Follow up to irc on CREATE INDEX vs. maintenance_work_mem on 9.3

From: Alex Shulgin <ash(at)commandprompt(dot)com>
To: Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>
Cc: pgsql-hackers(at)postgresql(dot)org
Subject: Follow up to irc on CREATE INDEX vs. maintenance_work_mem on 9.3
Date: 2014-11-26 12:53:48
Message-ID: 87bnnuazs3.fsf@commandprompt.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers


Tom,

First of all, thanks for your help on IRC last time with that CREATE
INDEX memory consumption problem.

As has been pointed out in a stackexchange answer to my question[1], it
is indeed the limitation of pre-9.4 versions, but the limit is imposed
on memtuples array, rather than total memory the sort in CREATE INDEX
may allocate. The memtuples won't grow further than MaxAllocSize and
I've got 24x50x10^6 = 1200MB, which just doesn't fit.

We've got a customer who is testing a migration to PostgreSQL-9.3 (from
$some_other_db), thus they load the tables first (some of their tables
have 10-100 million rows), then create the indexes and they constantly
see disk sort being used despite lots of available RAM and
maintenance_work_mem set to increasingly higher values.

Now my question, is it feasible to back-patch this to 9.3? Or should we
tell the customer to wait before 9.4 is released?

Thanks.
--
Alex

[1] http://dba.stackexchange.com/questions/83600/postgresql-create-index-memory-requirement

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Robert Haas 2014-11-26 13:02:19 Re: superuser() shortcuts
Previous Message Stephen Frost 2014-11-26 12:37:38 Re: Role Attribute Bitmask Catalog Representation