Fix PD failover #2570
Fix PD failover #2570
Conversation
pkg/manager/member/pd_failover.go
Outdated
| for podName, pdMember := range tc.Status.PD.Members { | ||
| if !podNames.Has(podName) { |
There was a problem hiding this comment.
I think we should update tryToMarkAPeerAsFailure instead of here, why ignore the non-managed members in the cluster health check?
There was a problem hiding this comment.
It's ok for me. updated.
pkg/manager/member/pd_failover.go
Outdated
| for podName, pdMember := range tc.Status.PD.Members { | ||
| if !podNames.Has(podName) { |
There was a problem hiding this comment.
Should we keep consistent with tikv handling here?
| @@ -125,6 +125,12 @@ func (pf *pdFailover) tryToMarkAPeerAsFailure(tc *v1alpha1.TidbCluster) error { | |||
| if pdMember.LastTransitionTime.IsZero() { | |||
| continue | |||
| } | |||
| if !pf.isPodDesired(tc, podName) { | |||
There was a problem hiding this comment.
Should we also delete the non desired members in the failure members as done in https://github.com/pingcap/tidb-operator/pull/2560/files#diff-cff8f7143431f3d6302182e300a81909R58?
There was a problem hiding this comment.
I'm not sure if it's necessary right now since all failure members will be cleared. anyway, we can revise it later.
Co-authored-by: Yecheng Fu <[email protected]>
|
/merge |
|
Your auto merge job has been accepted, waiting for:
|
|
/run-all-tests |
|
/run-cherry-picker |
Signed-off-by: sre-bot <[email protected]>
|
cherry pick to release-1.1 in PR #2577 |
What problem does this PR solve?
Limit the PD Member in checking PD failover. If the PD Member didn't manage by the operator, the operator wouldn't failover if the unmanaged pd member failed.
Does this PR introduce a user-facing change?: