Re: [PERFORM] Connections "Startup" - Mailing list pgsql-admin

From Pavel Stehule
Subject Re: [PERFORM] Connections "Startup"
Date
Msg-id CAFj8pRA-vcD_J+QnxebifT3BaugG173gJdpgTtGXb14Lcz-taQ@mail.gmail.com
Whole thread Raw
In response to Connections "Startup"  (Artem Tomyuk <admin@leboutique.com>)
Responses Re: [PERFORM] Connections "Startup"  (Jim Nasby <Jim.Nasby@BlueTreble.com>)
List pgsql-admin
Hi

2015-12-22 8:59 GMT+01:00 Artem Tomyuk <admin@leboutique.com>:
Hi.

I've noticed huge decrease in performance.
During this in htop i see a lot (200 - 300) of connections in state "startup", each of them eats 3-3% of CPU time. This processes are not visible in pg_stat_activity so i cant understand what they are doing, and i cant kill them. I cant see the bottleneck in Disk IO to. The logs of postgres says nothing to. I am confused.....
What can be the cause of  huge amount of "startup" connections.... 
Maybe its better to start use connection pooler such as pgbouncer? 
Thanks a lot.

What is your max_connections? Can you ran "perf top" ? What is there.

Too high number can enforce system overloading. You cannot to see these connections in pg_stat_activity because the process in this state isn't fully initialized.

There was lot of bugfix releases after 9.1.2 - currently there is PostgreSQL 9.2.19. Try to upgrade first.

Regards

Pavel
 

PS.
Server config is:
2 * Intel Xeon 2660 CPU with 64 gigs of RAM. 
Hardware RAID10.
Centos 6.6, PostgreSQL 9.1.2 




pgsql-admin by date:

Previous
From: Artem Tomyuk
Date:
Subject: Connections "Startup"
Next
From: CN
Date:
Subject: Re: feature request: pg_restore renaming target schema