Message-ID: <55931EB0.1000003@agliodbs.com>
Date: Tue, 30 Jun 2015 15:56:48 -0700
From: Josh Berkus <josh@agliodbs.com>
Organization: PostgreSQL Experts Inc.
User-Agent: Mozilla/5.0 (X11; Linux x86_64;
 rv:31.0) Gecko/20100101 Thunderbird/31.6.0
MIME-Version: 1.0
To: eudald_v <reaven.galaeindael@gmail.com>, pgsql-performance@postgresql.org
Subject: Re: Sudden connection and load average spikes with postgresql
 9.3
References: <1435675936590-5855895.post@n5.nabble.com>
 <WM!c7bd7562c7bece63187a13788e45741062e0d671e243b767856b62e6b994fa96ffbf6af1c5663cb27c147d6d67cd9bc5!@asav-3.01.com>
In-Reply-To: 
 <WM!c7bd7562c7bece63187a13788e45741062e0d671e243b767856b62e6b994fa96ffbf6af1c5663cb27c147d6d67cd9bc5!@asav-3.01.com>
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit
Precedence: bulk
Sender: pgsql-performance-owner@postgresql.org

On 06/30/2015 07:52 AM, eudald_v wrote:
> Two days from now, I've been experiencing that, randomly, the connections
> rise up till they reach max connections, and the load average of the server
> goes arround 300~400, making every command issued on the server take
> forever. When this happens, ram is relatively low (70Gb used), cores
> activity is lower than usual and sometimes swap happens (I've swappiness
> configured to 10%)

As Tom said, the most likely reason for this is application behavior and
blocking locks.  Try some of these queries on our scripts page:

https://github.com/pgexperts/pgx_scripts/tree/master/locks

However, I have seem some other things which cause these kinds of stalls:

* runaway connection generation by the application, due to either a
programming bug or an irresponsible web crawler (see
https://www.pgexperts.com/blog/quinn_weaver/)

* issues evicting blocks from shared_buffers: what is your
shared_buffers set to?  How large is your database?

* Checkpoint stalls: what FS are you on?  What are your transaction log
settings for PostgreSQL?

* Issues with the software/hardware stack around your storage, causing
total IO stalls periodically.  What does IO throughput look like
before/during/after the stalls?

The last was the cause the last time I dealt with a situation like
yours; it turned out the issue was bad RAID card firmware where the card
would lock up whenever the write-through buffer got too much pressure.

-- 
Josh Berkus
PostgreSQL Experts Inc.
http://pgexperts.com


-- 
Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-performance