Closed Unix4ever closed 1 month ago
It won't work for such machines, so treat their etcd member ID as orphaned.
This fix mitigates the following issue:
A
Link
MachineSetEtcdAudit
B
Before the fix, after these operations the cluster would get stuck in scaling down state due to controllers deadlock:
MachineSetStatus
MachineStatus
/m
It won't work for such machines, so treat their etcd member ID as orphaned.
This fix mitigates the following issue:
A
is added and joined etcd, now the cluster has 4 members.A
is removed ungracefully by removing it'sLink
.MachineSetEtcdAudit
controller removes controlplaneA
member controlplaneB
is added and removed ungracefully by removing it'sLink
.Before the fix, after these operations the cluster would get stuck in scaling down state due to controllers deadlock:
MachineSetStatus
controller waits for the etcd memberA
to be removed.MachineSetEtcdAudit
controller tries to reach controlplaneB
which will never be available because it'sMachineStatus
no longer exist.