/etc/cni/net.d/10-canal.conflist not updated when flannel subnet changes
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
Canal Charm |
Triaged
|
Medium
|
Unassigned |
Bug Description
We have seen that when the flannel network gets changed, as can be observed in either the flannel interface or in `/var/run/
This causes new pods created on the worker to be spun up with an ip from the older flannel subnet, and thus cutting off network connectivity for the pod.
Symptoms can be seen when running tcpdumps on the flannel interface, where you can find stale MAC being reported.
For example:
1) set up flannel with a subnet
2) Change the subnet for flannel
3) spin up pods, and you will see the old subnet being used from `/etc/cni/
I am not entirely sure how this flannel subnet changed on this host, but regardless I would expect `/etc/cni/
Env:
During our upgrade steps, we ended with an intermediate setup of,
-kubernetes-master:
charm: kubernetes-
channel: 1.24/stable
revision: 171
...
options:
channel: 1.23/stable
-kubernetes-worker:
charm: kubernetes-worker
channel: 1.24/stable
revision: 44
...
options:
channel: 1.23/stable
Changed in charm-canal: | |
status: | New → Triaged |
importance: | Undecided → Medium |
milestone: | none → 1.28 |
Changed in charm-canal: | |
milestone: | 1.28 → 1.28+ck1 |
Changed in charm-canal: | |
milestone: | 1.28+ck1 → 1.29 |
Changed in charm-canal: | |
milestone: | 1.29 → 1.29+ck1 |
Changed in charm-canal: | |
milestone: | 1.29+ck1 → 1.29+ck2 |
Changed in charm-canal: | |
milestone: | 1.29+ck2 → 1.30 |
Changed in charm-canal: | |
milestone: | 1.30 → 1.31 |