Partitioning versus autovacuum

From: Greg Stark <stark(at)mit(dot)edu>
To: PostgreSQL Hackers <pgsql-hackers(at)lists(dot)postgresql(dot)org>
Subject: Partitioning versus autovacuum
Date: 2019-09-30 17:48:19
Message-ID: CAM-w4HMQKC8hw7nB9TW3OV+hkB5OUcPtvr_U_EiSOjByoa-e4Q@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

So we now support `ANALYZE partitioned_table` which will gather statistics
for the main table by gathering stats from all the partitions.

However as far as I can tell autovacuum will never actually trigger this
analyze. Because we never generate any update records for the parent table
in the statistics. Have I missed something?

I didn't find any discussion of this in the threads from when partitioning
was committed but there were a lot of discussions and I could easily have
missed it.

Is there a story for this? Some way to configure things so that autovacuum
will analyze partitioned tables?

Or should we look at doing something? Maybe whether we analyze a child we
should also update the parent -- and if there's no stats yet run analyze on
it?

This may be a serious enough problem for users that it may warrant
backpatching. Not having any stats is resulting in some pretty weird plans
for us.

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Alexander Korotkov 2019-09-30 17:59:49 Re: Connections hang indefinitely while taking a gin index's LWLock buffer_content lock(PG10.7)
Previous Message Fabrízio de Royes Mello 2019-09-30 17:46:42 Re: Hooks for session start and end, take two