[k8s-R5.0]: Sync between Control and data plane required to handle situations where CNI miss the messages
Bug #1765941 reported by
Pulkit Tandon
This bug affects 1 person
Affects | Status | Importance | Assigned to | Milestone | ||
---|---|---|---|---|---|---|
Juniper Openstack | Status tracked in Trunk | |||||
R5.0 |
Fix Committed
|
High
|
Venkatraman Venkatapathy | |||
Trunk |
Fix Committed
|
High
|
Venkatraman Venkatapathy |
Bug Description
Synch is required between Control and Data plane for K8s config.
e.g.:
If pod deletion is in progress and compute goes down, the CNI on agent won't get to know and would not free up the resources.
It can further lead to problems of new Pod creations.
Another case is when deployments are created.
Compute reboot will result in creating a replication of a different compute.
But when the node comes up after reboot, it will still have the resource as the CNI missed the notification.
To post a comment you must log in.
Hi Pulkit,
Do you have output of "kubectl get pod -o wide" when these conditions happen ? When state Pod is in ?
Thanks,
Sachin