Horton

Re: wake up logical workers after ALTER SUBSCRIPTION

Nathan Bossart <nathandbossart@gmail.com>

View thread View on postgr.es
From: Nathan Bossart <nathandbossart@gmail.com>
To: Melih Mutlu <m.melihmutlu@gmail.com>
Cc: Thomas Munro <thomas.munro@gmail.com>, "Hayato Kuroda (Fujitsu)" <kuroda.hayato@fujitsu.com>, "pgsql-hackers@postgresql.org" <pgsql-hackers@postgresql.org>
Date: 2022-12-06T21:29:54Z
Lists: pgsql-hackers

Attachments

On Tue, Dec 06, 2022 at 11:25:51AM -0800, Nathan Bossart wrote:
> On Tue, Dec 06, 2022 at 07:44:46PM +0300, Melih Mutlu wrote:
>> - When the state is SYNCDONE and the apply worker has to wake up to change
>> the state to READY.
>> 
>> I think we already call logicalrep_worker_wakeup_ptr wherever it's needed
>> for the above cases? What am I missing here?
> 
> IIUC we must restart all the apply workers for a subscription to enable
> two_phase mode.  It looks like finish_sync_worker() only wakes up its own
> apply worker.  I moved this logic to where the sync worker marks the state
> as SYNCDONE and added a check that two_phase mode is pending.  Even so,
> there can still be unnecessary wakeups, but this adjustment should limit
> them.

Actually, that's not quite right.  The sync worker will wake up the apply
worker to change the state from SYNCDONE to READY.  AllTablesyncsReady()
checks that all tables are READY, so we need to wake up all the workers
when an apply worker changes the state to READY.  Each worker will then
evaluate whether to restart for two_phase mode.

-- 
Nathan Bossart
Amazon Web Services: https://aws.amazon.com

Commits

  1. Avoid type cheats for invalid dsa_handles and dshash_table_handles.

  2. Track logrep apply workers' last start times to avoid useless waits.

  3. Check for two_phase change at end of process_syncing_tables_for_apply.

  4. Wake up a subscription's replication worker processes after DDL.