Thread: High Disk Utilization on Master with Streaming Replication Running 9.2.4

Hello,
What is the disk overhead for streaming replication? I have noticed that PostgreSQL’s disk utilization goes up by 3x on the master when a slave is connected. The setup only uses streaming replication, i.e. archive is off, there is no archive command. I see with iotop that the wal writer process is averaging 80% of IO. If I stop the slave, then IO utilization on the master drops to ~20%. Is there something that I am doing wrong? I have wal_keep_segments at 2048, checkpoint_segments at 96. 
Thank you,
Joel

Re: High Disk Utilization on Master with Streaming Replication Running 9.2.4

From
John R Pierce
Date:
On 7/18/2014 4:44 PM, Joel Avni wrote:
> Hello,
> What is the disk overhead for streaming replication? I have noticed
> that PostgreSQL’s disk utilization goes up by 3x on the master when a
> slave is connected. The setup only uses streaming replication, i.e.
> archive is off, there is no archive command. I see with iotop that the
> wal writer process is averaging 80% of IO. If I stop the slave, then
> IO utilization on the master drops to ~20%. Is there something that I
> am doing wrong? I have wal_keep_segments at 2048, checkpoint_segments
> at 96.
> Thank you,
> Joel

there's bug fixes in 9.2 since .4 that could easily be causing what
you're seeing.

for instance,
http://www.postgresql.org/docs/current/static/release-9-2-5.html
(checkpoint memory leak when wal_level=hot_standby)

9.2.8 is out, 9.2.9 is coming out really soon.

--
john r pierce                                      37N 122W
somewhere on the middle of the left coast