BUG #5767: Memory Leak

On Thu, Nov 25, 2010 at 05:28, Darryl Pye <darrylpye@hotmail.com> wrote:

The following bug has been logged online:

Bug reference:      5767
Logged by:          Darryl Pye
Email address:      darrylpye@hotmail.com
PostgreSQL version: 9.01
Operating system:   Windows Server  2008
Description:        Memory Leak
Details:

A single postgres  process  appears to be constantly growing in memory usage
 until it reaches 100% of memory on the machine.

I have rebooted the server and the problem goes away but then  a single
postgres.exe process continues to  eat memory until it uses all available
ram then i have to restart again.

Machine has 4 gig of ram available.
I am able to watch the process grow constantly at approximately  8kb every
second.

Until it reaches  4gig.

I have made the following configuration changes from default
shared_buffers = 1GB
wal_buffers = 1MB
effective_cache_size = 2GB

What are your work_mem and maintenance_work_mem settings?

And which of the postgres processes is it, and what's it doing? You
can use Process Explorer to look at the process, and find a handle
called something starting with "pgident" - that one will contain a
description of the process, just like the title does on Unix. Or you
can look up the process id in the pg_stat_activity view to identify
it.

--
 Magnus Hagander
 Me: http://www.hagander.net/
 Work: http://www.redpill-linpro.com/

On 11/25/2010 12:28 PM, Darryl Pye wrote:

PostgreSQL version: 9.01

32-bit or 64-bit ?

A single postgres process appears to be constantly growing in memory usage
until it reaches 100% of memory on the machine.

How are you watching it? Task manager?

You need to use something that can differentiate between the shared
memory block and the per-task private memory. Process Explorer can do
that, IIRC.

I have rebooted the server and the problem goes away but then a single
postgres.exe process continues to eat memory until it uses all available
ram then i have to restart again.

Are you able to identify which postgres process it is? Is it shown in
"SELECT * from pg_stat_activity;" ?

Is it one of the postgres processes that appears when PostgreSQL is
first started, before any connections are made to the database server?

Are you sure you're not looking at totals for all postgres.exe instances?

I have made the following configuration changes from default
shared_buffers = 1GB
wal_buffers = 1MB
effective_cache_size = 2GB

With those settings, you should expect PostgreSQL to appear to use a bit
over 1GB of RAM per postgres.exe, though the vast majority of that 1GB
should be shared between all the postgres.exe processes so the total
memory of all postgres.exe instances will appear to vastly exceed your
machine's RAM. That's just because of the bad accounting for shared
memory in most OSes, including windows, and doesn't affect how it runs.

--
Craig Ringer