Horton

RE: Synchronizing slots from primary to standby

Zhijie Hou (Fujitsu) <houzj.fnst@fujitsu.com>

View thread View on postgr.es
From: "Zhijie Hou (Fujitsu)" <houzj.fnst@fujitsu.com>
To: Amit Kapila <amit.kapila16@gmail.com>, Masahiko Sawada <sawada.mshk@gmail.com>
Cc: Bharath Rupireddy <bharath.rupireddyforpostgres@gmail.com>, Bertrand Drouvot <bertranddrouvot.pg@gmail.com>, shveta malik <shveta.malik@gmail.com>, Ajin Cherian <itsajin@gmail.com>, Peter Smith <smithpb2250@gmail.com>, Dilip Kumar <dilipbalaut@gmail.com>, Nisha Moond <nisha.moond412@gmail.com>, "Hayato Kuroda (Fujitsu)" <kuroda.hayato@fujitsu.com>, Peter Eisentraut <peter.eisentraut@enterprisedb.com>, Bruce Momjian <bruce@momjian.us>, Ashutosh Sharma <ashu.coek88@gmail.com>, Andres Freund <andres@anarazel.de>, pgsql-hackers <pgsql-hackers@postgresql.org>, Alvaro Herrera <alvherre@alvh.no-ip.org>
Date: 2024-04-10T11:58:37Z
Lists: pgsql-hackers

Attachments

On Thursday, April 4, 2024 5:37 PM Amit Kapila <amit.kapila16@gmail.com> wrote:
> 
> BTW, while thinking on this one, I
> noticed that in the function LogicalConfirmReceivedLocation(), we first update
> the disk copy, see comment [1] and then in-memory whereas the same is not
> true in
> update_local_synced_slot() for the case when snapshot exists. Now, do we have
> the same risk here in case of standby? Because I think we will use these xmins
> while sending the feedback message (in XLogWalRcvSendHSFeedback()).
>
> * We have to write the changed xmin to disk *before* we change
> * the in-memory value, otherwise after a crash we wouldn't know
> * that some catalog tuples might have been removed already.

Yes, I think we have the risk on the standby, I can reproduce the case that if
the server crashes after updating the in-memory value and before saving them to
disk, the synced slot could be invalidated after restarting from crash, because
the necessary rows have been removed on the primary. The steps can be found in
[1].

I think we'd better fix the order in update_local_synced_slot() as well. I
tried to make the fix in 0002, 0001 is Shveta's patch to fix another issue in this thread. Since
they are touching the same function, so attach them together for review.

[1]
-- Primary:
SELECT 'init' FROM pg_create_logical_replication_slot('logicalslot', 'test_decoding', false, false, true);

-- Standby:
SELECT 'init' FROM pg_create_logical_replication_slot('standbylogicalslot', 'test_decoding', false, false, false);
SELECT pg_sync_replication_slots();

-- Primary:
CREATE TABLE test (a int);
INSERT INTO test VALUES(1);
DROP TABLE test;

SELECT txid_current();
SELECT txid_current();
SELECT txid_current();
SELECT pg_log_standby_snapshot();

SELECT pg_replication_slot_advance('logicalslot', pg_current_wal_lsn());

-- Standby:
- wait for standby to replay all the changes on the primary.

- this is to serialize snapshots.
SELECT pg_replication_slot_advance('standbylogicalslot', pg_last_wal_replay_lsn());

- Use gdb to stop at the place after calling ReplicationSlotsComputexx()
  functions and before calling ReplicationSlotSave().
SELECT pg_sync_replication_slots();

-- Primary:

- First, wait for the primary slot(the physical slot)'s catalog xmin to be
  updated to the same as the failover slot.

VACUUM FULL;

- Wait for VACUMM FULL to be replayed on standby.

-- Standby:

- For the process which is blocked by gdb, let the process crash (elog(PANIC,
  ...)).

After restarting the standby from crash, we can see the synced slot is invalidated.

LOG:  invalidating obsolete replication slot "logicalslot"
DETAIL:  The slot conflicted with xid horizon 741.
CONTEXT:  WAL redo at 0/3059B90 for Heap2/PRUNE_ON_ACCESS: snapshotConflictHorizon: 741, isCatalogRel: T, nplans: 0, nredirected: 0, ndead: 7, nunused: 0, dead: [22, 23, 24, 25, 26, 27, 28]; blkref #0: rel 1663/5/1249, blk 16


Best Regards,
Hou zj

Commits

  1. Doc: Add the new section "Logical Replication Failover".

  2. Fix the review comments and a bug in the slot sync code.

  3. Fix a test in failover slots regression test.

  4. Fix the intermittent buildfarm failures in 040_standby_failover_slots_sync.

  5. Split XLogCtl->LogwrtResult into separate struct members

  6. Ensure that the sync slots reach a consistent state after promotion without losing data.

  7. Introduce a new GUC 'standby_slot_names'.

  8. Fix BF failure introduced by commit b3f6b14cf4.

  9. Fixups for commit 93db6cbda0.

  10. Fix BF failure in commit 93db6cbda0.

  11. Add a new slot sync worker to synchronize logical slots.

  12. Improve ERROR/LOG messages added by commits ddd5f4f54a and 7a424ece48.

  13. Disable autovacuum on primary in 040_standby_failover_slots_sync test.

  14. Fix the incorrect format specifier used in commit 7a424ece48.

  15. Change the LOG level in 040_standby_failover_slots_sync.pl to DEBUG2.

  16. Add more LOG and DEBUG messages for slot synchronization.

  17. Another try to fix BF failure introduced in commit ddd5f4f54a.

  18. Fix BF introduced in commit ddd5f4f54a.

  19. Add a slot synchronization function.

  20. Clean-ups for 776621a5e4 and 7329240437.

  21. Enhance libpqrcv APIs to support slot synchronization.

  22. Add a failover option to subscriptions.

  23. Allow setting failover property in the replication command.

  24. Allow to enable failover property for replication slots via SQL API.

  25. Add macros for looping through a List without a ListCell.

  26. Track conflict_reason in pg_replication_slots.

  27. Allow upgrades to preserve the full subscription's state.

  28. Add support for incremental backup.

  29. Simplify some logic in CreateReplicationSlot()

  30. Fix uninitialized access to InitialRunningXacts during decoding after ERROR.

  31. Flexible options for CREATE_REPLICATION_SLOT.

  32. Empty search_path in logical replication apply worker and walsender.

  33. Fix oldest xmin and LSN computation across repslots after advancing

  34. Require C99 (and thus MSCV 2013 upwards).