[ngw] GroupWise POA Memory Usage

Anthony Harper aharper at psc.ac.uk
Thu Mar 28 09:33:39 UTC 2019

Hi Marvin,

I also have POA crashes.  What usually happens is that either the poa crashes (with a mention of libc in /var/log/messages), the oom killer steps in or I manually restart the poa out of hours.  I've recently migrated GW from OES 2015 SP1 to SLES 15 (all running on top of ESX), and the memory usage issue and crashes have followed.

I use Adrem Netcrunch to monitor our servers, and I'm tracking "% Memory Used by Processes" for the server and "% Memory Utilization" of the poa process.  They track in near perfect correlation.

I am sending cores to support (and have done for a few years) - the only steps I have to create the core is editing the /etc/sysconfig/grpwise file's GROUPWISE_DEBUG_OPTION to "on,fence" - do you have other steps?

>>> "Marvin Huffaker" <mhuffaker at redjuju.com> 27/03/2019 18:14 >>>
I've been chasing POA crashes on larger systems for a couple years now.   I haven't actually observed the behavior you're mentioning (I've only see after the POA Crashes) but with the cores we've obtained, I'm told it's a memory corruption issue.   Where/how are you monitoring the memory consumption? I could monitor my customer systems for the same behavior on the problem systems.  One of my systems is SLES 15 but it doesn't seem to matter which OS, I've seen it on SLES11 and SLES 12 also.  Are you getting Cores to the support team for analysis?  If so that could help get it resolved.  While it may be common, it shouldn't be considered normal.   I have a list of steps to take to prepare your server for a core, it's more  of a challenge on SLES 12 and 15, but you actually get better cores than on SLES 11.

>>> "Anthony Harper" <aharper at psc.ac.uk> 3/27/2019 8:25 AM >>>

The poa processes on my two GroupWise servers gradually increase their memory consumption until the server's oom-killer steps in a kills the poa.  Is this normal behaviour?

We're currently running 18.1.0 -132861 on SLES 15, however I've seen this behaviour for a number of years now.



More information about the ngw mailing list