Thanks for the hin. My bad. The backup db and 9.5 had a different type on
one of the foreign-key constrains char(36) vs varchar(36).
The schema was screwed couple of days ago, byt performance numbers I checked only
after migration to 9.5.
Sorry for the noise.
Tigran.
----- Original Message -----
> From: "Tom Lane" <tgl@sss.pgh.pa.us>
> To: "Andres Freund" <andres@anarazel.de>
> Cc: "Mkrtchyan, Tigran" <tigran.mkrtchyan@desy.de>, "pgsql-performance" <pgsql-performance@postgresql.org>
> Sent: Sunday, July 5, 2015 4:33:25 PM
> Subject: Re: [PERFORM] 9.5alpha1 vs 9.4
> Andres Freund <andres@anarazel.de> writes:
>> On 2015-07-05 13:10:51 +0200, Mkrtchyan, Tigran wrote:
>>> today I have update my test system to 9.5alpha1.
>>> Most of the operations are ok, except delete.
>>> I get ~1000 times slower!
>
>>> 255.88 | 566.11 | 452 | DELETE FROM t_inodes WHERE ipnfsid=$1 AND
>>> inlink = ?
>
>> That certainly should not be the case. Could you show the query plan for
>> this statement in both versions?
>
> EXPLAIN ANALYZE, please. I'm wondering about a missing index on some
> foreign-key-involved column. That would show up as excessive time in
> the relevant trigger ...
>
> regards, tom lane