Hi

We saw a rather extreme performance problem in a cluster upgraded from
9.1 to 9.3.  It uses a largish number of child tables (partitions) and
many columns.  Planning a simple UPDATE via the base table started
using a very large amount of memory and CPU time.

My colleague Rushabh Lathia tracked the performance change down to
http://git.postgresql.org/gitweb/?p=postgresql.git;a=commitdiff;h=c03ad5602f529787968fa3201b35c119bbc6d782
.

The call to copyObject in the loop introduced here seems to be
problematic (copying append_rel_list for every element in
append_rel_list unconditionally), though we're still trying to figure
it out.  Attached is a simple repro script, with variables to tweak.

Quite a few others have posted about this sort of thing and been
politely reminded of the 100 table caveat [1][2] which is fair enough,
but the situations seems to have got dramatically worse for some users
after an upgrade.

[1] 
http://www.postgresql.org/message-id/[email protected]
[2] 
http://www.postgresql.org/message-id/flat/[email protected]#[email protected]

-- 
Thomas Munro
http://www.enterprisedb.com

Attachment: repro-planner-explosion.sh
Description: Bourne shell script

-- 
Sent via pgsql-hackers mailing list ([email protected])
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers

Reply via email to