On 2013-09-10 08:45:33 -0400, Andrew Dunstan wrote:
>
> On 09/10/2013 08:20 AM, Andres Freund wrote:
>
> >A backtrace for this would be useful. Alternatively you could recompile
> >postgres using -fno-omit-frame-pointer in CFLAGS and use perf record -g.
>
> It's using a custom build, so this should be doable.
Great.
> >Any chance you have older prepared xacts, older sessions or something
> >like that around? I'd expect heap_prune* to be present in workloads that
> >spend significant time in heap_hot_search_buffer...
>
>
> Not sure about prepared transactions. There are certainly probably old
> prepared statements around, and long running transactions alongside this
> one.
Ok, long running transactions will do the trick. I quicky checked and
doing an index lookup for min/max histogram lookups was added *after*
8.4 which would explain why you're not seing the issue there
(c.f. 40608e7f949fb7e4025c0ddd5be01939adc79eec).
It getting slower and slower during a testrun would be explained by the
additional tuple versions amassing which cannot be marked dead because
of older transactions around. I guess those are also part of the test?
If I interpret things correctly you're using serializable? I guess there
is no chance to use repeatable read instead?
Greetings,
Andres Freund
--
Andres Freund http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services