-
Notifications
You must be signed in to change notification settings - Fork 815
Open
Description
Summary
When rebooting or otherwise making changes to microk8s that require it to be stopped and restarted, I noticed that flannel periodically "leaks" IP addresses. More IPs are used until the range is exhausted and then microk8s is no longer able to run pods.
This happens even if the cluster is otherwise static, no deployments, no new nodes, etc.
- HA is disabled
- Single node
sudo ls /var/lib/cni/networks/microk8s-flannel-network
10.1.42.10 10.1.42.14 10.1.42.18 10.1.42.22 10.1.42.26 10.1.42.3 10.1.42.33 10.1.42.37 10.1.42.41 10.1.42.45 10.1.42.49 10.1.42.52 10.1.42.56 10.1.42.8
10.1.42.11 10.1.42.15 10.1.42.19 10.1.42.23 10.1.42.27 10.1.42.30 10.1.42.34 10.1.42.38 10.1.42.42 10.1.42.46 10.1.42.5 10.1.42.53 10.1.42.57 10.1.42.9
10.1.42.12 10.1.42.16 10.1.42.2 10.1.42.24 10.1.42.28 10.1.42.31 10.1.42.35 10.1.42.4 10.1.42.43 10.1.42.47 10.1.42.50 10.1.42.54 10.1.42.6 last_reserved_ip.0
10.1.42.13 10.1.42.17 10.1.42.21 10.1.42.25 10.1.42.29 10.1.42.32 10.1.42.36 10.1.42.40 10.1.42.44 10.1.42.48 10.1.42.51 10.1.42.55 10.1.42.7 lock
What Should Happen Instead?
Microk8s should gracefully stop/start without having flannel network failing periodically.
Reproduction Steps
- microk8s stop
- Make changes for example:
/var/lib/cni/networks/microk8s-flannel-network
- microk8s start
or
- Reboot
- Rinse and repeat
Introspection Report
N/A
Microk8s inspect was run and it didn't flag the network issue nor otherwise resolve the problem.
Can you suggest a fix?
The "hacky" workaround is to stop microk8s and remove the dangling IPs
sudo rm -rf /var/lib/cni/networks/microk8s-flannel-network
Are you interested in contributing with a fix?
christoph-h, ASk1 and IvanJosipovic
Metadata
Metadata
Assignees
Labels
No labels