Movatterモバイル変換

[0]ホーム
URL:

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

 Sign up
Appearance settings

Commit0434033

Browse files
author
Amit Kapila
committed
Fix data loss in logical replication.
Data loss can happen when the DDLs like ALTER PUBLICATION ... ADD TABLE ...or ALTER TYPE ... that don't take a strong lock on table happensconcurrently to DMLs on the tables involved in the DDL. This happensbecause logical decoding doesn't distribute invalidations to concurrenttransactions and those transactions use stale cache data to decode thechanges. The problem becomes bigger because we keep using the stale cacheeven after those in-progress transactions are finished and skip thechanges required to be sent to the client.This commit fixes the issue by distributing invalidation messages fromcatalog-modifying transactions to all concurrent in-progress transactions.This allows the necessary rebuild of the catalog cache when decoding newchanges after concurrent DDL.We observed performance regression primarily during frequent execution of*publication DDL* statements that modify the published tables. Theregression is minor or nearly nonexistent for DDLs that do not affect thepublished tables or occur infrequently, making this a worthwhile cost toresolve a longstanding data loss issue.An alternative approach considered was to take a strong lock on eachaffected table during publication modification. However, this would onlyaddress issues related to publication DDLs (but not the ALTER TYPE ...)and require locking every relation in the database for publicationscreated 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 fixfor 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: 14Discussion:https://postgr.es/m/de52b282-1166-1180-45a2-8d8917ca74c6@enterprisedb.comDiscussion:https://postgr.es/m/CAD21AoAenVqiMjpN-PvGHL1N9DWnHSq673bfgr6phmBUzx=kLQ@mail.gmail.com
1 parent115f45e commit0434033

File tree

6 files changed

+134
-15
lines changed

6 files changed

+134
-15
lines changed

‎contrib/test_decoding/Makefile

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -8,7 +8,8 @@ REGRESS = ddl xact rewrite toast permissions decoding_in_xact \
88
spill slot truncate stream stats twophase twophase_stream
99
ISOLATION = mxact delayed_startup ondisk_startup concurrent_ddl_dml\
1010
oldest_xmin snapshot_transfer subxact_without_top concurrent_stream\
11-
twophase_snapshot catalog_change_snapshot skip_snapshot_restore
11+
twophase_snapshot catalog_change_snapshot skip_snapshot_restore\
12+
invalidation_distrubution
1213

1314
REGRESS_OPTS = --temp-config$(top_srcdir)/contrib/test_decoding/logical.conf
1415
ISOLATION_OPTS = --temp-config$(top_srcdir)/contrib/test_decoding/logical.conf
Lines changed: 20 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,20 @@
1+
Parsed test spec with 2 sessions
2+
3+
starting permutation: s1_insert_tbl1 s1_begin s1_insert_tbl1 s2_alter_pub_add_tbl s1_commit s1_insert_tbl1 s2_get_binary_changes
4+
step s1_insert_tbl1: INSERT INTO tbl1 (val1, val2) VALUES (1, 1);
5+
step s1_begin: BEGIN;
6+
step s1_insert_tbl1: INSERT INTO tbl1 (val1, val2) VALUES (1, 1);
7+
step s2_alter_pub_add_tbl: ALTER PUBLICATION pub ADD TABLE tbl1;
8+
step s1_commit: COMMIT;
9+
step s1_insert_tbl1: INSERT INTO tbl1 (val1, val2) VALUES (1, 1);
10+
step s2_get_binary_changes: SELECT count(data) FROM pg_logical_slot_get_binary_changes('isolation_slot', NULL, NULL, 'proto_version', '2', 'publication_names', 'pub') WHERE get_byte(data, 0) = 73;
11+
count
12+
-----
13+
1
14+
(1 row)
15+
16+
?column?
17+
--------
18+
stop
19+
(1 row)
20+
Lines changed: 32 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,32 @@
1+
# Test that catalog cache invalidation messages are distributed to ongoing
2+
# transactions, ensuring they can access the updated catalog content after
3+
# processing these messages.
4+
setup
5+
{
6+
SELECT'init'FROMpg_create_logical_replication_slot('isolation_slot','pgoutput');
7+
CREATETABLEtbl1(val1integer,val2integer);
8+
CREATEPUBLICATIONpub;
9+
}
10+
11+
teardown
12+
{
13+
DROPTABLEtbl1;
14+
DROPPUBLICATIONpub;
15+
SELECT'stop'FROMpg_drop_replication_slot('isolation_slot');
16+
}
17+
18+
session"s1"
19+
setup {SETsynchronous_commit=on; }
20+
21+
step"s1_begin" {BEGIN; }
22+
step"s1_insert_tbl1" {INSERTINTOtbl1 (val1,val2)VALUES (1,1); }
23+
step"s1_commit" {COMMIT; }
24+
25+
session"s2"
26+
setup {SETsynchronous_commit=on; }
27+
28+
step"s2_alter_pub_add_tbl" {ALTERPUBLICATIONpubADDTABLEtbl1; }
29+
step"s2_get_binary_changes" {SELECTcount(data)FROMpg_logical_slot_get_binary_changes('isolation_slot',NULL,NULL,'proto_version','2','publication_names','pub')WHEREget_byte(data,0)=73; }
30+
31+
# Expect to get one insert change. LOGICAL_REP_MSG_INSERT = 'I'
32+
permutation"s1_insert_tbl1""s1_begin""s1_insert_tbl1""s2_alter_pub_add_tbl""s1_commit""s1_insert_tbl1""s2_get_binary_changes"

‎src/backend/replication/logical/reorderbuffer.c

Lines changed: 23 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -5196,3 +5196,26 @@ ResolveCminCmaxDuringDecoding(HTAB *tuplecid_data,
51965196
*cmax=ent->cmax;
51975197
return true;
51985198
}
5199+
5200+
/*
5201+
* Count invalidation messages of specified transaction.
5202+
*
5203+
* Returns number of messages, and msgs is set to the pointer of the linked
5204+
* list for the messages.
5205+
*/
5206+
uint32
5207+
ReorderBufferGetInvalidations(ReorderBuffer*rb,TransactionIdxid,
5208+
SharedInvalidationMessage**msgs)
5209+
{
5210+
ReorderBufferTXN*txn;
5211+
5212+
txn=ReorderBufferTXNByXid(rb,xid, false,NULL,InvalidXLogRecPtr,
5213+
false);
5214+
5215+
if (txn==NULL)
5216+
return0;
5217+
5218+
*msgs=txn->invalidations;
5219+
5220+
returntxn->ninvalidations;
5221+
}

‎src/backend/replication/logical/snapbuild.c

Lines changed: 53 additions & 14 deletions
Original file line numberDiff line numberDiff line change
@@ -290,7 +290,7 @@ static void SnapBuildFreeSnapshot(Snapshot snap);
290290

291291
staticvoidSnapBuildSnapIncRefcount(Snapshotsnap);
292292

293-
staticvoidSnapBuildDistributeNewCatalogSnapshot(SnapBuild*builder,XLogRecPtrlsn);
293+
staticvoidSnapBuildDistributeSnapshotAndInval(SnapBuild*builder,XLogRecPtrlsn,TransactionIdxid);
294294

295295
/* xlog reading helper functions for SnapBuildProcessRunningXacts */
296296
staticboolSnapBuildFindSnapshot(SnapBuild*builder,XLogRecPtrlsn,xl_running_xacts*running);
@@ -843,23 +843,24 @@ SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
843843
}
844844

845845
/*
846-
* Add a new Snapshot to all transactions we're decoding that currently are
847-
* in-progress so they can see new catalog contents made by the transaction
848-
* that just committed. This is necessary because those in-progress
849-
* transactions will use the new catalog's contents from here on (at the very
850-
* least everything they do needs to be compatible with newer catalog
851-
* contents).
846+
* Add a new Snapshotand invalidation messagesto all transactions we're
847+
*decoding that currently arein-progress so they can see new catalog contents
848+
*made by the transactionthat just committed. This is necessary because those
849+
*in-progresstransactions will use the new catalog's contents from here on
850+
*(at the veryleast everything they do needs to be compatible with newer
851+
*catalogcontents).
852852
*/
853853
staticvoid
854-
SnapBuildDistributeNewCatalogSnapshot(SnapBuild*builder,XLogRecPtrlsn)
854+
SnapBuildDistributeSnapshotAndInval(SnapBuild*builder,XLogRecPtrlsn,TransactionIdxid)
855855
{
856856
dlist_itertxn_i;
857857
ReorderBufferTXN*txn;
858858

859859
/*
860860
* Iterate through all toplevel transactions. This can include
861861
* subtransactions which we just don't yet know to be that, but that's
862-
* fine, they will just get an unnecessary snapshot queued.
862+
* fine, they will just get an unnecessary snapshot and invalidations
863+
* queued.
863864
*/
864865
dlist_foreach(txn_i,&builder->reorder->toplevel_by_lsn)
865866
{
@@ -872,6 +873,14 @@ SnapBuildDistributeNewCatalogSnapshot(SnapBuild *builder, XLogRecPtr lsn)
872873
* transaction which in turn implies we don't yet need a snapshot at
873874
* all. We'll add a snapshot when the first change gets queued.
874875
*
876+
* Similarly, we don't need to add invalidations to a transaction whose
877+
* base snapshot is not yet set. Once a base snapshot is built, it will
878+
* include the xids of committed transactions that have modified the
879+
* catalog, thus reflecting the new catalog contents. The existing
880+
* catalog cache will have already been invalidated after processing
881+
* the invalidations in the transaction that modified catalogs,
882+
* ensuring that a fresh cache is constructed during decoding.
883+
*
875884
* NB: This works correctly even for subtransactions because
876885
* ReorderBufferAssignChild() takes care to transfer the base snapshot
877886
* to the top-level transaction, and while iterating the changequeue
@@ -881,13 +890,13 @@ SnapBuildDistributeNewCatalogSnapshot(SnapBuild *builder, XLogRecPtr lsn)
881890
continue;
882891

883892
/*
884-
* We don't need to add snapshotto prepared transactions as they
885-
* should not see the new catalog contents.
893+
* We don't need to add snapshotor invalidations to prepared
894+
*transactions as theyshould not see the new catalog contents.
886895
*/
887896
if (rbtxn_prepared(txn)||rbtxn_skip_prepared(txn))
888897
continue;
889898

890-
elog(DEBUG2,"adding a new snapshot to %u at %X/%X",
899+
elog(DEBUG2,"adding a new snapshotand invalidationsto %u at %X/%X",
891900
txn->xid,LSN_FORMAT_ARGS(lsn));
892901

893902
/*
@@ -897,6 +906,33 @@ SnapBuildDistributeNewCatalogSnapshot(SnapBuild *builder, XLogRecPtr lsn)
897906
SnapBuildSnapIncRefcount(builder->snapshot);
898907
ReorderBufferAddSnapshot(builder->reorder,txn->xid,lsn,
899908
builder->snapshot);
909+
910+
/*
911+
* Add invalidation messages to the reorder buffer of in-progress
912+
* transactions except the current committed transaction, for which we
913+
* will execute invalidations at the end.
914+
*
915+
* It is required, otherwise, we will end up using the stale catcache
916+
* contents built by the current transaction even after its decoding,
917+
* which should have been invalidated due to concurrent catalog
918+
* changing transaction.
919+
*/
920+
if (txn->xid!=xid)
921+
{
922+
uint32ninvalidations;
923+
SharedInvalidationMessage*msgs=NULL;
924+
925+
ninvalidations=ReorderBufferGetInvalidations(builder->reorder,
926+
xid,&msgs);
927+
928+
if (ninvalidations>0)
929+
{
930+
Assert(msgs!=NULL);
931+
932+
ReorderBufferAddInvalidations(builder->reorder,txn->xid,lsn,
933+
ninvalidations,msgs);
934+
}
935+
}
900936
}
901937
}
902938

@@ -1175,8 +1211,11 @@ SnapBuildCommitTxn(SnapBuild *builder, XLogRecPtr lsn, TransactionId xid,
11751211
/* refcount of the snapshot builder for the new snapshot */
11761212
SnapBuildSnapIncRefcount(builder->snapshot);
11771213

1178-
/* add a new catalog snapshot to all currently running transactions */
1179-
SnapBuildDistributeNewCatalogSnapshot(builder,lsn);
1214+
/*
1215+
* Add a new catalog snapshot and invalidations messages to all
1216+
* currently running transactions.
1217+
*/
1218+
SnapBuildDistributeSnapshotAndInval(builder,lsn,xid);
11801219
}
11811220
}
11821221

‎src/include/replication/reorderbuffer.h

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -676,6 +676,10 @@ TransactionId ReorderBufferGetOldestXmin(ReorderBuffer *rb);
676676

677677
voidReorderBufferSetRestartPoint(ReorderBuffer*,XLogRecPtrptr);
678678

679+
uint32ReorderBufferGetInvalidations(ReorderBuffer*rb,
680+
TransactionIdxid,
681+
SharedInvalidationMessage**msgs);
682+
679683
voidStartupReorderBuffer(void);
680684

681685
#endif

0 commit comments

Comments
 (0)
[8]ページ先頭
©2009-2025 Movatter.jp