Horton

Re: First steps with 8.3 and autovacuum launcher

Tom Lane <tgl@sss.pgh.pa.us>

View thread View on postgr.es
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Simon Riggs <simon@2ndquadrant.com>
Cc: Alvaro Herrera <alvherre@commandprompt.com>, Michael Paesold <mpaesold@gmx.at>, Heikki Linnakangas <heikki@enterprisedb.com>, Gregory Stark <stark@enterprisedb.com>, Guillaume Smet <guillaume.smet@gmail.com>, "Matthew T. O'Connor" <matthew@zeut.net>, Stefan Kaltenbrunner <stefan@kaltenbrunner.cc>, PostgreSQL-development <pgsql-hackers@postgresql.org>
Date: 2007-10-12T17:51:42Z
Lists: pgsql-hackers 
Simon Riggs <simon@2ndquadrant.com> writes:
> Can you explain further what you meant by "don't disable manual
> cancels".

I meant that pg_cancel_backend() should still work on autovac workers,
contrary to Alvaro's suggestion that autovac workers should sometimes
ignore SIGINT.

Basically the implementation vision I have is that the SIGINT catcher in
an autovac worker should remain stupid, and any intelligence involved
should be on the side where we're deciding whether to send a signal or
not.  This probably does involve exposing more state in PGPROC but I see
nothing much wrong with that.  (It might be time to merge inVacuum,
isAutovacuum, and the additional state into a bitwise vacuumFlags field.)

			regards, tom lane