Re: Hungry postmaster - Mailing list pgsql-general

From Tom Lane
Subject Re: Hungry postmaster
Date
Msg-id 14550.1118925960@sss.pgh.pa.us
Whole thread Raw
In response to Hungry postmaster  ("Ilja Golshtein" <ilejn@yandex.ru>)
Responses Re: Hungry postmaster
List pgsql-general
"Ilja Golshtein" <ilejn@yandex.ru> writes:
> At the moment on my Linux box I have process
> 'postmaster' eats all CPU. It corresponds
> with connection closed day before -
> not sure if it was normal disconnect.
> strace shows the process constantly
> makes 'send' syscall with EPIPE result.
> Any ideas how to trace this issue
> and/or prevent such thing in the future?

Send it a SIGINT and see if it goes away.

If so, I would bet that someone did an unconstrained join (ie SELECT
the cross product of some large tables) and killed their client instead
of waiting for the result.  It's not hard to write a SQL query that will
produce terabytes of output :-(

If SIGINT doesn't kill it pretty quickly, try attaching to it with gdb
so you can get a stack trace.

            regards, tom lane

pgsql-general by date:

Previous
From: Tom Lane
Date:
Subject: Re: [HACKERS] INHERITS and planning
Next
From: "Ilja Golshtein"
Date:
Subject: Re: Hungry postmaster