Fix data loss in logical replication.

Amit Kapila · Amit Kapila · commit cadaf0ac4637 · 2025-04-10T12:57:10.000+05:30
Data loss can happen when the DDLs like ALTER PUBLICATION ... ADD TABLE ... or ALTER TYPE ... that don't take a strong lock on table happens concurrently to DMLs on the tables involved in the DDL. This happens because logical decoding doesn't distribute invalidations to concurrent transactions and those transactions use stale cache data to decode the changes. The problem becomes bigger because we keep using the stale cache even after those in-progress transactions are finished and skip the changes required to be sent to the client. This commit fixes the issue by distributing invalidation messages from catalog-modifying transactions to all concurrent in-progress transactions. This allows the necessary rebuild of the catalog cache when decoding new changes after concurrent DDL. We observed performance regression primarily during frequent execution of *publication DDL* statements that modify the published tables. The regression is minor or nearly nonexistent for DDLs that do not affect the published tables or occur infrequently, making this a worthwhile cost to resolve a longstanding data loss issue. An alternative approach considered was to take a strong lock on each affected table during publication modification. However, this would only address issues related to publication DDLs (but not the ALTER TYPE ...) and require locking every relation in the database for publications created as FOR ALL TABLES, which is impractical. The bug exists in all supported branches, but we are backpatching till 14. The fix for 13 requires somewhat bigger changes than this fix, so the fix for that branch is still under discussion. Reported-by: hubert depesz lubaczewski <depesz@depesz.com> Reported-by: Tomas Vondra <tomas.vondra@enterprisedb.com> Author: Shlok Kyal <shlok.kyal.oss@gmail.com> Author: Hayato Kuroda <kuroda.hayato@fujitsu.com> Reviewed-by: Zhijie Hou <houzj.fnst@fujitsu.com> Reviewed-by: Masahiko Sawada <sawada.mshk@gmail.com> Reviewed-by: Amit Kapila <amit.kapila16@gmail.com> Tested-by: Benoit Lobréau <benoit.lobreau@dalibo.com> Backpatch-through: 14 Discussion: https://postgr.es/m/de52b282-1166-1180-45a2-8d8917ca74c6@enterprisedb.com Discussion: https://postgr.es/m/CAD21AoAenVqiMjpN-PvGHL1N9DWnHSq673bfgr6phmBUzx=kLQ@mail.gmail.com
diff --git a/contrib/test_decoding/Makefile b/contrib/test_decoding/Makefile
@@ -9,7 +9,7 @@ REGRESS = ddl xact rewrite toast permissions decoding_in_xact \
 ISOLATION = mxact delayed_startup ondisk_startup concurrent_ddl_dml \
 	oldest_xmin snapshot_transfer subxact_without_top concurrent_stream \
 	twophase_snapshot slot_creation_error catalog_change_snapshot \
-	skip_snapshot_restore
+	skip_snapshot_restore invalidation_distrubution
 
 REGRESS_OPTS = --temp-config $(top_srcdir)/contrib/test_decoding/logical.conf
 ISOLATION_OPTS = --temp-config $(top_srcdir)/contrib/test_decoding/logical.conf
diff --git a/contrib/test_decoding/expected/invalidation_distrubution.out b/contrib/test_decoding/expected/invalidation_distrubution.out
@@ -0,0 +1,20 @@
+Parsed test spec with 2 sessions
+
+starting permutation: s1_insert_tbl1 s1_begin s1_insert_tbl1 s2_alter_pub_add_tbl s1_commit s1_insert_tbl1 s2_get_binary_changes
+step s1_insert_tbl1: INSERT INTO tbl1 (val1, val2) VALUES (1, 1);
+step s1_begin: BEGIN;
+step s1_insert_tbl1: INSERT INTO tbl1 (val1, val2) VALUES (1, 1);
+step s2_alter_pub_add_tbl: ALTER PUBLICATION pub ADD TABLE tbl1;
+step s1_commit: COMMIT;
+step s1_insert_tbl1: INSERT INTO tbl1 (val1, val2) VALUES (1, 1);
+step s2_get_binary_changes: SELECT count(data) FROM pg_logical_slot_get_binary_changes('isolation_slot', NULL, NULL, 'proto_version', '4', 'publication_names', 'pub') WHERE get_byte(data, 0) = 73;
+count
+-----
+    1
+(1 row)
+
+?column?
+--------
+stop    
+(1 row)
+
diff --git a/contrib/test_decoding/meson.build b/contrib/test_decoding/meson.build
@@ -63,6 +63,7 @@ tests += {
       'twophase_snapshot',
       'slot_creation_error',
       'skip_snapshot_restore',
+      'invalidation_distrubution',
     ],
     'regress_args': [
       '--temp-config', files('logical.conf'),
diff --git a/contrib/test_decoding/specs/invalidation_distrubution.spec b/contrib/test_decoding/specs/invalidation_distrubution.spec
@@ -0,0 +1,32 @@
+# Test that catalog cache invalidation messages are distributed to ongoing
+# transactions, ensuring they can access the updated catalog content after
+# processing these messages.
+setup
+{
+    SELECT 'init' FROM pg_create_logical_replication_slot('isolation_slot', 'pgoutput');
+    CREATE TABLE tbl1(val1 integer, val2 integer);
+    CREATE PUBLICATION pub;
+}
+
+teardown
+{
+    DROP TABLE tbl1;
+    DROP PUBLICATION pub;
+    SELECT 'stop' FROM pg_drop_replication_slot('isolation_slot');
+}
+
+session "s1"
+setup { SET synchronous_commit=on; }
+
+step "s1_begin" { BEGIN; }
+step "s1_insert_tbl1" { INSERT INTO tbl1 (val1, val2) VALUES (1, 1); }
+step "s1_commit" { COMMIT; }
+
+session "s2"
+setup { SET synchronous_commit=on; }
+
+step "s2_alter_pub_add_tbl" { ALTER PUBLICATION pub ADD TABLE tbl1; }
+step "s2_get_binary_changes" { SELECT count(data) FROM pg_logical_slot_get_binary_changes('isolation_slot', NULL, NULL, 'proto_version', '4', 'publication_names', 'pub') WHERE get_byte(data, 0) = 73; }
+
+# Expect to get one insert change. LOGICAL_REP_MSG_INSERT = 'I'
+permutation "s1_insert_tbl1" "s1_begin" "s1_insert_tbl1" "s2_alter_pub_add_tbl" "s1_commit" "s1_insert_tbl1" "s2_get_binary_changes"
diff --git a/src/backend/replication/logical/reorderbuffer.c b/src/backend/replication/logical/reorderbuffer.c
@@ -5337,3 +5337,26 @@ ResolveCminCmaxDuringDecoding(HTAB *tuplecid_data,
 		*cmax = ent->cmax;
 	return true;
 }
+
+/*
+ * Count invalidation messages of specified transaction.
+ *
+ * Returns number of messages, and msgs is set to the pointer of the linked
+ * list for the messages.
+ */
+uint32
+ReorderBufferGetInvalidations(ReorderBuffer *rb, TransactionId xid,
+							  SharedInvalidationMessage **msgs)
+{
+	ReorderBufferTXN *txn;
+
+	txn = ReorderBufferTXNByXid(rb, xid, false, NULL, InvalidXLogRecPtr,
+								false);
+
+	if (txn == NULL)
+		return 0;
+
+	*msgs = txn->invalidations;
+
+	return txn->ninvalidations;
+}
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
@@ -300,7 +300,7 @@ static void SnapBuildFreeSnapshot(Snapshot snap);
 
 static void SnapBuildSnapIncRefcount(Snapshot snap);
 
-static void SnapBuildDistributeNewCatalogSnapshot(SnapBuild *builder, XLogRecPtr lsn);
+static void SnapBuildDistributeSnapshotAndInval(SnapBuild *builder, XLogRecPtr lsn, TransactionId xid);
 
 static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
 												 uint32 xinfo);
@@ -859,23 +859,24 @@ SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 }
 
 /*
- * Add a new Snapshot to all transactions we're decoding that currently are
- * in-progress so they can see new catalog contents made by the transaction
- * that just committed. This is necessary because those in-progress
- * transactions will use the new catalog's contents from here on (at the very
- * least everything they do needs to be compatible with newer catalog
- * contents).
+ * Add a new Snapshot and invalidation messages to all transactions we're
+ * decoding that currently are in-progress so they can see new catalog contents
+ * made by the transaction that just committed. This is necessary because those
+ * in-progress transactions will use the new catalog's contents from here on
+ * (at the very least everything they do needs to be compatible with newer
+ * catalog contents).
  */
 static void
-SnapBuildDistributeNewCatalogSnapshot(SnapBuild *builder, XLogRecPtr lsn)
+SnapBuildDistributeSnapshotAndInval(SnapBuild *builder, XLogRecPtr lsn, TransactionId xid)
 {
 	dlist_iter	txn_i;
 	ReorderBufferTXN *txn;
 
 	/*
 	 * Iterate through all toplevel transactions. This can include
 	 * subtransactions which we just don't yet know to be that, but that's
-	 * fine, they will just get an unnecessary snapshot queued.
+	 * fine, they will just get an unnecessary snapshot and invalidations
+	 * queued.
 	 */
 	dlist_foreach(txn_i, &builder->reorder->toplevel_by_lsn)
 	{
@@ -888,6 +889,14 @@ SnapBuildDistributeNewCatalogSnapshot(SnapBuild *builder, XLogRecPtr lsn)
 		 * transaction which in turn implies we don't yet need a snapshot at
 		 * all. We'll add a snapshot when the first change gets queued.
 		 *
+		 * Similarly, we don't need to add invalidations to a transaction whose
+		 * base snapshot is not yet set. Once a base snapshot is built, it will
+		 * include the xids of committed transactions that have modified the
+		 * catalog, thus reflecting the new catalog contents. The existing
+		 * catalog cache will have already been invalidated after processing
+		 * the invalidations in the transaction that modified catalogs,
+		 * ensuring that a fresh cache is constructed during decoding.
+		 *
 		 * NB: This works correctly even for subtransactions because
 		 * ReorderBufferAssignChild() takes care to transfer the base snapshot
 		 * to the top-level transaction, and while iterating the changequeue
@@ -897,13 +906,13 @@ SnapBuildDistributeNewCatalogSnapshot(SnapBuild *builder, XLogRecPtr lsn)
 			continue;
 
 		/*
-		 * We don't need to add snapshot to prepared transactions as they
-		 * should not see the new catalog contents.
+		 * We don't need to add snapshot or invalidations to prepared
+		 * transactions as they should not see the new catalog contents.
 		 */
 		if (rbtxn_prepared(txn) || rbtxn_skip_prepared(txn))
 			continue;
 
-		elog(DEBUG2, "adding a new snapshot to %u at %X/%X",
+		elog(DEBUG2, "adding a new snapshot and invalidations to %u at %X/%X",
 			 txn->xid, LSN_FORMAT_ARGS(lsn));
 
 		/*
@@ -913,6 +922,33 @@ SnapBuildDistributeNewCatalogSnapshot(SnapBuild *builder, XLogRecPtr lsn)
 		SnapBuildSnapIncRefcount(builder->snapshot);
 		ReorderBufferAddSnapshot(builder->reorder, txn->xid, lsn,
 								 builder->snapshot);
+
+		/*
+		 * Add invalidation messages to the reorder buffer of in-progress
+		 * transactions except the current committed transaction, for which we
+		 * will execute invalidations at the end.
+		 *
+		 * It is required, otherwise, we will end up using the stale catcache
+		 * contents built by the current transaction even after its decoding,
+		 * which should have been invalidated due to concurrent catalog
+		 * changing transaction.
+		 */
+		if (txn->xid != xid)
+		{
+			uint32 ninvalidations;
+			SharedInvalidationMessage *msgs = NULL;
+
+			ninvalidations = ReorderBufferGetInvalidations(builder->reorder,
+														   xid, &msgs);
+
+			if (ninvalidations > 0)
+			{
+				Assert(msgs != NULL);
+
+				ReorderBufferAddInvalidations(builder->reorder, txn->xid, lsn,
+											  ninvalidations, msgs);
+			}
+		}
 	}
 }
 
@@ -1184,8 +1220,11 @@ SnapBuildCommitTxn(SnapBuild *builder, XLogRecPtr lsn, TransactionId xid,
 		/* refcount of the snapshot builder for the new snapshot */
 		SnapBuildSnapIncRefcount(builder->snapshot);
 
-		/* add a new catalog snapshot to all currently running transactions */
-		SnapBuildDistributeNewCatalogSnapshot(builder, lsn);
+		/*
+		 * Add a new catalog snapshot and invalidations messages to all
+		 * currently running transactions.
+		 */
+		SnapBuildDistributeSnapshotAndInval(builder, lsn, xid);
 	}
 }
 
diff --git a/src/include/replication/reorderbuffer.h b/src/include/replication/reorderbuffer.h
@@ -729,6 +729,10 @@ extern TransactionId *ReorderBufferGetCatalogChangesXacts(ReorderBuffer *rb);
 
 extern void ReorderBufferSetRestartPoint(ReorderBuffer *rb, XLogRecPtr ptr);
 
+extern uint32 ReorderBufferGetInvalidations(ReorderBuffer *rb,
+											TransactionId xid,
+											SharedInvalidationMessage **msgs);
+
 extern void StartupReorderBuffer(void);
 
 #endif
