Horton

Re: pgsql: Delay commit status checks until freezing executes.

Andres Freund <andres@anarazel.de>

View thread View on postgr.es
From: Andres Freund <andres@anarazel.de>
To: Peter Geoghegan <pg@bowt.ie>
Cc: pgsql-hackers@postgresql.org
Date: 2023-01-08T03:25:49Z
Lists: pgsql-hackers 
Hi,

On 2023-01-07 15:41:29 -0800, Peter Geoghegan wrote:
> Do we need to do anything about this to the "pg_xact and pg_subtrans"
> section of the transam README?

Probably a good idea, although it doesn't neatly fit right now.


> Also, does amcheck's get_xid_status() need a reference to these rules?

Don't think so? Whad made you ask?


> FWIW, I found an existing comment about this rule in the call to
> TransactionIdAbortTree() from RecordTransactionAbort() -- which took
> me quite a while to find. So you might have been remembering that
> comment before.

Possible, my memory is vague enough that it's hard to be sure...

Greetings,

Andres Freund

Commits

  1. Improve TransactionIdDidAbort() documentation.

  2. Check that xmax didn't commit in freeze check.

  3. Delay commit status checks until freezing executes.

  4. Add page-level freezing to VACUUM.

  5. Set relfrozenxid to oldest extant XID seen by VACUUM.