Conversation
| # only keep those that have been unschedulable for more than 5min over the past 30min | ||
| [30m:]) > 5 |
There was a problem hiding this comment.
Is it "firing only when there are more than 5" or "more than 5mn" ?
There was a problem hiding this comment.
5 minutes.
I replaced min with minutes in my comment to make it clearer.
| [30m:]) > 5 | ||
| # count per cluster | ||
| ) by (cluster_id, cluster_type, customer, installation, pipeline, provider, region) | ||
| > 2 |
There was a problem hiding this comment.
I wanted to start with a high enough value to not get paged for nothing, but probably "at least 2 pods consistently failing" is a high enough value.
I'll reduce it to "at least 2".
There was a problem hiding this comment.
Ok, maybe add a comment about this
| # Let's start sending these alerts to atlas, then we can swich to provider team when properly tuned. | ||
| team: atlas |
There was a problem hiding this comment.
I am not so sure about this I would rather hand this over to kaas, this is not an observability topic.
There was a problem hiding this comment.
That's the plan, but only once we're happy with how the alert behaves.
There was a problem hiding this comment.
Still I think it's up to kaas to adjust the alert and get familiar with it.
There was a problem hiding this comment.
All right, let them deal with this.
I've updated the code, but I'll need to warn them in Slack as well before merging then.
hervenicol
force-pushed
the
podsunschedulable
branch
2 times, most recently
from
e9a32ca to
314cb95
Compare
|
Shouldn't this already be covered by an alert about unsatisfied Deployments/DaemonSets? |
|
The alert you think about targets usually 1 deployment. This alert as for goal to detect cluster instabilities |
|
Are you sure? I'm pretty sure we have a generic one. But yeah, rather be safe than sorry. 🙂 |
hervenicol
force-pushed
the
podsunschedulable
branch
2 times, most recently
from
0e61e0d to
f333ef0
Compare
hervenicol
force-pushed
the
podsunschedulable
branch
from
f333ef0 to
92867b7
Compare
|
Ping @giantswarm/team-phoenix and @giantswarm/team-rocket This new alert would page you. |
4 participants
Towards: https://github.com/giantswarm/giantswarm/issues/33710
This PR alerts when multiple pods are unschedulable in the
kube-systemnamespace.This shows a general problem with the cluster, and in the future this should trigger an inhibition for alerts about individual broken components (like
LoggingAgentDown).Currently paging atlas even though this is kaas-related.
We'll probably discuss paging kaas when the alert is proven to work as intended.
Checklist
oncall-kaas-cloudGitHub group).