Restore the portal-level snapshot after procedure COMMIT/ROLLBACK.

COMMIT/ROLLBACK necessarily destroys all snapshots within the session.
The original implementation of intra-procedure transactions just
cavalierly did that, ignoring the fact that this left us executing in
a rather different environment than normal.  In particular, it turns
out that handling of toasted datums depends rather critically on there
being an outer ActiveSnapshot: otherwise, when SPI or the core
executor pop whatever snapshot they used and return, it's unsafe to
dereference any toasted datums that may appear in the query result.
It's possible to demonstrate "no known snapshots" and "missing chunk
number N for toast value" errors as a result of this oversight.

Historically this outer snapshot has been held by the Portal code,
and that seems like a good plan to preserve.  So add infrastructure
to pquery.c to allow re-establishing the Portal-owned snapshot if it's
not there anymore, and add enough bookkeeping support that we can tell
whether it is or not.

We can't, however, just re-establish the Portal snapshot as part of
COMMIT/ROLLBACK.  As in normal transaction start, acquiring the first
snapshot should wait until after SET and LOCK commands.  Hence, teach
spi.c about doing this at the right time.  (Note that this patch
doesn't fix the problem for any PLs that try to run intra-procedure
transactions without using SPI to execute SQL commands.)

This makes SPI's no_snapshots parameter rather a misnomer, so in HEAD,
rename that to allow_nonatomic.

replication/logical/worker.c also needs some fixes, because it wasn't
careful to hold a snapshot open around AFTER trigger execution.
That code doesn't use a Portal, which I suspect someday we're gonna
have to fix.  But for now, just rearrange the order of operations.
This includes back-patching the recent addition of finish_estate()
to centralize the cleanup logic there.

This also back-patches commit 2ecfeda into v13, to improve the
test coverage for worker.c (it was that test that exposed that
worker.c's snapshot management is wrong).

Per bug #15990 from Andreas Wicht.  Back-patch to v11 where
intra-procedure COMMIT was added.

Discussion: https://postgr.es/m/[email protected]

src/backend/commands/functioncmds.c

@@ -61,6 +61,7 @@
 #include "parser/parse_func.h"
 #include "parser/parse_type.h"
 #include "pgstat.h"
+#include "tcop/pquery.h"
 #include "utils/acl.h"
 #include "utils/builtins.h"
 #include "utils/fmgroids.h"
@@ -2342,6 +2343,20 @@ ExecuteCallStmt(CallStmt *stmt, ParamListInfo params, bool atomic, DestReceiver
 		if (fcinfo.isnull)
 			elog(ERROR, "procedure returned null record");
 
+		/*
+		 * Ensure there's an active snapshot whilst we execute whatever's
+		 * involved here.  Note that this is *not* sufficient to make the
+		 * world safe for TOAST pointers to be included in the returned data:
+		 * the referenced data could have gone away while we didn't hold a
+		 * snapshot.  Hence, it's incumbent on PLs that can do COMMIT/ROLLBACK
+		 * to not return TOAST pointers, unless those pointers were fetched
+		 * after the last COMMIT/ROLLBACK in the procedure.
+		 *
+		 * XXX that is a really nasty, hard-to-test requirement.  Is there a
+		 * way to remove it?
+		 */
+		EnsurePortalSnapshotExists();
+
 		td = DatumGetHeapTupleHeader(retval);
 		tupType = HeapTupleHeaderGetTypeId(td);
 		tupTypmod = HeapTupleHeaderGetTypMod(td);

src/backend/executor/spi.c

@@ -260,12 +260,8 @@ SPI_commit(void)
 	/* Start the actual commit */
 	_SPI_current->internal_xact = true;
 
-	/*
-	 * Before committing, pop all active snapshots to avoid error about
-	 * "snapshot %p still active".
-	 */
-	while (ActiveSnapshotSet())
-		PopActiveSnapshot();
+	/* Release snapshots associated with portals */
+	ForgetPortalSnapshots();
 
 	CommitTransactionCommand();
 	MemoryContextSwitchTo(oldcontext);
@@ -300,6 +296,9 @@ SPI_rollback(void)
 	/* Start the actual rollback */
 	_SPI_current->internal_xact = true;
 
+	/* Release snapshots associated with portals */
+	ForgetPortalSnapshots();
+
 	AbortCurrentTransaction();
 	MemoryContextSwitchTo(oldcontext);
 
@@ -2102,6 +2101,7 @@ _SPI_execute_plan(SPIPlanPtr plan, ParamListInfo paramLI,
 	Oid			my_lastoid = InvalidOid;
 	SPITupleTable *my_tuptable = NULL;
 	int			res = 0;
+	bool		allow_nonatomic = plan->no_snapshots;	/* legacy API name */
 	bool		pushed_active_snap = false;
 	ErrorContextCallback spierrcontext;
 	CachedPlan *cplan = NULL;
@@ -2134,11 +2134,12 @@ _SPI_execute_plan(SPIPlanPtr plan, ParamListInfo paramLI,
 	 * In the first two cases, we can just push the snap onto the stack once
 	 * for the whole plan list.
 	 *
-	 * But if the plan has no_snapshots set to true, then don't manage
-	 * snapshots at all.  The caller should then take care of that.
+	 * Note that snapshot != InvalidSnapshot implies an atomic execution
+	 * context.
 	 */
-	if (snapshot != InvalidSnapshot && !plan->no_snapshots)
+	if (snapshot != InvalidSnapshot)
 	{
+		Assert(!allow_nonatomic);
 		if (read_only)
 		{
 			PushActiveSnapshot(snapshot);
@@ -2225,15 +2226,39 @@ _SPI_execute_plan(SPIPlanPtr plan, ParamListInfo paramLI,
 		stmt_list = cplan->stmt_list;
 
 		/*
-		 * In the default non-read-only case, get a new snapshot, replacing
-		 * any that we pushed in a previous cycle.
+		 * If we weren't given a specific snapshot to use, and the statement
+		 * list requires a snapshot, set that up.
 		 */
-		if (snapshot == InvalidSnapshot && !read_only && !plan->no_snapshots)
+		if (snapshot == InvalidSnapshot &&
+			(list_length(stmt_list) > 1 ||
+			 (list_length(stmt_list) == 1 &&
+			  PlannedStmtRequiresSnapshot(linitial_node(PlannedStmt,
+														stmt_list)))))
 		{
-			if (pushed_active_snap)
-				PopActiveSnapshot();
-			PushActiveSnapshot(GetTransactionSnapshot());
-			pushed_active_snap = true;
+			/*
+			 * First, ensure there's a Portal-level snapshot.  This back-fills
+			 * the snapshot stack in case the previous operation was a COMMIT
+			 * or ROLLBACK inside a procedure or DO block.  (We can't put back
+			 * the Portal snapshot any sooner, or we'd break cases like doing
+			 * SET or LOCK just after COMMIT.)  It's enough to check once per
+			 * statement list, since COMMIT/ROLLBACK/CALL/DO can't appear
+			 * within a multi-statement list.
+			 */
+			EnsurePortalSnapshotExists();
+
+			/*
+			 * In the default non-read-only case, get a new per-statement-list
+			 * snapshot, replacing any that we pushed in a previous cycle.
+			 * Skip it when doing non-atomic execution, though (we rely
+			 * entirely on the Portal snapshot in that case).
+			 */
+			if (!read_only && !allow_nonatomic)
+			{
+				if (pushed_active_snap)
+					PopActiveSnapshot();
+				PushActiveSnapshot(GetTransactionSnapshot());
+				pushed_active_snap = true;
+			}
 		}
 
 		foreach(lc2, stmt_list)
@@ -2246,6 +2271,7 @@ _SPI_execute_plan(SPIPlanPtr plan, ParamListInfo paramLI,
 			_SPI_current->lastoid = InvalidOid;
 			_SPI_current->tuptable = NULL;
 
+			/* Check for unsupported cases. */
 			if (stmt->utilityStmt)
 			{
 				if (IsA(stmt->utilityStmt, CopyStmt))
@@ -2277,9 +2303,10 @@ _SPI_execute_plan(SPIPlanPtr plan, ParamListInfo paramLI,
 
 			/*
 			 * If not read-only mode, advance the command counter before each
-			 * command and update the snapshot.
+			 * command and update the snapshot.  (But skip it if the snapshot
+			 * isn't under our control.)
 			 */
-			if (!read_only && !plan->no_snapshots)
+			if (!read_only && pushed_active_snap)
 			{
 				CommandCounterIncrement();
 				UpdateActiveSnapshotCommandId();
@@ -2313,13 +2340,11 @@ _SPI_execute_plan(SPIPlanPtr plan, ParamListInfo paramLI,
 				ProcessUtilityContext context;
 
 				/*
-				 * If the SPI context is atomic, or we are asked to manage
-				 * snapshots, then we are in an atomic execution context.
-				 * Conversely, to propagate a nonatomic execution context, the
-				 * caller must be in a nonatomic SPI context and manage
-				 * snapshots itself.
+				 * If the SPI context is atomic, or we were not told to allow
+				 * nonatomic operations, tell ProcessUtility this is an atomic
+				 * execution context.
 				 */
-				if (_SPI_current->atomic || !plan->no_snapshots)
+				if (_SPI_current->atomic || !allow_nonatomic)
 					context = PROCESS_UTILITY_QUERY;
 				else
 					context = PROCESS_UTILITY_QUERY_NONATOMIC;

src/backend/replication/logical/worker.c

@@ -194,6 +194,13 @@ create_estate_for_relation(LogicalRepRelMapEntry *rel)
 	ResultRelInfo *resultRelInfo;
 	RangeTblEntry *rte;
 
+	/*
+	 * Input functions may need an active snapshot, as may AFTER triggers
+	 * invoked during finish_estate.  For safety, ensure an active snapshot
+	 * exists throughout all our usage of the executor.
+	 */
+	PushActiveSnapshot(GetTransactionSnapshot());
+
 	estate = CreateExecutorState();
 
 	rte = makeNode(RangeTblEntry);
@@ -221,6 +228,22 @@ create_estate_for_relation(LogicalRepRelMapEntry *rel)
 	return estate;
 }
 
+/*
+ * Finish any operations related to the executor state created by
+ * create_estate_for_relation().
+ */
+static void
+finish_estate(EState *estate)
+{
+	/* Handle any queued AFTER triggers. */
+	AfterTriggerEndQuery(estate);
+
+	/* Cleanup. */
+	ExecResetTupleTable(estate->es_tupleTable, false);
+	FreeExecutorState(estate);
+	PopActiveSnapshot();
+}
+
 /*
  * Executes default values for columns for which we can't map to remote
  * relation columns.
@@ -627,9 +650,6 @@ apply_handle_insert(StringInfo s)
 	remoteslot = ExecInitExtraTupleSlot(estate,
 										RelationGetDescr(rel->localrel));
 
-	/* Input functions may need an active snapshot, so get one */
-	PushActiveSnapshot(GetTransactionSnapshot());
-
 	/* Process and store remote tuple in the slot */
 	oldctx = MemoryContextSwitchTo(GetPerTupleMemoryContext(estate));
 	slot_store_cstrings(remoteslot, rel, newtup.values);
@@ -643,13 +663,8 @@ apply_handle_insert(StringInfo s)
 
 	/* Cleanup. */
 	ExecCloseIndices(estate->es_result_relation_info);
-	PopActiveSnapshot();
 
-	/* Handle queued AFTER triggers. */
-	AfterTriggerEndQuery(estate);
-
-	ExecResetTupleTable(estate->es_tupleTable, false);
-	FreeExecutorState(estate);
+	finish_estate(estate);
 
 	logicalrep_rel_close(rel, NoLock);
 
@@ -760,7 +775,6 @@ apply_handle_update(StringInfo s)
 		}
 	}
 
-	PushActiveSnapshot(GetTransactionSnapshot());
 	ExecOpenIndices(estate->es_result_relation_info, false);
 
 	/* Build the search tuple. */
@@ -819,15 +833,10 @@ apply_handle_update(StringInfo s)
 	}
 
 	/* Cleanup. */
+	EvalPlanQualEnd(&epqstate);
 	ExecCloseIndices(estate->es_result_relation_info);
-	PopActiveSnapshot();
 
-	/* Handle queued AFTER triggers. */
-	AfterTriggerEndQuery(estate);
-
-	EvalPlanQualEnd(&epqstate);
-	ExecResetTupleTable(estate->es_tupleTable, false);
-	FreeExecutorState(estate);
+	finish_estate(estate);
 
 	logicalrep_rel_close(rel, NoLock);
 
@@ -878,7 +887,6 @@ apply_handle_delete(StringInfo s)
 									   RelationGetDescr(rel->localrel));
 	EvalPlanQualInit(&epqstate, estate, NULL, NIL, -1);
 
-	PushActiveSnapshot(GetTransactionSnapshot());
 	ExecOpenIndices(estate->es_result_relation_info, false);
 
 	/* Find the tuple using the replica identity index. */
@@ -919,15 +927,10 @@ apply_handle_delete(StringInfo s)
 	}
 
 	/* Cleanup. */
+	EvalPlanQualEnd(&epqstate);
 	ExecCloseIndices(estate->es_result_relation_info);
-	PopActiveSnapshot();
 
-	/* Handle queued AFTER triggers. */
-	AfterTriggerEndQuery(estate);
-
-	EvalPlanQualEnd(&epqstate);
-	ExecResetTupleTable(estate->es_tupleTable, false);
-	FreeExecutorState(estate);
+	finish_estate(estate);
 
 	logicalrep_rel_close(rel, NoLock);