Ok

df -h /dev/shm
Filesystem      Size  Used Avail Use% Mounted on
tmpfs           406G     0  406G   0% /run/shm

Ok I will try lowering it.

From: Pavel Stehule [mailto:pavel.steh...@gmail.com]
Sent: Wednesday, October 21, 2015 11:24 AM
To: Jamie Koceniak
Cc: pgsql-performance@postgresql.org
Subject: Re: [PERFORM] Recursive query performance issue



2015-10-21 19:55 GMT+02:00 Jamie Koceniak 
<jkocen...@mediamath.com<mailto:jkocen...@mediamath.com>>:
Hi Pavel,

Thanks for the reply.

1. The queries aren’t waiting on any locks.
The query has a recursive join that uses a table with only 80k records and that 
table is not updated often.

2. The I/O load was not high. CPU utilization was very high and load was very 
high.
We have a large effective_cache_size = 512GB (25% of total memory)

so your server has 2TB RAM? It is not usual server - so this issue can be 
pretty strange :(
What is size of shared memory? Probably is significantly lower than 
effective_cache_size? Try to reduce effective cache size to be lower than 
shared buffers
Regards
Pavel



Thanks,
Jamie

From: Pavel Stehule 
[mailto:pavel.steh...@gmail.com<mailto:pavel.steh...@gmail.com>]
Sent: Wednesday, October 21, 2015 12:04 AM
To: Jamie Koceniak
Cc: pgsql-performance@postgresql.org<mailto:pgsql-performance@postgresql.org>
Subject: Re: [PERFORM] Recursive query performance issue

Hi

2015-10-20 19:34 GMT+02:00 Jamie Koceniak 
<jkocen...@mediamath.com<mailto:jkocen...@mediamath.com>>:
Version:
-----------------------------------------------------------------------------------------------
PostgreSQL 9.1.14 on x86_64-unknown-linux-gnu, compiled by gcc (Debian 4.7.2-5) 
4.7.2, 64-bit

Query Plan
http://explain.depesz.com/s/4s37

Normally, this query takes around 200-300 ms to execute.
However when several queries are run concurrently, query performance drops to 
30-60 seconds.


there can be few reasons:
1. locking - are you sure, so your queries don't wait on locks?
2. issues with cache stability - is there high IO load? You can try to increase 
effective_cache_size (or decrease if you have not enough memory)
Regards
Pavel




Reply via email to