On 15/03/17 15:52, Joe Talerico wrote: > Can we start looking at CPU usage as well? Not sure if your data has > this as well... Usage by Heat specifically? Or just in general? We're limited by what is logged in the gate, so CPU usage by Heat is definitely a non-starter. Picking a random gate run, the Heat memory use comes from this file: http://logs.openstack.org/27/445627/2/check-tripleo/gate-tripleo-ci-centos-7-ovb-nonha/9232979/logs/ps.txt.gz which is generated by running `ps` at the end of the test. We also have this file (including historical data) from dstat: http://logs.openstack.org/27/445627/2/check-tripleo/gate-tripleo-ci-centos-7-ovb-nonha/9232979/logs/dstat.txt.gz so there is _some_ data there, it's mostly a question of how to process it down to something we can plot against time. My first guess would be to do a box-and-whisker-style plot showing the distribution of the 1m load average during the test. (CPU usage itself is generally a pretty bad measure of... CPU usage.) What problems are you hoping to catch? cheers, Zane.