiting for the caller's transaction to end. Therefore wait here until
-	 * the worker indicates that it has the logical decoding initialized.
+	 * for any reason, otherwise the worker might end up waiting for the
+	 * caller's transaction to end. (Deadlock detector does not consider this
+	 * a conflict because the worker is in the same locking group as the
+	 * backend that launched it.) Therefore wait here until the worker
+	 * indicates that it has the logical decoding initialized.
 	 */
 	ConditionVariablePrepareToSleep(&shared->cv);
 	for (;;)
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index b8962d875b6..3d3ec0da111 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -620,6 +620,21 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 						proc->statusFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
+					/*
+					 * Similarly, if we note that we're blocked by some process
+					 * running REPACK (CONCURRENTLY), just fail.  That process
+					 * is going to upgrade its lock at some point, and it would
+					 * be inappropriate for any other process to cause that
+					 * to fail.
+					 */
+					if (checkProc == MyProc &&
+						proc->statusFlags & PROC_IN_CONCURRENT_REPACK)
+						ereport(ERROR,
+								errcode(ERRCODE_OBJECT_IN_USE),
+								errmsg("could not wait for concurrent REPACK"),
+								errdetail("Process %d waits for REPACK running on process %d",
+									   MyProc->pid, proc->pid));
+
 					/* We're done looking at this proclock */
 					break;
 				}
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1dad125706e..8ad9718f3d6 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -69,10 +69,12 @@ struct XidCache
 #define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be
 												 * included in vacuum horizons
 												 * in all databases */
+#define		PROC_IN_CONCURRENT_REPACK	0x40	/* REPACK (CONCURRENTLY) */
 
-/* flags reset at EOXact */
+/* flags reset at EOXact.  A bit of a misnomer ... */
 #define		PROC_VACUUM_STATE_MASK \
-	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)
+	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | \
+	 PROC_IN_CONCURRENT_REPACK)
 
 /*
  * Xmin-related flags. Make sure any flags that affect how the process' Xmin
diff --git a/src/test/modules/injection_points/Makefile b/src/test/modules/injection_points/Makefile
index 2cd7d87c533..f7663859fe2 100644
--- a/src/test/modules/injection_points/Makefile
+++ b/src/test/modules/injection_points/Makefile
@@ -15,6 +15,7 @@ REGRESS_OPTS = --dlpath=$(top_builddir)/src/test/regress
 ISOLATION = basic \
 	    inplace \
 	    repack \
+	    repack_deadlock \
 	    repack_toast \
 	    syscache-update-pruned \
 	    heap_lock_update
diff --git a/src/test/modules/injection_points/expected/repack_deadlock.out b/src/test/modules/injection_points/expected/repack_deadlock.out
new file mode 100644
index 00000000000..a86e4767536
--- /dev/null
+++ b/src/test/modules/injection_points/expected/repack_deadlock.out
@@ -0,0 +1,63 @@
+Parsed test spec with 2 sessions
+
+starting permutation: wait_before_lock add_column wakeup_before_lock check1
+injection_points_attach
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: 
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+ <waiting ...>
+step add_column: 
+	alter table repack_deadlock add column noise text;
+ <waiting ...>
+step add_column: <... completed>
+ERROR:  could not wait for concurrent REPACK
+step wakeup_before_lock: 
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+
+injection_points_wakeup
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: <... completed>
+step check1: 
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+
+count
+-----
+    1
+(1 row)
+
+i|j
+-+-
+1|1
+2|2
+3|3
+4|4
+(4 rows)
+
+count
+-----
+    4
+(1 row)
+
+injection_points_detach
+-----------------------
+                       
+(1 row)
+
diff --git a/src/test/modules/injection_points/meson.build b/src/test/modules/injection_points/meson.build
index a414abb924b..1cd88d6db65 100644
--- a/src/test/modules/injection_points/meson.build
+++ b/src/test/modules/injection_points/meson.build
@@ -46,6 +46,7 @@ tests += {
       'basic',
       'inplace',
       'repack',
+      'repack_deadlock',
       'repack_toast',
       'syscache-update-pruned',
       'heap_lock_update',
diff --git a/src/test/modules/injection_points/specs/repack_deadlock.spec b/src/test/modules/injection_points/specs/repack_deadlock.spec
new file mode 100644
index 00000000000..9d23a6588c2
--- /dev/null
+++ b/src/test/modules/injection_points/specs/repack_deadlock.spec
@@ -0,0 +1,83 @@
+# Test REPACK with a concurrent transaction that would cause a deadlock
+setup
+{
+	CREATE EXTENSION injection_points;
+
+	CREATE TABLE repack_deadlock(i int PRIMARY KEY, j int);
+	INSERT INTO repack_deadlock(i, j) VALUES (1, 1), (2, 2), (3, 3), (4, 4);
+
+	CREATE TABLE relfilenodes(node oid);
+
+	CREATE TABLE data_s1(i int, j int);
+	CREATE TABLE data_s2(i int, j int);
+}
+
+teardown
+{
+	DROP TABLE repack_deadlock;
+	DROP EXTENSION injection_points;
+
+	DROP TABLE relfilenodes;
+	DROP TABLE data_s1;
+	DROP TABLE data_s2;
+}
+
+session s1
+setup
+{
+	SELECT injection_points_set_local();
+	SELECT injection_points_attach('repack-concurrently-before-lock', 'wait');
+}
+# Perform the initial load and wait for s2 to do some data changes.
+step wait_before_lock
+{
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+}
+# Check the table from the perspective of s1.
+#
+# Besides the contents, we also check that relfilenode has changed.
+
+# Have each session write the contents into a table and use FULL JOIN to check
+# if the outputs are identical.
+step check1
+{
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+}
+teardown
+{
+	SELECT injection_points_detach('repack-concurrently-before-lock');
+}
+
+session s2
+# Change the existing data. UPDATE changes both key and non-key columns. Also
+# update one row twice to test whether tuple version generated by this session
+# can be found.
+step add_column
+{
+	alter table repack_deadlock add column noise text;
+}
+
+step wakeup_before_lock
+{
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+}
+
+# Test if data changes introduced while one session is performing REPACK
+# CONCURRENTLY find their way into the table.
+permutation
+	wait_before_lock
+	add_column
+	wakeup_before_lock
+	check1
-- 
2.47.3


--mdsv5bdgfjjj6d77
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v46-0007-Teach-snapshot-builder-to-skip-transactions-runn.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread

* [PATCH v48 6/7] Error out any process that would block at REPACK
@ 2026-03-25 19:35 Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Álvaro Herrera @ 2026-03-25 19:35 UTC (permalink / raw)

Any process waiting on REPACK to release its lock would actually cause
it to deadlock when it tries to upgrade its lock to AEL, losing all work
done to that point.  We avoid this by teaching the deadlock detector to
raise an error when this condition is detected.
---
 src/backend/commands/repack.c                 | 52 ++++++++----
 src/backend/storage/lmgr/deadlock.c           | 15 ++++
 src/include/storage/proc.h                    |  6 +-
 src/test/modules/injection_points/Makefile    |  1 +
 .../expected/repack_deadlock.out              | 63 ++++++++++++++
 src/test/modules/injection_points/meson.build |  1 +
 .../specs/repack_deadlock.spec                | 83 +++++++++++++++++++
 7 files changed, 202 insertions(+), 19 deletions(-)
 create mode 100644 src/test/modules/injection_points/expected/repack_deadlock.out
 create mode 100644 src/test/modules/injection_points/specs/repack_deadlock.spec

diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c
index a514053b777..781a471fc1d 100644
--- a/src/backend/commands/repack.c
+++ b/src/backend/commands/repack.c
@@ -276,6 +276,16 @@ ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
 	/* Determine the lock mode to use. */
 	lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
 
+	/*
+	 * If in concurrent mode, set the PROC_IN_CONCURRENT_REPACK flag.  This
+	 * makes the deadlock checker cause anyone that would conflict with us to
+	 * error out.  It's important to set this flag ahead of actually locking
+	 * the relation; it won't of course affect anyone until we do have a lock
+	 * that others can conflict with.
+	 */
+	if ((params.options & CLUOPT_CONCURRENT) != 0)
+		MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK;
+
 	/*
 	 * If a single relation is specified, process it and we're done ... unless
 	 * the relation is a partitioned table, in which case we fall through.
@@ -476,11 +486,8 @@ RepackLockLevel(bool concurrent)
  * If indexOid is InvalidOid, the table will be rewritten in physical order
  * instead of index order.
  *
- * Note that, in the concurrent case, the function releases the lock at some
- * point, in order to get AccessExclusiveLock for the final steps (i.e. to
- * swap the relation files). To make things simpler, the caller should expect
- * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
- * AccessExclusiveLock is kept till the end of the transaction.)
+ * On return, OldHeap is closed but locked with AccessExclusiveLock - the lock
+ * will be released at end of the transaction.
  *
  * 'cmd' indicates which command is being executed, to be used for error
  * messages.
@@ -512,10 +519,12 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
 		/*
 		 * Make sure we're not in a transaction block.
 		 *
-		 * The reason is that repack_setup_logical_decoding() could deadlock
-		 * if there's an XID already assigned.  It would be possible to run in
-		 * a transaction block if we had no XID, but this restriction is
-		 * simpler for users to understand and we don't lose anything.
+		 * The reason is that repack_setup_logical_decoding() could wait
+		 * indefinitely for our XID to complete. (The deadlock detector would
+		 * not recognize it because we'd be waiting for ourselves, i.e. no
+		 * real lock conflict.) It would be possible to run in a transaction
+		 * block if we had no XID, but this restriction is simpler for users
+		 * to understand and we don't lose anything.
 		 */
 		PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
 
@@ -998,10 +1007,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose,
 		 * Note that the worker has to wait for all transactions with XID
 		 * already assigned to finish. If some of those transactions is
 		 * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
-		 * table (e.g.  it runs CREATE INDEX), we can end up in a deadlock.
-		 * Not sure this risk is worth unlocking/locking the table (and its
-		 * clustering index) and checking again if it's still eligible for
-		 * REPACK CONCURRENTLY.
+		 * table (e.g. it runs CREATE INDEX), it should encounter ERROR in the
+		 * deadlock checking code.
 		 */
 		start_repack_decoding_worker(tableOid);
 
@@ -3090,7 +3097,16 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
 		LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
 
 	/*
-	 * Tuples and pages of the old heap will be gone, but the heap will stay.
+	 * Now that we have all access-exclusive locks on all relations, we no
+	 * longer want other processes to error out when trying to acquire a
+	 * conflicting lock.  Therefore, unset our flag.
+	 */
+	MyProc->statusFlags &= ~PROC_IN_CONCURRENT_REPACK;
+
+	/*
+	 * Tuples and pages of the old heap will be gone, but the heap itself will
+	 * stay.  In order for predicate locks to continue to work, convert them
+	 * to relation-level locks.  We do this both for table and indexes.
 	 */
 	TransferPredicateLocksToHeapRelation(OldHeap);
 	foreach_ptr(RelationData, index, indexrels)
@@ -3288,9 +3304,11 @@ start_repack_decoding_worker(Oid relid)
 
 	/*
 	 * The decoding setup must be done before the caller can have XID assigned
-	 * for any reason, otherwise the worker might end up in a deadlock,
-	 * waiting for the caller's transaction to end. Therefore wait here until
-	 * the worker indicates that it has the logical decoding initialized.
+	 * for any reason, otherwise the worker might end up waiting for the
+	 * caller's transaction to end. (Deadlock detector does not consider this
+	 * a conflict because the worker is in the same locking group as the
+	 * backend that launched it.) Therefore wait here until the worker
+	 * indicates that it has the logical decoding initialized.
 	 */
 	ConditionVariablePrepareToSleep(&shared->cv);
 	for (;;)
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index b8962d875b6..c20ac682b0d 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -620,6 +620,21 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 						proc->statusFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
+					/*
+					 * Similarly, if we note that we're blocked by some
+					 * process running REPACK (CONCURRENTLY), just fail.  That
+					 * process is going to upgrade its lock at some point, and
+					 * it would be inappropriate for any other process to
+					 * cause that to fail.
+					 */
+					if (checkProc == MyProc &&
+						proc->statusFlags & PROC_IN_CONCURRENT_REPACK)
+						ereport(ERROR,
+								errcode(ERRCODE_OBJECT_IN_USE),
+								errmsg("could not wait for concurrent REPACK"),
+								errdetail("Process %d waits for REPACK running on process %d",
+										  MyProc->pid, proc->pid));
+
 					/* We're done looking at this proclock */
 					break;
 				}
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1dad125706e..8ad9718f3d6 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -69,10 +69,12 @@ struct XidCache
 #define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be
 												 * included in vacuum horizons
 												 * in all databases */
+#define		PROC_IN_CONCURRENT_REPACK	0x40	/* REPACK (CONCURRENTLY) */
 
-/* flags reset at EOXact */
+/* flags reset at EOXact.  A bit of a misnomer ... */
 #define		PROC_VACUUM_STATE_MASK \
-	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)
+	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | \
+	 PROC_IN_CONCURRENT_REPACK)
 
 /*
  * Xmin-related flags. Make sure any flags that affect how the process' Xmin
diff --git a/src/test/modules/injection_points/Makefile b/src/test/modules/injection_points/Makefile
index 2cd7d87c533..f7663859fe2 100644
--- a/src/test/modules/injection_points/Makefile
+++ b/src/test/modules/injection_points/Makefile
@@ -15,6 +15,7 @@ REGRESS_OPTS = --dlpath=$(top_builddir)/src/test/regress
 ISOLATION = basic \
 	    inplace \
 	    repack \
+	    repack_deadlock \
 	    repack_toast \
 	    syscache-update-pruned \
 	    heap_lock_update
diff --git a/src/test/modules/injection_points/expected/repack_deadlock.out b/src/test/modules/injection_points/expected/repack_deadlock.out
new file mode 100644
index 00000000000..a86e4767536
--- /dev/null
+++ b/src/test/modules/injection_points/expected/repack_deadlock.out
@@ -0,0 +1,63 @@
+Parsed test spec with 2 sessions
+
+starting permutation: wait_before_lock add_column wakeup_before_lock check1
+injection_points_attach
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: 
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+ <waiting ...>
+step add_column: 
+	alter table repack_deadlock add column noise text;
+ <waiting ...>
+step add_column: <... completed>
+ERROR:  could not wait for concurrent REPACK
+step wakeup_before_lock: 
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+
+injection_points_wakeup
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: <... completed>
+step check1: 
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+
+count
+-----
+    1
+(1 row)
+
+i|j
+-+-
+1|1
+2|2
+3|3
+4|4
+(4 rows)
+
+count
+-----
+    4
+(1 row)
+
+injection_points_detach
+-----------------------
+                       
+(1 row)
+
diff --git a/src/test/modules/injection_points/meson.build b/src/test/modules/injection_points/meson.build
index a414abb924b..1cd88d6db65 100644
--- a/src/test/modules/injection_points/meson.build
+++ b/src/test/modules/injection_points/meson.build
@@ -46,6 +46,7 @@ tests += {
       'basic',
       'inplace',
       'repack',
+      'repack_deadlock',
       'repack_toast',
       'syscache-update-pruned',
       'heap_lock_update',
diff --git a/src/test/modules/injection_points/specs/repack_deadlock.spec b/src/test/modules/injection_points/specs/repack_deadlock.spec
new file mode 100644
index 00000000000..9d23a6588c2
--- /dev/null
+++ b/src/test/modules/injection_points/specs/repack_deadlock.spec
@@ -0,0 +1,83 @@
+# Test REPACK with a concurrent transaction that would cause a deadlock
+setup
+{
+	CREATE EXTENSION injection_points;
+
+	CREATE TABLE repack_deadlock(i int PRIMARY KEY, j int);
+	INSERT INTO repack_deadlock(i, j) VALUES (1, 1), (2, 2), (3, 3), (4, 4);
+
+	CREATE TABLE relfilenodes(node oid);
+
+	CREATE TABLE data_s1(i int, j int);
+	CREATE TABLE data_s2(i int, j int);
+}
+
+teardown
+{
+	DROP TABLE repack_deadlock;
+	DROP EXTENSION injection_points;
+
+	DROP TABLE relfilenodes;
+	DROP TABLE data_s1;
+	DROP TABLE data_s2;
+}
+
+session s1
+setup
+{
+	SELECT injection_points_set_local();
+	SELECT injection_points_attach('repack-concurrently-before-lock', 'wait');
+}
+# Perform the initial load and wait for s2 to do some data changes.
+step wait_before_lock
+{
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+}
+# Check the table from the perspective of s1.
+#
+# Besides the contents, we also check that relfilenode has changed.
+
+# Have each session write the contents into a table and use FULL JOIN to check
+# if the outputs are identical.
+step check1
+{
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+}
+teardown
+{
+	SELECT injection_points_detach('repack-concurrently-before-lock');
+}
+
+session s2
+# Change the existing data. UPDATE changes both key and non-key columns. Also
+# update one row twice to test whether tuple version generated by this session
+# can be found.
+step add_column
+{
+	alter table repack_deadlock add column noise text;
+}
+
+step wakeup_before_lock
+{
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+}
+
+# Test if data changes introduced while one session is performing REPACK
+# CONCURRENTLY find their way into the table.
+permutation
+	wait_before_lock
+	add_column
+	wakeup_before_lock
+	check1
-- 
2.47.3


--qfkt2ktdpcfeypib
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v48-0007-Teach-snapshot-builder-to-skip-transactions-runn.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread

* [PATCH v46 6/7] Error out any process that would block at REPACK
@ 2026-03-25 19:35 Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Álvaro Herrera @ 2026-03-25 19:35 UTC (permalink / raw)

Any process waiting on REPACK to release its lock would actually cause
it to deadlock when it tries to upgrade its lock to AEL, losing all work
done to that point.  We avoid this by teaching the deadlock detector to
raise an error when this condition is detected.
---
 src/backend/commands/repack.c                 | 53 ++++++++----
 src/backend/storage/lmgr/deadlock.c           | 15 ++++
 src/include/storage/proc.h                    |  6 +-
 src/test/modules/injection_points/Makefile    |  1 +
 .../expected/repack_deadlock.out              | 63 ++++++++++++++
 src/test/modules/injection_points/meson.build |  1 +
 .../specs/repack_deadlock.spec                | 83 +++++++++++++++++++
 7 files changed, 202 insertions(+), 20 deletions(-)
 create mode 100644 src/test/modules/injection_points/expected/repack_deadlock.out
 create mode 100644 src/test/modules/injection_points/specs/repack_deadlock.spec

diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c
index a58277bb6c1..21ab81c6f7f 100644
--- a/src/backend/commands/repack.c
+++ b/src/backend/commands/repack.c
@@ -276,6 +276,16 @@ ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
 	/* Determine the lock mode to use. */
 	lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
 
+	/*
+	 * If in concurrent mode, set the PROC_IN_CONCURRENT_REPACK flag.  This
+	 * makes the deadlock checker cause anyone that would conflict with us
+	 * to error out.  It's important to set this flag ahead of actually locking
+	 * the relation; it won't of course affect anyone until we do have a lock
+	 * that others can conflict with.
+	 */
+	if ((params.options & CLUOPT_CONCURRENT) != 0)
+		MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK;
+
 	/*
 	 * If a single relation is specified, process it and we're done ... unless
 	 * the relation is a partitioned table, in which case we fall through.
@@ -476,11 +486,8 @@ RepackLockLevel(bool concurrent)
  * If indexOid is InvalidOid, the table will be rewritten in physical order
  * instead of index order.
  *
- * Note that, in the concurrent case, the function releases the lock at some
- * point, in order to get AccessExclusiveLock for the final steps (i.e. to
- * swap the relation files). To make things simpler, the caller should expect
- * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
- * AccessExclusiveLock is kept till the end of the transaction.)
+ * On return, OldHeap is closed but locked with AccessExclusiveLock - the lock
+ * will be released at end of the transaction.
  *
  * 'cmd' indicates which command is being executed, to be used for error
  * messages.
@@ -512,10 +519,12 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
 		/*
 		 * Make sure we're not in a transaction block.
 		 *
-		 * The reason is that repack_setup_logical_decoding() could deadlock
-		 * if there's an XID already assigned.  It would be possible to run in
-		 * a transaction block if we had no XID, but this restriction is
-		 * simpler for users to understand and we don't lose anything.
+		 * The reason is that repack_setup_logical_decoding() could wait
+		 * indefinitely for our XID to complete. (The deadlock detector would
+		 * not recognize it because we'd be waiting for ourselves, i.e. no
+		 * real lock conflict.) It would be possible to run in a transaction
+		 * block if we had no XID, but this restriction is simpler for users
+		 * to understand and we don't lose anything.
 		 */
 		PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
 
@@ -998,10 +1007,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose,
 		 * Note that the worker has to wait for all transactions with XID
 		 * already assigned to finish. If some of those transactions is
 		 * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
-		 * table (e.g.  it runs CREATE INDEX), we can end up in a deadlock.
-		 * Not sure this risk is worth unlocking/locking the table (and its
-		 * clustering index) and checking again if it's still eligible for
-		 * REPACK CONCURRENTLY.
+		 * table (e.g. it runs CREATE INDEX), it should encounter ERROR in the
+		 * deadlock checking code.
 		 */
 		start_repack_decoding_worker(tableOid);
 
@@ -3123,10 +3130,18 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
 		LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
 
 	/*
-	 * Tuples and pages of the old heap will be gone, but the heap will stay.
+	 * Now that we have all access-exclusive locks on all relations, we no
+	 * longer want other processes to error out when trying to acquire a
+	 * conflicting lock.  Therefore, unset our flag.
+	 */
+	MyProc->statusFlags &= ~PROC_IN_CONCURRENT_REPACK;
+
+	/*
+	 * Tuples and pages of the old heap will be gone, but the heap itself will
+	 * stay.  In order for predicate locks to continue to work, convert them
+	 * to relation-level locks.  We do this both for table and indexes.
 	 */
 	TransferPredicateLocksToHeapRelation(OldHeap);
-	/* The same for indexes. */
 	for (int i = 0; i < nind; i++)
 	{
 		Relation	index = ind_refs[i];
@@ -3346,9 +3361,11 @@ start_repack_decoding_worker(Oid relid)
 
 	/*
 	 * The decoding setup must be done before the caller can have XID assigned
-	 * for any reason, otherwise the worker might end up in a deadlock,
-	 * waiting for the caller's transaction to end. Therefore wait here until
-	 * the worker indicates that it has the logical decoding initialized.
+	 * for any reason, otherwise the worker might end up waiting for the
+	 * caller's transaction to end. (Deadlock detector does not consider this
+	 * a conflict because the worker is in the same locking group as the
+	 * backend that launched it.) Therefore wait here until the worker
+	 * indicates that it has the logical decoding initialized.
 	 */
 	ConditionVariablePrepareToSleep(&shared->cv);
 	for (;;)
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index b8962d875b6..3d3ec0da111 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -620,6 +620,21 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 						proc->statusFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
+					/*
+					 * Similarly, if we note that we're blocked by some process
+					 * running REPACK (CONCURRENTLY), just fail.  That process
+					 * is going to upgrade its lock at some point, and it would
+					 * be inappropriate for any other process to cause that
+					 * to fail.
+					 */
+					if (checkProc == MyProc &&
+						proc->statusFlags & PROC_IN_CONCURRENT_REPACK)
+						ereport(ERROR,
+								errcode(ERRCODE_OBJECT_IN_USE),
+								errmsg("could not wait for concurrent REPACK"),
+								errdetail("Process %d waits for REPACK running on process %d",
+									   MyProc->pid, proc->pid));
+
 					/* We're done looking at this proclock */
 					break;
 				}
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1dad125706e..8ad9718f3d6 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -69,10 +69,12 @@ struct XidCache
 #define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be
 												 * included in vacuum horizons
 												 * in all databases */
+#define		PROC_IN_CONCURRENT_REPACK	0x40	/* REPACK (CONCURRENTLY) */
 
-/* flags reset at EOXact */
+/* flags reset at EOXact.  A bit of a misnomer ... */
 #define		PROC_VACUUM_STATE_MASK \
-	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)
+	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | \
+	 PROC_IN_CONCURRENT_REPACK)
 
 /*
  * Xmin-related flags. Make sure any flags that affect how the process' Xmin
diff --git a/src/test/modules/injection_points/Makefile b/src/test/modules/injection_points/Makefile
index 2cd7d87c533..f7663859fe2 100644
--- a/src/test/modules/injection_points/Makefile
+++ b/src/test/modules/injection_points/Makefile
@@ -15,6 +15,7 @@ REGRESS_OPTS = --dlpath=$(top_builddir)/src/test/regress
 ISOLATION = basic \
 	    inplace \
 	    repack \
+	    repack_deadlock \
 	    repack_toast \
 	    syscache-update-pruned \
 	    heap_lock_update
diff --git a/src/test/modules/injection_points/expected/repack_deadlock.out b/src/test/modules/injection_points/expected/repack_deadlock.out
new file mode 100644
index 00000000000..a86e4767536
--- /dev/null
+++ b/src/test/modules/injection_points/expected/repack_deadlock.out
@@ -0,0 +1,63 @@
+Parsed test spec with 2 sessions
+
+starting permutation: wait_before_lock add_column wakeup_before_lock check1
+injection_points_attach
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: 
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+ <waiting ...>
+step add_column: 
+	alter table repack_deadlock add column noise text;
+ <waiting ...>
+step add_column: <... completed>
+ERROR:  could not wait for concurrent REPACK
+step wakeup_before_lock: 
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+
+injection_points_wakeup
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: <... completed>
+step check1: 
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+
+count
+-----
+    1
+(1 row)
+
+i|j
+-+-
+1|1
+2|2
+3|3
+4|4
+(4 rows)
+
+count
+-----
+    4
+(1 row)
+
+injection_points_detach
+-----------------------
+                       
+(1 row)
+
diff --git a/src/test/modules/injection_points/meson.build b/src/test/modules/injection_points/meson.build
index a414abb924b..1cd88d6db65 100644
--- a/src/test/modules/injection_points/meson.build
+++ b/src/test/modules/injection_points/meson.build
@@ -46,6 +46,7 @@ tests += {
       'basic',
       'inplace',
       'repack',
+      'repack_deadlock',
       'repack_toast',
       'syscache-update-pruned',
       'heap_lock_update',
diff --git a/src/test/modules/injection_points/specs/repack_deadlock.spec b/src/test/modules/injection_points/specs/repack_deadlock.spec
new file mode 100644
index 00000000000..9d23a6588c2
--- /dev/null
+++ b/src/test/modules/injection_points/specs/repack_deadlock.spec
@@ -0,0 +1,83 @@
+# Test REPACK with a concurrent transaction that would cause a deadlock
+setup
+{
+	CREATE EXTENSION injection_points;
+
+	CREATE TABLE repack_deadlock(i int PRIMARY KEY, j int);
+	INSERT INTO repack_deadlock(i, j) VALUES (1, 1), (2, 2), (3, 3), (4, 4);
+
+	CREATE TABLE relfilenodes(node oid);
+
+	CREATE TABLE data_s1(i int, j int);
+	CREATE TABLE data_s2(i int, j int);
+}
+
+teardown
+{
+	DROP TABLE repack_deadlock;
+	DROP EXTENSION injection_points;
+
+	DROP TABLE relfilenodes;
+	DROP TABLE data_s1;
+	DROP TABLE data_s2;
+}
+
+session s1
+setup
+{
+	SELECT injection_points_set_local();
+	SELECT injection_points_attach('repack-concurrently-before-lock', 'wait');
+}
+# Perform the initial load and wait for s2 to do some data changes.
+step wait_before_lock
+{
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+}
+# Check the table from the perspective of s1.
+#
+# Besides the contents, we also check that relfilenode has changed.
+
+# Have each session write the contents into a table and use FULL JOIN to check
+# if the outputs are identical.
+step check1
+{
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+}
+teardown
+{
+	SELECT injection_points_detach('repack-concurrently-before-lock');
+}
+
+session s2
+# Change the existing data. UPDATE changes both key and non-key columns. Also
+# update one row twice to test whether tuple version generated by this session
+# can be found.
+step add_column
+{
+	alter table repack_deadlock add column noise text;
+}
+
+step wakeup_before_lock
+{
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+}
+
+# Test if data changes introduced while one session is performing REPACK
+# CONCURRENTLY find their way into the table.
+permutation
+	wait_before_lock
+	add_column
+	wakeup_before_lock
+	check1
-- 
2.47.3


--mdsv5bdgfjjj6d77
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v46-0007-Teach-snapshot-builder-to-skip-transactions-runn.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread

* [PATCH v45 7/8] Error out any process that would block at REPACK
@ 2026-03-25 19:35 Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Álvaro Herrera @ 2026-03-25 19:35 UTC (permalink / raw)

Any process waiting on REPACK to release its lock would actually cause
it to deadlock when it tries to upgrade its lock to AEL, losing all work
done to that point.  We avoid this by teaching the deadlock detector to
raise an error when this condition is detected.
---
 src/backend/commands/repack.c                 | 53 ++++++++----
 src/backend/storage/lmgr/deadlock.c           | 15 ++++
 src/include/storage/proc.h                    |  6 +-
 src/test/modules/injection_points/Makefile    |  1 +
 .../expected/repack_deadlock.out              | 63 ++++++++++++++
 src/test/modules/injection_points/meson.build |  1 +
 .../specs/repack_deadlock.spec                | 83 +++++++++++++++++++
 7 files changed, 202 insertions(+), 20 deletions(-)
 create mode 100644 src/test/modules/injection_points/expected/repack_deadlock.out
 create mode 100644 src/test/modules/injection_points/specs/repack_deadlock.spec

diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c
index d25ab43da7b..7291b4180ab 100644
--- a/src/backend/commands/repack.c
+++ b/src/backend/commands/repack.c
@@ -276,6 +276,16 @@ ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
 	/* Determine the lock mode to use. */
 	lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
 
+	/*
+	 * If in concurrent mode, set the PROC_IN_CONCURRENT_REPACK flag.  This
+	 * makes the deadlock checker cause anyone that would conflict with us
+	 * to error out.  It's important to set this flag ahead of actually locking
+	 * the relation; it won't of course affect anyone until we do have a lock
+	 * that others can conflict with.
+	 */
+	if ((params.options & CLUOPT_CONCURRENT) != 0)
+		MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK;
+
 	/*
 	 * If a single relation is specified, process it and we're done ... unless
 	 * the relation is a partitioned table, in which case we fall through.
@@ -476,11 +486,8 @@ RepackLockLevel(bool concurrent)
  * If indexOid is InvalidOid, the table will be rewritten in physical order
  * instead of index order.
  *
- * Note that, in the concurrent case, the function releases the lock at some
- * point, in order to get AccessExclusiveLock for the final steps (i.e. to
- * swap the relation files). To make things simpler, the caller should expect
- * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
- * AccessExclusiveLock is kept till the end of the transaction.)
+ * On return, OldHeap is closed but locked with AccessExclusiveLock - the lock
+ * will be released at end of the transaction.
  *
  * 'cmd' indicates which command is being executed, to be used for error
  * messages.
@@ -512,10 +519,12 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
 		/*
 		 * Make sure we're not in a transaction block.
 		 *
-		 * The reason is that repack_setup_logical_decoding() could deadlock
-		 * if there's an XID already assigned.  It would be possible to run in
-		 * a transaction block if we had no XID, but this restriction is
-		 * simpler for users to understand and we don't lose anything.
+		 * The reason is that repack_setup_logical_decoding() could wait
+		 * indefinitely for our XID to complete. (The deadlock detector would
+		 * not recognize it because we'd be waiting for ourselves, i.e. no
+		 * real lock conflict.) It would be possible to run in a transaction
+		 * block if we had no XID, but this restriction is simpler for users
+		 * to understand and we don't lose anything.
 		 */
 		PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
 
@@ -998,10 +1007,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose,
 		 * Note that the worker has to wait for all transactions with XID
 		 * already assigned to finish. If some of those transactions is
 		 * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
-		 * table (e.g.  it runs CREATE INDEX), we can end up in a deadlock.
-		 * Not sure this risk is worth unlocking/locking the table (and its
-		 * clustering index) and checking again if it's still eligible for
-		 * REPACK CONCURRENTLY.
+		 * table (e.g. it runs CREATE INDEX), it should encounter ERROR in the
+		 * deadlock checking code.
 		 */
 		start_repack_decoding_worker(tableOid);
 
@@ -3119,10 +3126,18 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
 		LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
 
 	/*
-	 * Tuples and pages of the old heap will be gone, but the heap will stay.
+	 * Now that we have all access-exclusive locks on all relations, we no
+	 * longer want other processes to error out when trying to acquire a
+	 * conflicting lock.  Therefore, unset our flag.
+	 */
+	MyProc->statusFlags &= ~PROC_IN_CONCURRENT_REPACK;
+
+	/*
+	 * Tuples and pages of the old heap will be gone, but the heap itself will
+	 * stay.  In order for predicate locks to continue to work, convert them
+	 * to relation-level locks.  We do this both for table and indexes.
 	 */
 	TransferPredicateLocksToHeapRelation(OldHeap);
-	/* The same for indexes. */
 	for (int i = 0; i < nind; i++)
 	{
 		Relation	index = ind_refs[i];
@@ -3342,9 +3357,11 @@ start_repack_decoding_worker(Oid relid)
 
 	/*
 	 * The decoding setup must be done before the caller can have XID assigned
-	 * for any reason, otherwise the worker might end up in a deadlock,
-	 * waiting for the caller's transaction to end. Therefore wait here until
-	 * the worker indicates that it has the logical decoding initialized.
+	 * for any reason, otherwise the worker might end up waiting for the
+	 * caller's transaction to end. (Deadlock detector does not consider this
+	 * a conflict because the worker is in the same locking group as the
+	 * backend that launched it.) Therefore wait here until the worker
+	 * indicates that it has the logical decoding initialized.
 	 */
 	ConditionVariablePrepareToSleep(&shared->cv);
 	for (;;)
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index b8962d875b6..3d3ec0da111 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -620,6 +620,21 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 						proc->statusFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
+					/*
+					 * Similarly, if we note that we're blocked by some process
+					 * running REPACK (CONCURRENTLY), just fail.  That process
+					 * is going to upgrade its lock at some point, and it would
+					 * be inappropriate for any other process to cause that
+					 * to fail.
+					 */
+					if (checkProc == MyProc &&
+						proc->statusFlags & PROC_IN_CONCURRENT_REPACK)
+						ereport(ERROR,
+								errcode(ERRCODE_OBJECT_IN_USE),
+								errmsg("could not wait for concurrent REPACK"),
+								errdetail("Process %d waits for REPACK running on process %d",
+									   MyProc->pid, proc->pid));
+
 					/* We're done looking at this proclock */
 					break;
 				}
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1dad125706e..8ad9718f3d6 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -69,10 +69,12 @@ struct XidCache
 #define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be
 												 * included in vacuum horizons
 												 * in all databases */
+#define		PROC_IN_CONCURRENT_REPACK	0x40	/* REPACK (CONCURRENTLY) */
 
-/* flags reset at EOXact */
+/* flags reset at EOXact.  A bit of a misnomer ... */
 #define		PROC_VACUUM_STATE_MASK \
-	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)
+	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | \
+	 PROC_IN_CONCURRENT_REPACK)
 
 /*
  * Xmin-related flags. Make sure any flags that affect how the process' Xmin
diff --git a/src/test/modules/injection_points/Makefile b/src/test/modules/injection_points/Makefile
index 2cd7d87c533..f7663859fe2 100644
--- a/src/test/modules/injection_points/Makefile
+++ b/src/test/modules/injection_points/Makefile
@@ -15,6 +15,7 @@ REGRESS_OPTS = --dlpath=$(top_builddir)/src/test/regress
 ISOLATION = basic \
 	    inplace \
 	    repack \
+	    repack_deadlock \
 	    repack_toast \
 	    syscache-update-pruned \
 	    heap_lock_update
diff --git a/src/test/modules/injection_points/expected/repack_deadlock.out b/src/test/modules/injection_points/expected/repack_deadlock.out
new file mode 100644
index 00000000000..a86e4767536
--- /dev/null
+++ b/src/test/modules/injection_points/expected/repack_deadlock.out
@@ -0,0 +1,63 @@
+Parsed test spec with 2 sessions
+
+starting permutation: wait_before_lock add_column wakeup_before_lock check1
+injection_points_attach
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: 
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+ <waiting ...>
+step add_column: 
+	alter table repack_deadlock add column noise text;
+ <waiting ...>
+step add_column: <... completed>
+ERROR:  could not wait for concurrent REPACK
+step wakeup_before_lock: 
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+
+injection_points_wakeup
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: <... completed>
+step check1: 
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+
+count
+-----
+    1
+(1 row)
+
+i|j
+-+-
+1|1
+2|2
+3|3
+4|4
+(4 rows)
+
+count
+-----
+    4
+(1 row)
+
+injection_points_detach
+-----------------------
+                       
+(1 row)
+
diff --git a/src/test/modules/injection_points/meson.build b/src/test/modules/injection_points/meson.build
index a414abb924b..1cd88d6db65 100644
--- a/src/test/modules/injection_points/meson.build
+++ b/src/test/modules/injection_points/meson.build
@@ -46,6 +46,7 @@ tests += {
       'basic',
       'inplace',
       'repack',
+      'repack_deadlock',
       'repack_toast',
       'syscache-update-pruned',
       'heap_lock_update',
diff --git a/src/test/modules/injection_points/specs/repack_deadlock.spec b/src/test/modules/injection_points/specs/repack_deadlock.spec
new file mode 100644
index 00000000000..9d23a6588c2
--- /dev/null
+++ b/src/test/modules/injection_points/specs/repack_deadlock.spec
@@ -0,0 +1,83 @@
+# Test REPACK with a concurrent transaction that would cause a deadlock
+setup
+{
+	CREATE EXTENSION injection_points;
+
+	CREATE TABLE repack_deadlock(i int PRIMARY KEY, j int);
+	INSERT INTO repack_deadlock(i, j) VALUES (1, 1), (2, 2), (3, 3), (4, 4);
+
+	CREATE TABLE relfilenodes(node oid);
+
+	CREATE TABLE data_s1(i int, j int);
+	CREATE TABLE data_s2(i int, j int);
+}
+
+teardown
+{
+	DROP TABLE repack_deadlock;
+	DROP EXTENSION injection_points;
+
+	DROP TABLE relfilenodes;
+	DROP TABLE data_s1;
+	DROP TABLE data_s2;
+}
+
+session s1
+setup
+{
+	SELECT injection_points_set_local();
+	SELECT injection_points_attach('repack-concurrently-before-lock', 'wait');
+}
+# Perform the initial load and wait for s2 to do some data changes.
+step wait_before_lock
+{
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+}
+# Check the table from the perspective of s1.
+#
+# Besides the contents, we also check that relfilenode has changed.
+
+# Have each session write the contents into a table and use FULL JOIN to check
+# if the outputs are identical.
+step check1
+{
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+}
+teardown
+{
+	SELECT injection_points_detach('repack-concurrently-before-lock');
+}
+
+session s2
+# Change the existing data. UPDATE changes both key and non-key columns. Also
+# update one row twice to test whether tuple version generated by this session
+# can be found.
+step add_column
+{
+	alter table repack_deadlock add column noise text;
+}
+
+step wakeup_before_lock
+{
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+}
+
+# Test if data changes introduced while one session is performing REPACK
+# CONCURRENTLY find their way into the table.
+permutation
+	wait_before_lock
+	add_column
+	wakeup_before_lock
+	check1
-- 
2.47.3


--cldir6umqisr673d
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v45-0008-Teach-snapshot-builder-to-skip-transactions-runn.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread

* [PATCH v48 6/7] Error out any process that would block at REPACK
@ 2026-03-25 19:35 Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Álvaro Herrera @ 2026-03-25 19:35 UTC (permalink / raw)

Any process waiting on REPACK to release its lock would actually cause
it to deadlock when it tries to upgrade its lock to AEL, losing all work
done to that point.  We avoid this by teaching the deadlock detector to
raise an error when this condition is detected.
---
 src/backend/commands/repack.c                 | 52 ++++++++----
 src/backend/storage/lmgr/deadlock.c           | 15 ++++
 src/include/storage/proc.h                    |  6 +-
 src/test/modules/injection_points/Makefile    |  1 +
 .../expected/repack_deadlock.out              | 63 ++++++++++++++
 src/test/modules/injection_points/meson.build |  1 +
 .../specs/repack_deadlock.spec                | 83 +++++++++++++++++++
 7 files changed, 202 insertions(+), 19 deletions(-)
 create mode 100644 src/test/modules/injection_points/expected/repack_deadlock.out
 create mode 100644 src/test/modules/injection_points/specs/repack_deadlock.spec

diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c
index a514053b777..781a471fc1d 100644
--- a/src/backend/commands/repack.c
+++ b/src/backend/commands/repack.c
@@ -276,6 +276,16 @@ ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
 	/* Determine the lock mode to use. */
 	lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
 
+	/*
+	 * If in concurrent mode, set the PROC_IN_CONCURRENT_REPACK flag.  This
+	 * makes the deadlock checker cause anyone that would conflict with us to
+	 * error out.  It's important to set this flag ahead of actually locking
+	 * the relation; it won't of course affect anyone until we do have a lock
+	 * that others can conflict with.
+	 */
+	if ((params.options & CLUOPT_CONCURRENT) != 0)
+		MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK;
+
 	/*
 	 * If a single relation is specified, process it and we're done ... unless
 	 * the relation is a partitioned table, in which case we fall through.
@@ -476,11 +486,8 @@ RepackLockLevel(bool concurrent)
  * If indexOid is InvalidOid, the table will be rewritten in physical order
  * instead of index order.
  *
- * Note that, in the concurrent case, the function releases the lock at some
- * point, in order to get AccessExclusiveLock for the final steps (i.e. to
- * swap the relation files). To make things simpler, the caller should expect
- * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
- * AccessExclusiveLock is kept till the end of the transaction.)
+ * On return, OldHeap is closed but locked with AccessExclusiveLock - the lock
+ * will be released at end of the transaction.
  *
  * 'cmd' indicates which command is being executed, to be used for error
  * messages.
@@ -512,10 +519,12 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
 		/*
 		 * Make sure we're not in a transaction block.
 		 *
-		 * The reason is that repack_setup_logical_decoding() could deadlock
-		 * if there's an XID already assigned.  It would be possible to run in
-		 * a transaction block if we had no XID, but this restriction is
-		 * simpler for users to understand and we don't lose anything.
+		 * The reason is that repack_setup_logical_decoding() could wait
+		 * indefinitely for our XID to complete. (The deadlock detector would
+		 * not recognize it because we'd be waiting for ourselves, i.e. no
+		 * real lock conflict.) It would be possible to run in a transaction
+		 * block if we had no XID, but this restriction is simpler for users
+		 * to understand and we don't lose anything.
 		 */
 		PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
 
@@ -998,10 +1007,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose,
 		 * Note that the worker has to wait for all transactions with XID
 		 * already assigned to finish. If some of those transactions is
 		 * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
-		 * table (e.g.  it runs CREATE INDEX), we can end up in a deadlock.
-		 * Not sure this risk is worth unlocking/locking the table (and its
-		 * clustering index) and checking again if it's still eligible for
-		 * REPACK CONCURRENTLY.
+		 * table (e.g. it runs CREATE INDEX), it should encounter ERROR in the
+		 * deadlock checking code.
 		 */
 		start_repack_decoding_worker(tableOid);
 
@@ -3090,7 +3097,16 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
 		LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
 
 	/*
-	 * Tuples and pages of the old heap will be gone, but the heap will stay.
+	 * Now that we have all access-exclusive locks on all relations, we no
+	 * longer want other processes to error out when trying to acquire a
+	 * conflicting lock.  Therefore, unset our flag.
+	 */
+	MyProc->statusFlags &= ~PROC_IN_CONCURRENT_REPACK;
+
+	/*
+	 * Tuples and pages of the old heap will be gone, but the heap itself will
+	 * stay.  In order for predicate locks to continue to work, convert them
+	 * to relation-level locks.  We do this both for table and indexes.
 	 */
 	TransferPredicateLocksToHeapRelation(OldHeap);
 	foreach_ptr(RelationData, index, indexrels)
@@ -3288,9 +3304,11 @@ start_repack_decoding_worker(Oid relid)
 
 	/*
 	 * The decoding setup must be done before the caller can have XID assigned
-	 * for any reason, otherwise the worker might end up in a deadlock,
-	 * waiting for the caller's transaction to end. Therefore wait here until
-	 * the worker indicates that it has the logical decoding initialized.
+	 * for any reason, otherwise the worker might end up waiting for the
+	 * caller's transaction to end. (Deadlock detector does not consider this
+	 * a conflict because the worker is in the same locking group as the
+	 * backend that launched it.) Therefore wait here until the worker
+	 * indicates that it has the logical decoding initialized.
 	 */
 	ConditionVariablePrepareToSleep(&shared->cv);
 	for (;;)
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index b8962d875b6..c20ac682b0d 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -620,6 +620,21 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 						proc->statusFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
+					/*
+					 * Similarly, if we note that we're blocked by some
+					 * process running REPACK (CONCURRENTLY), just fail.  That
+					 * process is going to upgrade its lock at some point, and
+					 * it would be inappropriate for any other process to
+					 * cause that to fail.
+					 */
+					if (checkProc == MyProc &&
+						proc->statusFlags & PROC_IN_CONCURRENT_REPACK)
+						ereport(ERROR,
+								errcode(ERRCODE_OBJECT_IN_USE),
+								errmsg("could not wait for concurrent REPACK"),
+								errdetail("Process %d waits for REPACK running on process %d",
+										  MyProc->pid, proc->pid));
+
 					/* We're done looking at this proclock */
 					break;
 				}
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1dad125706e..8ad9718f3d6 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -69,10 +69,12 @@ struct XidCache
 #define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be
 												 * included in vacuum horizons
 												 * in all databases */
+#define		PROC_IN_CONCURRENT_REPACK	0x40	/* REPACK (CONCURRENTLY) */
 
-/* flags reset at EOXact */
+/* flags reset at EOXact.  A bit of a misnomer ... */
 #define		PROC_VACUUM_STATE_MASK \
-	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)
+	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | \
+	 PROC_IN_CONCURRENT_REPACK)
 
 /*
  * Xmin-related flags. Make sure any flags that affect how the process' Xmin
diff --git a/src/test/modules/injection_points/Makefile b/src/test/modules/injection_points/Makefile
index 2cd7d87c533..f7663859fe2 100644
--- a/src/test/modules/injection_points/Makefile
+++ b/src/test/modules/injection_points/Makefile
@@ -15,6 +15,7 @@ REGRESS_OPTS = --dlpath=$(top_builddir)/src/test/regress
 ISOLATION = basic \
 	    inplace \
 	    repack \
+	    repack_deadlock \
 	    repack_toast \
 	    syscache-update-pruned \
 	    heap_lock_update
diff --git a/src/test/modules/injection_points/expected/repack_deadlock.out b/src/test/modules/injection_points/expected/repack_deadlock.out
new file mode 100644
index 00000000000..a86e4767536
--- /dev/null
+++ b/src/test/modules/injection_points/expected/repack_deadlock.out
@@ -0,0 +1,63 @@
+Parsed test spec with 2 sessions
+
+starting permutation: wait_before_lock add_column wakeup_before_lock check1
+injection_points_attach
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: 
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+ <waiting ...>
+step add_column: 
+	alter table repack_deadlock add column noise text;
+ <waiting ...>
+step add_column: <... completed>
+ERROR:  could not wait for concurrent REPACK
+step wakeup_before_lock: 
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+
+injection_points_wakeup
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: <... completed>
+step check1: 
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+
+count
+-----
+    1
+(1 row)
+
+i|j
+-+-
+1|1
+2|2
+3|3
+4|4
+(4 rows)
+
+count
+-----
+    4
+(1 row)
+
+injection_points_detach
+-----------------------
+                       
+(1 row)
+
diff --git a/src/test/modules/injection_points/meson.build b/src/test/modules/injection_points/meson.build
index a414abb924b..1cd88d6db65 100644
--- a/src/test/modules/injection_points/meson.build
+++ b/src/test/modules/injection_points/meson.build
@@ -46,6 +46,7 @@ tests += {
       'basic',
       'inplace',
       'repack',
+      'repack_deadlock',
       'repack_toast',
       'syscache-update-pruned',
       'heap_lock_update',
diff --git a/src/test/modules/injection_points/specs/repack_deadlock.spec b/src/test/modules/injection_points/specs/repack_deadlock.spec
new file mode 100644
index 00000000000..9d23a6588c2
--- /dev/null
+++ b/src/test/modules/injection_points/specs/repack_deadlock.spec
@@ -0,0 +1,83 @@
+# Test REPACK with a concurrent transaction that would cause a deadlock
+setup
+{
+	CREATE EXTENSION injection_points;
+
+	CREATE TABLE repack_deadlock(i int PRIMARY KEY, j int);
+	INSERT INTO repack_deadlock(i, j) VALUES (1, 1), (2, 2), (3, 3), (4, 4);
+
+	CREATE TABLE relfilenodes(node oid);
+
+	CREATE TABLE data_s1(i int, j int);
+	CREATE TABLE data_s2(i int, j int);
+}
+
+teardown
+{
+	DROP TABLE repack_deadlock;
+	DROP EXTENSION injection_points;
+
+	DROP TABLE relfilenodes;
+	DROP TABLE data_s1;
+	DROP TABLE data_s2;
+}
+
+session s1
+setup
+{
+	SELECT injection_points_set_local();
+	SELECT injection_points_attach('repack-concurrently-before-lock', 'wait');
+}
+# Perform the initial load and wait for s2 to do some data changes.
+step wait_before_lock
+{
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+}
+# Check the table from the perspective of s1.
+#
+# Besides the contents, we also check that relfilenode has changed.
+
+# Have each session write the contents into a table and use FULL JOIN to check
+# if the outputs are identical.
+step check1
+{
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+}
+teardown
+{
+	SELECT injection_points_detach('repack-concurrently-before-lock');
+}
+
+session s2
+# Change the existing data. UPDATE changes both key and non-key columns. Also
+# update one row twice to test whether tuple version generated by this session
+# can be found.
+step add_column
+{
+	alter table repack_deadlock add column noise text;
+}
+
+step wakeup_before_lock
+{
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+}
+
+# Test if data changes introduced while one session is performing REPACK
+# CONCURRENTLY find their way into the table.
+permutation
+	wait_before_lock
+	add_column
+	wakeup_before_lock
+	check1
-- 
2.47.3


--qfkt2ktdpcfeypib
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v48-0007-Teach-snapshot-builder-to-skip-transactions-runn.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread

* [PATCH v46 6/7] Error out any process that would block at REPACK
@ 2026-03-25 19:35 Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Álvaro Herrera @ 2026-03-25 19:35 UTC (permalink / raw)

Any process waiting on REPACK to release its lock would actually cause
it to deadlock when it tries to upgrade its lock to AEL, losing all work
done to that point.  We avoid this by teaching the deadlock detector to
raise an error when this condition is detected.
---
 src/backend/commands/repack.c                 | 53 ++++++++----
 src/backend/storage/lmgr/deadlock.c           | 15 ++++
 src/include/storage/proc.h                    |  6 +-
 src/test/modules/injection_points/Makefile    |  1 +
 .../expected/repack_deadlock.out              | 63 ++++++++++++++
 src/test/modules/injection_points/meson.build |  1 +
 .../specs/repack_deadlock.spec                | 83 +++++++++++++++++++
 7 files changed, 202 insertions(+), 20 deletions(-)
 create mode 100644 src/test/modules/injection_points/expected/repack_deadlock.out
 create mode 100644 src/test/modules/injection_points/specs/repack_deadlock.spec

diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c
index a58277bb6c1..21ab81c6f7f 100644
--- a/src/backend/commands/repack.c
+++ b/src/backend/commands/repack.c
@@ -276,6 +276,16 @@ ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
 	/* Determine the lock mode to use. */
 	lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
 
+	/*
+	 * If in concurrent mode, set the PROC_IN_CONCURRENT_REPACK flag.  This
+	 * makes the deadlock checker cause anyone that would conflict with us
+	 * to error out.  It's important to set this flag ahead of actually locking
+	 * the relation; it won't of course affect anyone until we do have a lock
+	 * that others can conflict with.
+	 */
+	if ((params.options & CLUOPT_CONCURRENT) != 0)
+		MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK;
+
 	/*
 	 * If a single relation is specified, process it and we're done ... unless
 	 * the relation is a partitioned table, in which case we fall through.
@@ -476,11 +486,8 @@ RepackLockLevel(bool concurrent)
  * If indexOid is InvalidOid, the table will be rewritten in physical order
  * instead of index order.
  *
- * Note that, in the concurrent case, the function releases the lock at some
- * point, in order to get AccessExclusiveLock for the final steps (i.e. to
- * swap the relation files). To make things simpler, the caller should expect
- * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
- * AccessExclusiveLock is kept till the end of the transaction.)
+ * On return, OldHeap is closed but locked with AccessExclusiveLock - the lock
+ * will be released at end of the transaction.
  *
  * 'cmd' indicates which command is being executed, to be used for error
  * messages.
@@ -512,10 +519,12 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
 		/*
 		 * Make sure we're not in a transaction block.
 		 *
-		 * The reason is that repack_setup_logical_decoding() could deadlock
-		 * if there's an XID already assigned.  It would be possible to run in
-		 * a transaction block if we had no XID, but this restriction is
-		 * simpler for users to understand and we don't lose anything.
+		 * The reason is that repack_setup_logical_decoding() could wait
+		 * indefinitely for our XID to complete. (The deadlock detector would
+		 * not recognize it because we'd be waiting for ourselves, i.e. no
+		 * real lock conflict.) It would be possible to run in a transaction
+		 * block if we had no XID, but this restriction is simpler for users
+		 * to understand and we don't lose anything.
 		 */
 		PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
 
@@ -998,10 +1007,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose,
 		 * Note that the worker has to wait for all transactions with XID
 		 * already assigned to finish. If some of those transactions is
 		 * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
-		 * table (e.g.  it runs CREATE INDEX), we can end up in a deadlock.
-		 * Not sure this risk is worth unlocking/locking the table (and its
-		 * clustering index) and checking again if it's still eligible for
-		 * REPACK CONCURRENTLY.
+		 * table (e.g. it runs CREATE INDEX), it should encounter ERROR in the
+		 * deadlock checking code.
 		 */
 		start_repack_decoding_worker(tableOid);
 
@@ -3123,10 +3130,18 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
 		LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
 
 	/*
-	 * Tuples and pages of the old heap will be gone, but the heap will stay.
+	 * Now that we have all access-exclusive locks on all relations, we no
+	 * longer want other processes to error out when trying to acquire a
+	 * conflicting lock.  Therefore, unset our flag.
+	 */
+	MyProc->statusFlags &= ~PROC_IN_CONCURRENT_REPACK;
+
+	/*
+	 * Tuples and pages of the old heap will be gone, but the heap itself will
+	 * stay.  In order for predicate locks to continue to work, convert them
+	 * to relation-level locks.  We do this both for table and indexes.
 	 */
 	TransferPredicateLocksToHeapRelation(OldHeap);
-	/* The same for indexes. */
 	for (int i = 0; i < nind; i++)
 	{
 		Relation	index = ind_refs[i];
@@ -3346,9 +3361,11 @@ start_repack_decoding_worker(Oid relid)
 
 	/*
 	 * The decoding setup must be done before the caller can have XID assigned
-	 * for any reason, otherwise the worker might end up in a deadlock,
-	 * waiting for the caller's transaction to end. Therefore wait here until
-	 * the worker indicates that it has the logical decoding initialized.
+	 * for any reason, otherwise the worker might end up waiting for the
+	 * caller's transaction to end. (Deadlock detector does not consider this
+	 * a conflict because the worker is in the same locking group as the
+	 * backend that launched it.) Therefore wait here until the worker
+	 * indicates that it has the logical decoding initialized.
 	 */
 	ConditionVariablePrepareToSleep(&shared->cv);
 	for (;;)
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index b8962d875b6..3d3ec0da111 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -620,6 +620,21 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 						proc->statusFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
+					/*
+					 * Similarly, if we note that we're blocked by some process
+					 * running REPACK (CONCURRENTLY), just fail.  That process
+					 * is going to upgrade its lock at some point, and it would
+					 * be inappropriate for any other process to cause that
+					 * to fail.
+					 */
+					if (checkProc == MyProc &&
+						proc->statusFlags & PROC_IN_CONCURRENT_REPACK)
+						ereport(ERROR,
+								errcode(ERRCODE_OBJECT_IN_USE),
+								errmsg("could not wait for concurrent REPACK"),
+								errdetail("Process %d waits for REPACK running on process %d",
+									   MyProc->pid, proc->pid));
+
 					/* We're done looking at this proclock */
 					break;
 				}
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1dad125706e..8ad9718f3d6 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -69,10 +69,12 @@ struct XidCache
 #define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be
 												 * included in vacuum horizons
 												 * in all databases */
+#define		PROC_IN_CONCURRENT_REPACK	0x40	/* REPACK (CONCURRENTLY) */
 
-/* flags reset at EOXact */
+/* flags reset at EOXact.  A bit of a misnomer ... */
 #define		PROC_VACUUM_STATE_MASK \
-	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)
+	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | \
+	 PROC_IN_CONCURRENT_REPACK)
 
 /*
  * Xmin-related flags. Make sure any flags that affect how the process' Xmin
diff --git a/src/test/modules/injection_points/Makefile b/src/test/modules/injection_points/Makefile
index 2cd7d87c533..f7663859fe2 100644
--- a/src/test/modules/injection_points/Makefile
+++ b/src/test/modules/injection_points/Makefile
@@ -15,6 +15,7 @@ REGRESS_OPTS = --dlpath=$(top_builddir)/src/test/regress
 ISOLATION = basic \
 	    inplace \
 	    repack \
+	    repack_deadlock \
 	    repack_toast \
 	    syscache-update-pruned \
 	    heap_lock_update
diff --git a/src/test/modules/injection_points/expected/repack_deadlock.out b/src/test/modules/injection_points/expected/repack_deadlock.out
new file mode 100644
index 00000000000..a86e4767536
--- /dev/null
+++ b/src/test/modules/injection_points/expected/repack_deadlock.out
@@ -0,0 +1,63 @@
+Parsed test spec with 2 sessions
+
+starting permutation: wait_before_lock add_column wakeup_before_lock check1
+injection_points_attach
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: 
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+ <waiting ...>
+step add_column: 
+	alter table repack_deadlock add column noise text;
+ <waiting ...>
+step add_column: <... completed>
+ERROR:  could not wait for concurrent REPACK
+step wakeup_before_lock: 
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+
+injection_points_wakeup
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: <... completed>
+step check1: 
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+
+count
+-----
+    1
+(1 row)
+
+i|j
+-+-
+1|1
+2|2
+3|3
+4|4
+(4 rows)
+
+count
+-----
+    4
+(1 row)
+
+injection_points_detach
+-----------------------
+                       
+(1 row)
+
diff --git a/src/test/modules/injection_points/meson.build b/src/test/modules/injection_points/meson.build
index a414abb924b..1cd88d6db65 100644
--- a/src/test/modules/injection_points/meson.build
+++ b/src/test/modules/injection_points/meson.build
@@ -46,6 +46,7 @@ tests += {
       'basic',
       'inplace',
       'repack',
+      'repack_deadlock',
       'repack_toast',
       'syscache-update-pruned',
       'heap_lock_update',
diff --git a/src/test/modules/injection_points/specs/repack_deadlock.spec b/src/test/modules/injection_points/specs/repack_deadlock.spec
new file mode 100644
index 00000000000..9d23a6588c2
--- /dev/null
+++ b/src/test/modules/injection_points/specs/repack_deadlock.spec
@@ -0,0 +1,83 @@
+# Test REPACK with a concurrent transaction that would cause a deadlock
+setup
+{
+	CREATE EXTENSION injection_points;
+
+	CREATE TABLE repack_deadlock(i int PRIMARY KEY, j int);
+	INSERT INTO repack_deadlock(i, j) VALUES (1, 1), (2, 2), (3, 3), (4, 4);
+
+	CREATE TABLE relfilenodes(node oid);
+
+	CREATE TABLE data_s1(i int, j int);
+	CREATE TABLE data_s2(i int, j int);
+}
+
+teardown
+{
+	DROP TABLE repack_deadlock;
+	DROP EXTENSION injection_points;
+
+	DROP TABLE relfilenodes;
+	DROP TABLE data_s1;
+	DROP TABLE data_s2;
+}
+
+session s1
+setup
+{
+	SELECT injection_points_set_local();
+	SELECT injection_points_attach('repack-concurrently-before-lock', 'wait');
+}
+# Perform the initial load and wait for s2 to do some data changes.
+step wait_before_lock
+{
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+}
+# Check the table from the perspective of s1.
+#
+# Besides the contents, we also check that relfilenode has changed.
+
+# Have each session write the contents into a table and use FULL JOIN to check
+# if the outputs are identical.
+step check1
+{
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+}
+teardown
+{
+	SELECT injection_points_detach('repack-concurrently-before-lock');
+}
+
+session s2
+# Change the existing data. UPDATE changes both key and non-key columns. Also
+# update one row twice to test whether tuple version generated by this session
+# can be found.
+step add_column
+{
+	alter table repack_deadlock add column noise text;
+}
+
+step wakeup_before_lock
+{
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+}
+
+# Test if data changes introduced while one session is performing REPACK
+# CONCURRENTLY find their way into the table.
+permutation
+	wait_before_lock
+	add_column
+	wakeup_before_lock
+	check1
-- 
2.47.3


--mdsv5bdgfjjj6d77
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v46-0007-Teach-snapshot-builder-to-skip-transactions-runn.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread

* [PATCH v45 7/8] Error out any process that would block at REPACK
@ 2026-03-25 19:35 Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 664+ messages in thread

From: Álvaro Herrera @ 2026-03-25 19:35 UTC (permalink / raw)

Any process waiting on REPACK to release its lock would actually cause
it to deadlock when it tries to upgrade its lock to AEL, losing all work
done to that point.  We avoid this by teaching the deadlock detector to
raise an error when this condition is detected.
---
 src/backend/commands/repack.c                 | 53 ++++++++----
 src/backend/storage/lmgr/deadlock.c           | 15 ++++
 src/include/storage/proc.h                    |  6 +-
 src/test/modules/injection_points/Makefile    |  1 +
 .../expected/repack_deadlock.out              | 63 ++++++++++++++
 src/test/modules/injection_points/meson.build |  1 +
 .../specs/repack_deadlock.spec                | 83 +++++++++++++++++++
 7 files changed, 202 insertions(+), 20 deletions(-)
 create mode 100644 src/test/modules/injection_points/expected/repack_deadlock.out
 create mode 100644 src/test/modules/injection_points/specs/repack_deadlock.spec

diff --git a/src/backend/commands/repack.c b/src/backend/commands/repack.c
index d25ab43da7b..7291b4180ab 100644
--- a/src/backend/commands/repack.c
+++ b/src/backend/commands/repack.c
@@ -276,6 +276,16 @@ ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel)
 	/* Determine the lock mode to use. */
 	lockmode = RepackLockLevel((params.options & CLUOPT_CONCURRENT) != 0);
 
+	/*
+	 * If in concurrent mode, set the PROC_IN_CONCURRENT_REPACK flag.  This
+	 * makes the deadlock checker cause anyone that would conflict with us
+	 * to error out.  It's important to set this flag ahead of actually locking
+	 * the relation; it won't of course affect anyone until we do have a lock
+	 * that others can conflict with.
+	 */
+	if ((params.options & CLUOPT_CONCURRENT) != 0)
+		MyProc->statusFlags |= PROC_IN_CONCURRENT_REPACK;
+
 	/*
 	 * If a single relation is specified, process it and we're done ... unless
 	 * the relation is a partitioned table, in which case we fall through.
@@ -476,11 +486,8 @@ RepackLockLevel(bool concurrent)
  * If indexOid is InvalidOid, the table will be rewritten in physical order
  * instead of index order.
  *
- * Note that, in the concurrent case, the function releases the lock at some
- * point, in order to get AccessExclusiveLock for the final steps (i.e. to
- * swap the relation files). To make things simpler, the caller should expect
- * OldHeap to be closed on return, regardless CLUOPT_CONCURRENT. (The
- * AccessExclusiveLock is kept till the end of the transaction.)
+ * On return, OldHeap is closed but locked with AccessExclusiveLock - the lock
+ * will be released at end of the transaction.
  *
  * 'cmd' indicates which command is being executed, to be used for error
  * messages.
@@ -512,10 +519,12 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
 		/*
 		 * Make sure we're not in a transaction block.
 		 *
-		 * The reason is that repack_setup_logical_decoding() could deadlock
-		 * if there's an XID already assigned.  It would be possible to run in
-		 * a transaction block if we had no XID, but this restriction is
-		 * simpler for users to understand and we don't lose anything.
+		 * The reason is that repack_setup_logical_decoding() could wait
+		 * indefinitely for our XID to complete. (The deadlock detector would
+		 * not recognize it because we'd be waiting for ourselves, i.e. no
+		 * real lock conflict.) It would be possible to run in a transaction
+		 * block if we had no XID, but this restriction is simpler for users
+		 * to understand and we don't lose anything.
 		 */
 		PreventInTransactionBlock(isTopLevel, "REPACK (CONCURRENTLY)");
 
@@ -998,10 +1007,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose,
 		 * Note that the worker has to wait for all transactions with XID
 		 * already assigned to finish. If some of those transactions is
 		 * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
-		 * table (e.g.  it runs CREATE INDEX), we can end up in a deadlock.
-		 * Not sure this risk is worth unlocking/locking the table (and its
-		 * clustering index) and checking again if it's still eligible for
-		 * REPACK CONCURRENTLY.
+		 * table (e.g. it runs CREATE INDEX), it should encounter ERROR in the
+		 * deadlock checking code.
 		 */
 		start_repack_decoding_worker(tableOid);
 
@@ -3119,10 +3126,18 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
 		LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
 
 	/*
-	 * Tuples and pages of the old heap will be gone, but the heap will stay.
+	 * Now that we have all access-exclusive locks on all relations, we no
+	 * longer want other processes to error out when trying to acquire a
+	 * conflicting lock.  Therefore, unset our flag.
+	 */
+	MyProc->statusFlags &= ~PROC_IN_CONCURRENT_REPACK;
+
+	/*
+	 * Tuples and pages of the old heap will be gone, but the heap itself will
+	 * stay.  In order for predicate locks to continue to work, convert them
+	 * to relation-level locks.  We do this both for table and indexes.
 	 */
 	TransferPredicateLocksToHeapRelation(OldHeap);
-	/* The same for indexes. */
 	for (int i = 0; i < nind; i++)
 	{
 		Relation	index = ind_refs[i];
@@ -3342,9 +3357,11 @@ start_repack_decoding_worker(Oid relid)
 
 	/*
 	 * The decoding setup must be done before the caller can have XID assigned
-	 * for any reason, otherwise the worker might end up in a deadlock,
-	 * waiting for the caller's transaction to end. Therefore wait here until
-	 * the worker indicates that it has the logical decoding initialized.
+	 * for any reason, otherwise the worker might end up waiting for the
+	 * caller's transaction to end. (Deadlock detector does not consider this
+	 * a conflict because the worker is in the same locking group as the
+	 * backend that launched it.) Therefore wait here until the worker
+	 * indicates that it has the logical decoding initialized.
 	 */
 	ConditionVariablePrepareToSleep(&shared->cv);
 	for (;;)
diff --git a/src/backend/storage/lmgr/deadlock.c b/src/backend/storage/lmgr/deadlock.c
index b8962d875b6..3d3ec0da111 100644
--- a/src/backend/storage/lmgr/deadlock.c
+++ b/src/backend/storage/lmgr/deadlock.c
@@ -620,6 +620,21 @@ FindLockCycleRecurseMember(PGPROC *checkProc,
 						proc->statusFlags & PROC_IS_AUTOVACUUM)
 						blocking_autovacuum_proc = proc;
 
+					/*
+					 * Similarly, if we note that we're blocked by some process
+					 * running REPACK (CONCURRENTLY), just fail.  That process
+					 * is going to upgrade its lock at some point, and it would
+					 * be inappropriate for any other process to cause that
+					 * to fail.
+					 */
+					if (checkProc == MyProc &&
+						proc->statusFlags & PROC_IN_CONCURRENT_REPACK)
+						ereport(ERROR,
+								errcode(ERRCODE_OBJECT_IN_USE),
+								errmsg("could not wait for concurrent REPACK"),
+								errdetail("Process %d waits for REPACK running on process %d",
+									   MyProc->pid, proc->pid));
+
 					/* We're done looking at this proclock */
 					break;
 				}
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1dad125706e..8ad9718f3d6 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -69,10 +69,12 @@ struct XidCache
 #define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be
 												 * included in vacuum horizons
 												 * in all databases */
+#define		PROC_IN_CONCURRENT_REPACK	0x40	/* REPACK (CONCURRENTLY) */
 
-/* flags reset at EOXact */
+/* flags reset at EOXact.  A bit of a misnomer ... */
 #define		PROC_VACUUM_STATE_MASK \
-	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)
+	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND | \
+	 PROC_IN_CONCURRENT_REPACK)
 
 /*
  * Xmin-related flags. Make sure any flags that affect how the process' Xmin
diff --git a/src/test/modules/injection_points/Makefile b/src/test/modules/injection_points/Makefile
index 2cd7d87c533..f7663859fe2 100644
--- a/src/test/modules/injection_points/Makefile
+++ b/src/test/modules/injection_points/Makefile
@@ -15,6 +15,7 @@ REGRESS_OPTS = --dlpath=$(top_builddir)/src/test/regress
 ISOLATION = basic \
 	    inplace \
 	    repack \
+	    repack_deadlock \
 	    repack_toast \
 	    syscache-update-pruned \
 	    heap_lock_update
diff --git a/src/test/modules/injection_points/expected/repack_deadlock.out b/src/test/modules/injection_points/expected/repack_deadlock.out
new file mode 100644
index 00000000000..a86e4767536
--- /dev/null
+++ b/src/test/modules/injection_points/expected/repack_deadlock.out
@@ -0,0 +1,63 @@
+Parsed test spec with 2 sessions
+
+starting permutation: wait_before_lock add_column wakeup_before_lock check1
+injection_points_attach
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: 
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+ <waiting ...>
+step add_column: 
+	alter table repack_deadlock add column noise text;
+ <waiting ...>
+step add_column: <... completed>
+ERROR:  could not wait for concurrent REPACK
+step wakeup_before_lock: 
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+
+injection_points_wakeup
+-----------------------
+                       
+(1 row)
+
+step wait_before_lock: <... completed>
+step check1: 
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+
+count
+-----
+    1
+(1 row)
+
+i|j
+-+-
+1|1
+2|2
+3|3
+4|4
+(4 rows)
+
+count
+-----
+    4
+(1 row)
+
+injection_points_detach
+-----------------------
+                       
+(1 row)
+
diff --git a/src/test/modules/injection_points/meson.build b/src/test/modules/injection_points/meson.build
index a414abb924b..1cd88d6db65 100644
--- a/src/test/modules/injection_points/meson.build
+++ b/src/test/modules/injection_points/meson.build
@@ -46,6 +46,7 @@ tests += {
       'basic',
       'inplace',
       'repack',
+      'repack_deadlock',
       'repack_toast',
       'syscache-update-pruned',
       'heap_lock_update',
diff --git a/src/test/modules/injection_points/specs/repack_deadlock.spec b/src/test/modules/injection_points/specs/repack_deadlock.spec
new file mode 100644
index 00000000000..9d23a6588c2
--- /dev/null
+++ b/src/test/modules/injection_points/specs/repack_deadlock.spec
@@ -0,0 +1,83 @@
+# Test REPACK with a concurrent transaction that would cause a deadlock
+setup
+{
+	CREATE EXTENSION injection_points;
+
+	CREATE TABLE repack_deadlock(i int PRIMARY KEY, j int);
+	INSERT INTO repack_deadlock(i, j) VALUES (1, 1), (2, 2), (3, 3), (4, 4);
+
+	CREATE TABLE relfilenodes(node oid);
+
+	CREATE TABLE data_s1(i int, j int);
+	CREATE TABLE data_s2(i int, j int);
+}
+
+teardown
+{
+	DROP TABLE repack_deadlock;
+	DROP EXTENSION injection_points;
+
+	DROP TABLE relfilenodes;
+	DROP TABLE data_s1;
+	DROP TABLE data_s2;
+}
+
+session s1
+setup
+{
+	SELECT injection_points_set_local();
+	SELECT injection_points_attach('repack-concurrently-before-lock', 'wait');
+}
+# Perform the initial load and wait for s2 to do some data changes.
+step wait_before_lock
+{
+	REPACK (CONCURRENTLY) repack_deadlock USING INDEX repack_deadlock_pkey;
+}
+# Check the table from the perspective of s1.
+#
+# Besides the contents, we also check that relfilenode has changed.
+
+# Have each session write the contents into a table and use FULL JOIN to check
+# if the outputs are identical.
+step check1
+{
+	INSERT INTO relfilenodes(node)
+	SELECT relfilenode FROM pg_class WHERE relname='repack_deadlock';
+
+	SELECT count(DISTINCT node) FROM relfilenodes;
+
+	SELECT i, j FROM repack_deadlock ORDER BY i, j;
+
+	INSERT INTO data_s1(i, j)
+	SELECT i, j FROM repack_deadlock;
+
+	SELECT count(*)
+	FROM data_s1 d1 FULL JOIN data_s2 d2 USING (i, j)
+	WHERE d1.i ISNULL OR d2.i ISNULL;
+}
+teardown
+{
+	SELECT injection_points_detach('repack-concurrently-before-lock');
+}
+
+session s2
+# Change the existing data. UPDATE changes both key and non-key columns. Also
+# update one row twice to test whether tuple version generated by this session
+# can be found.
+step add_column
+{
+	alter table repack_deadlock add column noise text;
+}
+
+step wakeup_before_lock
+{
+	SELECT injection_points_wakeup('repack-concurrently-before-lock');
+}
+
+# Test if data changes introduced while one session is performing REPACK
+# CONCURRENTLY find their way into the table.
+permutation
+	wait_before_lock
+	add_column
+	wakeup_before_lock
+	check1
-- 
2.47.3


--cldir6umqisr673d
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v45-0008-Teach-snapshot-builder-to-skip-transactions-runn.patch"



^ permalink  raw  reply  [nested|flat] 664+ messages in thread