Date: Fri, 10 Aug 2012 13:35:16 -0400 (EDT)
From: Joseph Marlin <jmarlin@saucontech.com>
To: pgsql-performance@postgresql.org
Message-ID: <385621813.89890122.1344620116156.JavaMail.root@email>
In-Reply-To: <533006730.89877629.1344618806676.JavaMail.root@email>
Subject: Increasing WAL usage followed by sudden drop
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit

http://i.imgur.com/sva4H.png

In the image above, please find the traffic we have seen from our main postgresql node to our cloud backup. 

A few notes on that image:

a) We're only interested in looking at the blue - outbound - traffic
b) In general, this pipe is almost exclusively for WAL usage only. 

Hopefully you can see how generalized WAL traffic increases, until it cuts off sharply, only to begin increasing again.


http://i.imgur.com/2V8XY.png

In that image, you can see the traffic just after a cutoff - slowly ramping up again. You can also see our mysterious sawtooth pattern - spikes of WAL traffic that occur on the hour, quarter-hour, half-hour, and three-quarter-hour marks. We don't see any corresponding spikes in database activity at those times, so we're also mystified as to why these spikes are occurring. 


Any ideas on any of this? Why the sawteeth? Why the rise-then-drop? 

Thanks so much!