Horton

Re: Synchronizing slots from primary to standby

shveta malik <shveta.malik@gmail.com>

View thread View on postgr.es
From: shveta malik <shveta.malik@gmail.com>
To: Amit Kapila <amit.kapila16@gmail.com>
Cc: Bharath Rupireddy <bharath.rupireddyforpostgres@gmail.com>, "Drouvot, Bertrand" <bertranddrouvot.pg@gmail.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>, shveta malik <shveta.malik@gmail.com>
Date: 2023-07-26T05:01:32Z
Lists: pgsql-hackers 
On Mon, Jul 24, 2023 at 9:00 AM Amit Kapila <amit.kapila16@gmail.com> wrote:
>
> On Mon, Jul 24, 2023 at 8:03 AM Bharath Rupireddy
> <bharath.rupireddyforpostgres@gmail.com> wrote:
> >
> > On Fri, Jul 21, 2023 at 5:16 PM shveta malik <shveta.malik@gmail.com> wrote:
> > >
> > > Thanks Bharat for letting us know. It is okay to split the patch, it
> > > may definitely help to understand the modules better but shall we take
> > > a step back and try to reevaluate the design first before moving to
> > > other tasks?
> >
> > Agree that design comes first. FWIW, I'm attaching the v9 patch set
> > that I have with me. It can't be a perfect patch set unless the design
> > is finalized.
> >
> > > I analyzed more on the issues stated in [1] for replacing LIST_SLOTS
> > > with SELECT query. On rethinking, it might not be a good idea to
> > > replace this cmd with SELECT in Launcher code-path
> >
> > I think there are open fundamental design aspects, before optimizing
> > LIST_SLOTS, see below. I'm sure we can come back to this later.
> >
> > > Secondly, I was thinking if the design proposed in the patch is the
> > > best one. No doubt, it is the most simplistic design and thus may
> > > .......... Any feedback is appreciated.
> >
> > Here are my thoughts about this feature:
> >
> > Current design:
> >
> > 1. On primary, never allow walsenders associated with logical
> > replication slots to go ahead of physical standbys that are candidates
> > for future primary after failover. This enables subscribers to connect
> > to new primary after failover.
> > 2. On all candidate standbys, periodically sync logical slots from
> > primary (creating the slots if necessary) with one slot sync worker
> > per logical slot.
> >
> > Important considerations:
> >
> > 1. Does this design guarantee the row versions required by subscribers
> > aren't removed on candidate standbys as raised here -
> > https://www.postgresql.org/message-id/20220218222319.yozkbhren7vkjbi5%40alap3.anarazel.de?
> >
> > It seems safe with logical decoding on standbys feature. Also, a
> > test-case from upthread is already in patch sets (in v9 too)
> > https://www.postgresql.org/message-id/CAAaqYe9FdKODa1a9n%3Dqj%2Bw3NiB9gkwvhRHhcJNginuYYRCnLrg%40mail.gmail.com.
> > However, we need to verify the use cases extensively.
> >
>
> Agreed.
>
> > 2. All candidate standbys will start one slot sync worker per logical
> > slot which might not be scalable.
> >
>
> Yeah, that doesn't sound like a good idea but IIRC, the proposed patch
> is using one worker per database (for all slots corresponding to a
> database).
>
> > Is having one (or a few more - not
> > necessarily one for each logical slot) worker for all logical slots
> > enough?
> >
>
> I guess for a large number of slots the is a possibility of a large
> gap in syncing the slots which probably means we need to retain
> corresponding WAL for a much longer time on the primary. If we can
> prove that the gap won't be large enough to matter then this would be
> probably worth considering otherwise, I think we should find a way to
> scale the number of workers to avoid the large gap.
>

How about this:

1) On standby, spawn 1 worker per database in the start (as it is
doing currently).

2) Maintain statistics on activity against each primary's database on
standby by any means. Could be by maintaining 'last_synced_time' and
'last_activity_seen time'.  The last_synced_time is updated every time
we sync/recheck slots for that particular database. The
'last_activity_seen_time' changes only if we get any slot on that
database where actually confirmed_flush or say restart_lsn has changed
from what was maintained already.

3) If at any moment, we find that 'last_synced_time' -
'last_activity_seen' goes beyond a threshold, that means that DB is
not active currently. Add it to list of inactive DB

4) Launcher on the other hand is always checking if it needs to spawn
any other extra worker for any new DB. It will additionally check if
number of inactive databases (maintained on standby) has gone higher
(> some threshold), then it brings down the workers for those and
starts a common worker which takes care of all such inactive databases
(or merge all in 1), while workers for active databases remain as such
(i.e. one per db). Each worker maintains the list of DBs which it is
responsible for.

5) If in the list of these inactive databases, we again find any
active database using the above logic, then the launcher will spawn a
separate worker for that.

Pros:
Lesser workers on standby as per the load on primary.
Lesser poking of primary by standby i.e. standby will send queries to
get slot info for all inactive DBs in 1 run instead of each worker
sending such queries separately.

Cons:  We might see spawning and freeing of workers more frequently.

Please let me know your thoughts on this.

thanks
Shveta

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).