Join GitHub today
GitHub is home to over 31 million developers working together to host and review code, manage projects, and build software together.Sign up
kubelet: set low oom_score_adj for containers in critical pods #73758
@dashpole this is really all I was talking about in sig-node.
Currently, the only way to get the lowest
We have two reserved critical pod priority classes which users can explicitly convey:
We should treat this expression as the primary indicator of priority and, therefore, oom_adjust_score.
So just to walk through an example:
That is a pretty huge difference.
Given that we have priority, I think we should revisit the decision to set the
Generally, this issue occurs because we differentiate between QoS tiers in a non-continuous way. As mentioned at sig-node, using the guaranteed QoS tier requires setting CPU and memory limits, which isn't always desirable. Just as eviction doesn't take limits into account, I don't think setting the
My ideal formula would be something like:
This way, critical pods are still differentiated by requests, and the only input variable from the pod is the pod's memory request.
In that case, what would you think about something that differentiates between critical pods that have lower/higher requests, rather than setting all critical pods to -998?
I definitely think this an improvement. Just seeing what else we can accomplish while we have people's attention :).
[APPROVALNOTIFIER] This PR is APPROVED
The full list of commands accepted by this bot can be found here.
The pull request process is described here
Needs approval from an approver in each of these files:
Approvers can indicate their approval by writing
referenced this pull request
Feb 6, 2019
1 similar comment
Feb 7, 2019
15 of 16 checks passed
@sjenning: The following test failed, say
Thanks @sjenning it makes our configurations so much easier and we already had smaller outages because we forgot in one pod to set cpu limits....