Horton

Re: Minimal logical decoding on standbys

Amit Khandekar <amitdkhan.pg@gmail.com>

View thread
From: Amit Khandekar <amitdkhan.pg@gmail.com>
To: Robert Haas <robertmhaas@gmail.com>
Cc: Alvaro Herrera <alvherre@2ndquadrant.com>, Andres Freund <andres@anarazel.de>, tushar <tushar.ahuja@enterprisedb.com>, Petr Jelinek <petr.jelinek@2ndquadrant.com>, Craig Ringer <craig@2ndquadrant.com>, Petr Jelinek <petr@2ndquadrant.com>, pgsql-hackers <pgsql-hackers@postgresql.org>, Shaun Thomas <shaun.thomas@2ndquadrant.com>
Date: 2019-10-03T06:35:15Z
Lists: pgsql-hackers

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. Reduce the log level in 035_standby_logical_decoding.pl.

  2. 035_standby_logical_decoding: Add missing waits for replication

  3. For cascading replication, wake physical and logical walsenders separately

  4. Handle logical slot conflicts on standby

  5. Support invalidating replication slots due to horizon and wal_level

  6. Prevent use of invalidated logical slot in CreateDecodingContext()

  7. Replace replication slot's invalidated_at LSN with an enum

  8. Pass down table relation into more index relation functions

  9. Assert only valid flag bits are passed to visibilitymap_set()

  10. Remove unused _bt_delitems_delete() argument.

  11. Add xl_btree_delete optimization.

Attachments

On Mon, 30 Sep 2019 at 23:38, Robert Haas <robertmhaas@gmail.com> wrote:
>
> On Mon, Sep 30, 2019 at 7:35 AM Amit Khandekar <amitdkhan.pg@gmail.com> wrote:
> > Alright. Attached is the updated patch that splits the file into two
> > files, one that does only xmin related testing, and the other test
> > file that tests conflict recovery scenarios, and also one scenario
> > where drop-database drops the slots on the database on standby.
> > Removed get_slot_xmins() and get_node_from_slotname().
> > Renamed 'replica' to 'standby'.
> > Used node->backup() function instead of pg_basebackup command.
> > Renamed $master_slot to $master_slotname, similarly for $standby_slot.
>
> In general, I think this code is getting a lot clearer and easier to
> understand in these last few revisions.
>
> Why does create_logical_slot_on_standby include sleep(1)? Does the
> test fail if you take that out?
It has not failed for me, but I think sometimes it may happen that the
system command 'pg_recvlogical' is so slow to start that before it
tries to even create the slot, the subsequent checkpoint command
concurrently runs, causing a "running transactions" record to arrive
on standby *before* even pg_recvlogical decides the starting point
from which to receive records. So effectively pg_recvlogical can miss
this record.

> If so, it's probably going to fail on
> the buildfarm even with that included, because some of the buildfarm
> machines are really slow (e.g. because they use CLOBBER_CACHE_ALWAYS,
> or because they're running on a shared system with low hardware
> specifications and an ancient disk).

Yeah right, then it makes sense to explicitly wait for the slot to
calculate the restart_lsn, and only then run the checkpoint command.
Did that now.

>
> Similarly for the sleep(1) just after you VACUUM FREEZE all the databases.
I checked that VACUUM command returns only after updating the
pg_database.datfrozenxid. So now I think it's safe to immediately run
the checkpoint command after vacuum. So removed the sleep() now.

Attached is the updated patch series.

>
> I'm not sure wait the point of the wait_for_xmins() stuff is in
> 019_standby_logical_decoding_conflicts.pl. Isn't that just duplicating
> stuff we've already tested in 018?
Actually, in 019, the function call is more to wait for
hot_standby_feedback to take effect.


--
Thanks,
-Amit Khandekar
EnterpriseDB Corporation
The Postgres Database Company