[BUG] The cni pod network cache cleanup does not work in all cases #8150
Description
Describe the Bug
The file /system/oem/99_cni_reset.yaml targets to cleanup the cached files.
But on v1.5.0 cluster, it seemed not work all the time
harv41:~ # date (cluster newly rebooted)
Mon Apr 28 10:45:22 UTC 2025
harv41:~ # ls /var/lib/cni/networks/k8s-pod-network/ -alth
total 344K
drwxr-xr-x 2 root root 4.0K Apr 28 10:36 .
-rw------- 1 root root 70 Apr 28 10:36 10.52.0.160
-rw------- 1 root root 11 Apr 28 10:36 last_reserved_ip.0
-rw------- 1 root root 70 Apr 28 10:36 10.52.0.159
...
-rw------- 1 root root 70 Apr 28 10:35 10.52.0.100
-rw------- 1 root root 70 Apr 25 09:22 10.52.0.98
-rw------- 1 root root 70 Apr 25 09:22 10.52.0.97
-rw------- 1 root root 70 Apr 23 12:14 10.52.0.37
-rw------- 1 root root 70 Apr 23 12:14 10.52.0.36
-rw------- 1 root root 70 Apr 17 06:30 10.52.0.151
-rw------- 1 root root 70 Apr 17 06:30 10.52.0.150
-rw------- 1 root root 70 Apr 16 09:00 10.52.0.75
-rw------- 1 root root 70 Apr 16 09:00 10.52.0.74
-rw------- 1 root root 70 Apr 14 12:44 10.52.0.14
-rw------- 1 root root 70 Apr 14 12:41 10.52.0.10
-rw------- 1 root root 70 Apr 14 12:40 10.52.0.2
-rw------- 1 root root 70 Apr 14 12:40 10.52.0.250
...
-rw------- 1 root root 70 Apr 11 13:23 10.52.0.207
-rw------- 1 root root 70 Apr 9 19:14 10.52.0.138
-rw------- 1 root root 70 Apr 9 19:01 10.52.0.60
drwxr-xr-x 3 root root 4.0K Apr 9 18:57 ..
-rwxr-x--- 1 root root 0 Apr 9 18:57 lock
To Reproduce
-
Cold start a v150 cluster
-
Check files
ls /var/lib/cni/networks/k8s-pod-network/ -alth -
Per https://rancher.github.io/elemental-toolkit/docs/customizing/stages/#initramfs, when following file is used, the cached files are cleand.
harv41:~ # cat /oem/99_cni_reset.yaml
name: "reset container dhcp leases"
stages:
initramfs:
- name: "clean network"
commands:
- rm -rf /var/lib/cni/networks/k8s-pod-network
harv41:~ # ls /var/lib/cni/networks/k8s-pod-network
ls: cannot access '/var/lib/cni/networks/k8s-pod-network': No such file or directory
harv41:~ # ls /var/lib/cni/networks/
harv41:~ #
after some time when the node has deployed all workloads:
harv41:~ # ls /var/lib/cni/networks/k8s-pod-network/
10.52.0.10 10.52.0.15 10.52.0.2 10.52.0.24 10.52.0.29 10.52.0.33 10.52.0.38 10.52.0.42 10.52.0.47 10.52.0.52 10.52.0.57 last_reserved_ip.0
10.52.0.11 10.52.0.16 10.52.0.20 10.52.0.25 10.52.0.3 10.52.0.34 10.52.0.39 10.52.0.43 10.52.0.48 10.52.0.53 10.52.0.6 lock
10.52.0.12 10.52.0.17 10.52.0.21 10.52.0.26 10.52.0.30 10.52.0.35 10.52.0.4 10.52.0.44 10.52.0.5 10.52.0.54 10.52.0.7
10.52.0.13 10.52.0.18 10.52.0.22 10.52.0.27 10.52.0.31 10.52.0.36 10.52.0.40 10.52.0.45 10.52.0.50 10.52.0.55 10.52.0.8
10.52.0.14 10.52.0.19 10.52.0.23 10.52.0.28 10.52.0.32 10.52.0.37 10.52.0.41 10.52.0.46 10.52.0.51 10.52.0.56 10.52.0.9
Expected Behavior
Cached files are cleaned up.
Support Bundle for Troubleshooting
not required
Environment
- Harvester version: v1.5.0
- Impacted VM:
- Impacted volume (PV):
- Underlying Infrastructure (e.g., Baremetal with Dell PowerEdge R630): KVM VM based
- Rancher version: v2.11.0
Additional context
Original issue #7471 has been verified.
Maybe elemental has some changes.
Workaround and Mitigation
No response
Metadata
Metadata
Assignees
Labels
Type
Projects
Status