I'm setting up OpenShift Enterprise 2 and I'd like to create a district with a larger gear size. Changing
/etc/openshift/resource_limits.conf
on the nodes is straightforward for increasing memory and disk available to the gear, but CPU resource management is less intuitive (from resource_limits.conf):
# cpu cpu_rt_period_us=100000 cpu_rt_runtime_us=950000
cpu_shares=128
cpu_cfs_quota_us=100000
By default, a gear can only consume a maximum of 100% of a single processor core. If I want to allow a bigger gear size that could allow full utilization of 2 processor cores, how would I do that, or is it currently not possible at all in OpenShift?
Since all the gears are the same, and since 'cpu_shares' are compared on a relative basis when restricting a group, I'm not sure it makes sense to change 'cpu_shares'.
However, 'cpu_cfs_quota_us' looks like it might be the right knob to turn. From this page:
It appears that I should be able to double the quota to get a full 2 cores. However, it's not clear whether OpenShift will respect this, since the 'cpu_cfs_period_us' parameter is not even found in resource_limits.conf.
I performed an experiment using 'stress'. I first confirmed that I could load 2 cores under a normal ssh login (using 'stress --cpu 2'). Then I logged in to a gear on that host and ran the same thing. With cpu_cfs_quota_us=100000, I can only consume a max of 50% CPU for each stress process. But when I change to cpu_cfs_quota_us=200000, I can consume over 99% for each process, so it appears that it is now successful. Would be nice if this was called out in the OpenShift docs...