Hi Andrew. We were seeing exactly the same thing. On a 32 core machine all of the cores would regularly sit around 75-80% cpu utilization.
The only thing that alleviated it was to change the operations mode to distributed. As soon as that happened the CPU utilization dropped to very little, the occasional spike being the extend of it. Alen. http://docs.hortonworks.com/HDPDocuments/Ambari-2.1.0.0/bk_ambari_reference_guide/content/_configuring_ambari_metrics_for_distributed_mode.html -----Original Message----- From: Andrew Robertson [mailto:[email protected]] Sent: September 16, 2015 10:35 PM To: [email protected] Subject: Ambari metrics collector CPU usage How much CPU is the ambari metrics collector expected to use (Ambari 2.1.1)? In my production cluster with 12 machines the ams hbase master process has averaged 82% of a cpu core (sandybridge), based on total cpu usage in seconds over the past 2 weeks. A test cluster with 4 machines has also been behaving the same way with very high cpu on the ams hbase master java process. I'm not sure if this is the new normal for metrics (versus the negligible overhead with gmon/gmetad) or if there's something wrong somewhere in either ams, hbase, or in the (default) config. Are others seeing high CPU here as well? Is there something I should look at to see what it may be doing, or are there suggestions for reducing the cpu usage? Or is this a known issue? I looked through the open issues and didn't see anything obvious that was tracking this. Thank you!
