Check status of all the core pods for microshift #4009
Conversation
|
[APPROVALNOTIFIER] This PR is NOT APPROVED This pull-request has been approved by: The full list of commands accepted by this bot can be found here.
Needs approval from an approver in each of these files:
Approvers can indicate their approval by writing |
praveenkumar
force-pushed
the
issue_3852
branch
from
92394d6 to
5907b3b
Compare
pkg/crc/cluster/cluster.go
Outdated
| } | ||
|
|
||
| func podRunningForNamespace(ocConfig oc.Config, namespace string) bool { | ||
| stdin, stderr, err := ocConfig.WithFailFast().RunOcCommand("get", "pods", "-n", namespace, "--field-selector=status.phase!=Running") |
There was a problem hiding this comment.
Why !=Running? I would have expected =Running?
There was a problem hiding this comment.
@cfergeau If I use ==Running that means it will show all the running pods, what we want the pods which are not in Running phase in that specific namespace so we can reiterate in retry function.
$ kubectl get pod -n kube-system --field-selector=status.phase!=Running
NAME READY STATUS RESTARTS AGE
csi-snapshot-controller-85cc4fd76b-xznzw 1/1 Pending 0 45h
| if !podRunningForNamespace(ocConfig, namespace) { | ||
| logging.Debugf("Pods in %s namespace are not running", namespace) | ||
| return &errors.RetriableError{Err: fmt.Errorf("pods in %s namespace are not running", namespace)} | ||
| } |
There was a problem hiding this comment.
Fwiw, this is a bit wasteful as we'll try again and again the same namespaces even if we already found running pods. Maybe this can be done with a map? map keys are namespaces, iterate over the keys. When there are running pods in the namespace, remove it from the map?
There was a problem hiding this comment.
Yes, but it has bit of benefit in case some pod goes to reconciliation state (like in one iteration it is running but in second it is in pending state.) It is not full proof solution (with k8s context it is never going to be) but should be good for initial feedback if core pods are running.
There was a problem hiding this comment.
In the OpenShift case, once the oc get co check succeeds once, we retry 2 more times and we only decide the cluster is good when the oc get co check succeeds 3 times in a row. If you want to handle " in one iteration it is running but in second it is in pending state" it would be nice to have a consistent approach.
There was a problem hiding this comment.
In the OpenShift case, once the
oc get cocheck succeeds once, we retry 2 more times and we only decide the cluster is good when theoc get cocheck succeeds 3 times in a row. If you want to handle " in one iteration it is running but in second it is in pending state" it would be nice to have a consistent approach.
yes in case of openshift we can iterate over all the clusteroperator at once because those are not namespace specific resource. Here we are not able to have a single call which provide use all the pods status in core namespaces otherwise I would've use same logic. So now we iterate over namespace by namespace and check the pods status.
There was a problem hiding this comment.
I'm not questioning the way the iterations are done, I was reacting to
it has bit of benefit in case some pod goes to reconciliation state (like in one iteration it is running but in second it is in pending state.)
For an OpenShift cluster, we roughly do iterate over a isClusterReady() function until it returns true. Once it returns true, we still run it 2 times in case the cluster was ready, but in a transient/conciliation state.
If reconciliation is something you want to try to handle better, I would use the same approach as for OpenShift for consistency, cluster is not ready before isClusterReady() succeeded 3 times in a row.
In past we observed having kube api access doesn't mean all the required service pods are running and cluster is working as expected. This PR adds a list of core namespace for microshift preset and make sure all the pods in that namespace is running before letting user to know to consume the cluster.
praveenkumar
force-pushed
the
issue_3852
branch
from
5907b3b to
217216c
Compare
| return errors.Retry(ctx, 2*time.Minute, waitForPods, 2*time.Second) | ||
| } | ||
|
|
||
| func podRunningForNamespace(ocConfig oc.Config, namespace string) bool { |
There was a problem hiding this comment.
allPodsRunning(ocConfig oc.Config, namespace string) bool or checkAllPodsRunning is more descriptive/accurate.
There was a problem hiding this comment.
it should be in namespace context so checkAllPodsRunningInNamespace or allPodsRunningForNamespace ?
There was a problem hiding this comment.
There is a namespace argument, we don't have non-namespace function this could be confused with, so I don't think it's really useful to mention Namespace in the function name. It's more something for an api doc comment if you think it's important to inform API users that it will only iterate over a single namespace.
| if !podRunningForNamespace(ocConfig, namespace) { | ||
| logging.Debugf("Pods in %s namespace are not running", namespace) | ||
| return &errors.RetriableError{Err: fmt.Errorf("pods in %s namespace are not running", namespace)} | ||
| } |
There was a problem hiding this comment.
In the OpenShift case, once the oc get co check succeeds once, we retry 2 more times and we only decide the cluster is good when the oc get co check succeeds 3 times in a row. If you want to handle " in one iteration it is running but in second it is in pending state" it would be nice to have a consistent approach.
praveenkumar
added
the
has-to-be-in-release
praveenkumar
removed
the
has-to-be-in-release
This is not microshift specific, as it might also be able to solve issues with the readiness of the OCP and OKD preset. |
|
What was the reasoning behind this label, as the fixed issue: #3852 is merely an enhancement. |
|
/hold |
In past we observed having kube api access doesn't mean all the required service pods are running and cluster is working as expected. This PR adds a list of core namespace for microshift preset and make sure all the pods in that namespace is running before letting user to know to consume the cluster.
Fixes: Issue #3852