Re: [TRAP: FailedAssertion] causing server to crash

On 19 July 2017 at 20:26, Neha Sharma <neha(dot)sharma(at)enterprisedb(dot)com> wrote:

> Hi,
>
> I am getting FailedAssertion while executing the attached script.However,I
> am not able to produce the core dump for the same,the script runs in
> background and takes around a day time to produce the mentioned error.
>
> "TRAP: FailedAssertion("!(TransactionIdPrecedesOrEquals(oldestXact,
> ShmemVariableCache->oldestXid))", File: "clog.c", Line: 683)
> 2017-07-19 01:16:51.973 GMT [27873] LOG: server process (PID 28084) was
> terminated by signal 6: Aborted
> 2017-07-19 01:16:51.973 GMT [27873] DETAIL: Failed process was running:
> autovacuum: VACUUM pg_toast.pg_toast_13029 (to prevent wraparound)"
>

It's not much of a stretch to assume this was likely introduced
by ea42cc18c:

https://git.postgresql.org/gitweb/?p=postgresql.git;a=commit;h=ea42cc18c35381f639d45628d792e790ff39e271

so I'll have a look to see if I can spot how it could happen.

Please show your "SELECT version()" and if using a build from git, the "git
rev-parse --short HEAD" for the build you're running.

It's interesting and probably significant that your Pg was nearing
wraparound. Just a tip for that - if you want to create a situation where
you near wraparound quickly and easily, for testing, you can, on a test
instance you don't care about:

- select txid_current();
- stop pg
- pg_resetxlog -n $(( $the_xid + (1<<31) - 1001000 ))
- try to start pg, note the missing clog segment it complains about
- dd if=/dev/zero bs=1 count=262144 of=datadir/pg_clog/$MISSINGSEGNAME
- start Pg

That should put you about 1000 txn's from the 1 million xid limit, assuming
I got my maths right (don't assume that), and assuming your starting xid is
well short of the 2-million mark.

--
Craig Ringer http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services