Re: Multixacts wraparound monitoring

Hello,
If I get you right:
Latest checkpoint's NextMultiXactId: 2075246000
Latest checkpoint's oldestMultiXid: 2019511697
Number of members files: 10820
Size pg_multixact/members/ (bytes) (2.7Gb): 2887696384
Pages in file: 32
Members on page: 2045
Number of members (32*2045*10820): 708060800
Members per multixact (2075246000 - 2019511697)/708060800: 12,70421916
Multixact size (bytes) (2887696384/708060800): 4,078316981 - It's a lot?

Kind regards,
 
Vladimir Pavlov

-----Original Message-----
From: Alvaro Herrera [mailto:alvherre(at)2ndquadrant(dot)com]
Sent: Thursday, March 31, 2016 12:17 AM
To: Pavlov Vladimir
Cc: 'Adrian Klaver'; 'pgsql-general(at)postgresql(dot)org'
Subject: Re: [GENERAL] Multixacts wraparound monitoring

Pavlov, Vladimir wrote:
> Yes, VACUUM helps to solve the problem and the WARNING gone away.
> But, the problem is that the VACUUM for the entire database (2.4T) takes over 7 hours, and it has to run every 15-20 hours (about 300 millions transactions), otherwise:
> ERROR: multixact "members" limit exceeded - and server stops working.
> The question is how to start the VACUUM at least once in three days.

You should have *started* the thread with this information.

My bet is that your multixacts are overly large and that's causing excessive vacuuming work; this is likely due to bug #8470 (which is fixed in 9.5 and master but not 9.3 and 9.4) and my bet is that you would very much benefit from the patch I posted in https://www.postgresql.org/message-id/20150620043650.GX133018%40postgresql.org
I didn't actually verify this; you could with some arithmetic on the deltas in multixact counters in pg_controldata output that you could take periodically.

--
Álvaro Herrera http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services