-
Task
-
Resolution: Obsolete
-
Major
-
None
-
None
-
None
-
None
-
5
-
False
-
False
-
Undefined
-
-
SDN Sprint 203, SDN Sprint 204
-
0
-
0.0
Implement a pre-puller logic (https://coreos.slack.com/archives/CDCP2LA9L/p1623683204281100?thread_ts=1623666860.270300&cid=CDCP2LA9L) for reducing downtime during ovn-k8s upgades.
The CNI 2.0 fix for a prober that can identify when a node is ready or not which was the preferred way to solve https://bugzilla.redhat.com/show_bug.cgi?id=1943334 and https://bugzilla.redhat.com/show_bug.cgi?id=1943336 might not happen anytime soon or might be a bit tricky.
So the way we see now, we could do a combo of things:
1) Do the pre-puller and reduce the downtime from 60sec further.
2) Do the tainting as well, - even if its not efficient and doesn't cover all use cases like node restarts, at least from CI perspective it will solve the bugs.
3) We can utilize the `Status()` probes of the CRIO/CNI and leverage when the plugin deletes the config versus sets it up. Problem here is to make multus reflect the changes done by SDN/OVN plugins.