Re: [HACKERS] Aborted Transaction During Vacuum - Mailing list pgsql-hackers

From Tom Lane
Subject Re: [HACKERS] Aborted Transaction During Vacuum
Date
Msg-id 6178.934579907@sss.pgh.pa.us
Whole thread Raw
In response to Re: [HACKERS] Aborted Transaction During Vacuum  ("Oliver Elphick" <olly@lfix.co.uk>)
List pgsql-hackers
"Oliver Elphick" <olly@lfix.co.uk> writes:
>> But the vc_abort problem didn't cause a backend coredump --- it just
>> reported an error and failed to complete the vacuum, no?

> I got a coredump too; I never mentioned it, because I found the proximate
> cause and curing that made it go away.  When unlink failed in vc_shutdown,
> it called ELOG and a segfault occurred a little later.

Ah, I wish I'd known that.  So what Tony is seeing is exactly the same
behavior you observed.  OK, I feel better now --- I thought the coredump
was probably some platform-specific misbehavior that only Tony was seeing.

We still need to figure out what is causing it, because I can see no
reason for a coredump after vc_shutdown elog()s.  Something is being
clobbered that should not be.  But it sounds like installing the
vc_abort patch will get Tony on his feet, and then we can look for the
secondary bug at our leisure.
        regards, tom lane


pgsql-hackers by date:

Previous
From: Tom Lane
Date:
Subject: Re: [HACKERS] Aborted Transaction During Vacuum
Next
From: Tom Lane
Date:
Subject: Ah-hah, I see the problem: EndPortalAllocMode()