Re: [Ganglia-general] Ganglia gmond memory leak?
i observed this in the past as well. running valgrind for days did not yeild any clue. i had a hunch that remote spoofed metrics were involved, as the leak seemed to get better when i had coincidentally disabled the sending of some of those spoof metrics. but, we never found anything conclusive. there was also some odd race such that sometimes after restart the leak was much faster, but after restarting a few times the leak slowed (but was always still fast enough to be a burden). -scott From: Aidan Wong aidanw...@attinteractive.com To: Ave-Lallemant, Nathan P nathan.p.ave-lallem...@efleets.com; ganglia-general ganglia-general@lists.sourceforge.net Sent: Thursday, February 23, 2012 8:34 AM Subject: Re: [Ganglia-general] Ganglia gmond memory leak? I've restarted the gmond process and memory usage drops until gmond hogs memory over time. ?Any Ganglia contributors who may want to chime in on this memory leak issue? ?I'm on Ganglia 3.2.0. ?Are there any improvements on version 3.3.1 addressing this issue? Thanks -- Virtualization Cloud Management Using Capacity Planning Cloud computing makes use of virtualization - but cloud computing also focuses on allowing computing to be delivered as a service. http://www.accelacomm.com/jaw/sfnl/114/51521223/ ___ Ganglia-general mailing list Ganglia-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/ganglia-general
Re: [Ganglia-general] Ganglia gmond memory leak?
makes sense, but i know in my case the number of metrics was constant after the server gmond had been started for about 10 minutes all gmetric crons had a chance to submit an initial value. -scott On Thu, 23 Feb 2012, Matt Massie wrote: Each unique metric (keyed on metric name) requires memory space in gmond. A good test is to peek at the number of metrics in gmond over time, e.g. $ telnet localhost 8649 | grep METRIC | wc -l If the number of metrics over time increases, so will the memory use. Ganglia will release the metric and memory when the age of the metric is greater than DMAX. A DMAX value of zero will cause ganglia to hold the metric indefinitely. In order to make sure that ganglia is releasing old metrics, set the DMAX value to something like 5 minutes (300 secs). For example, lets assume you are doing per process monitoring and the metric name looks like cpu_user.%d % (pid,) Over time, you'll have lots of metrics (cpu_user.343493, cpu_user.343022, cpu_user.232323) that start accumulating and taking up memory space. -Matt-- Virtualization Cloud Management Using Capacity Planning Cloud computing makes use of virtualization - but cloud computing also focuses on allowing computing to be delivered as a service. http://www.accelacomm.com/jaw/sfnl/114/51521223/___ Ganglia-general mailing list Ganglia-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/ganglia-general