Re: BUG #13818: PostgreSQL crashes after cronjob runs as "postgres"

From: Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>
To: ryan(at)runscope(dot)com
Cc: pgsql-bugs(at)postgresql(dot)org
Subject: Re: BUG #13818: PostgreSQL crashes after cronjob runs as "postgres"
Date: 2015-12-15 22:16:55
Message-ID: 4665.1450217815@sss.pgh.pa.us
Views: Raw Message | Whole Thread | Download mbox
Thread:
Lists: pgsql-bugs

ryan(at)runscope(dot)com writes:
> We are running PostgreSQL 9.4.5 on CentOS 7, uising the RPM from
> postgresql.org. We perform automatic OS updates, so our server was just
> upgraded from CentOS 7.1 to CentOS 7.2. Since the upgrade we've been seeing
> PostgreSQL crash after a cronjob is run as the "postgres" user.

> Here's the syslog output from /var/log/messages when the cronjob runs:
> ...
> Dec 14 18:24:01 prod118 postgres[52611]: [2584-1] PANIC:
> semop(id=57311261) failed: Invalid argument

That seems to mean that something has decided to remove Postgres' SysV
semaphores --- the only plausible reason for EINVAL from semop() is that
the semaphore set no longer exists. Now, Postgres itself certainly did
not do that.

I would bet that what you are looking at is a kernel or systemd bug
associated with this "slice" feature. Perhaps it's miscategorizing
Postgres' existing semaphores as belonging to the new slice and then
destroying them when it destroys the slice.

regards, tom lane

In response to

Browse pgsql-bugs by date

  From Date Subject
Next Message Joe Conway 2015-12-15 23:17:25 Re: json_to_recordset not working with camelcase json keys
Previous Message Daniel Kellenberger 2015-12-15 21:46:33 json_to_recordset not working with camelcase json keys