Alvaro Herrera <alvherre(at)commandprompt(dot)com> writes:
> Hmmm, remember that DatabaseCancelAutovacuumActivity is called on CREATE
> DATABASE; but what it does is send SIGINT, not SIGTERM. Also, it's not
> in 8.2. SIGINT does terminate the autovac process however.
> I haven't read the whole problem report completely, so I'm not sure this
> has something to do or not.
AFAICT, SIGINT should be okay, because it will lead to an ERROR not a
FATAL elog; so control should fall out through the CATCH block before
the autovacuum process quits. The problem is with FATAL elogs.
Mark reports that the only FATAL lines in his logs are instances of
FATAL: terminating connection due to administrator command
FATAL: database "launchpad_ftest" does not exist
and the latter presumably isn't coming out from within the btree vacuum
code, so I don't see any other explanation for a FATAL exit than SIGTERM.
regards, tom lane
In response to
pgsql-hackers by date
|Next:||From: Tom Lane||Date: 2007-03-30 17:50:16|
|Subject: Re: Proposal: include PL/Proxy into core |
|Previous:||From: Pavan Deolasee||Date: 2007-03-30 17:21:25|
|Subject: Re: CREATE INDEX and HOT - revised design|