-
Bug
-
Resolution: Unresolved
-
Normal
-
None
-
rhos-17.1.3
-
None
-
False
-
-
False
-
?
-
rhos-conplat-observability
-
None
-
-
-
-
-
Refinement
-
1
-
Moderate
To Reproduce Steps to reproduce the behavior:
- Compute node up with "collectd" container with "ipmi" device present [1]
- Major iD of "/dev/ipmi0" device on host compute node gets modified due to any reason [2]
- "collectd" container will be unable to bind again the device "/dev/ipmi0", hence IPMI plugin will be impacted [3]
[1]
~$ grep -A 1 "--device" sos_commands/podman/containers/podman_inspect_9f3f11fa346f
"--device",
"/dev/ipmi0",~
[2]
~$ grep ipmi sos_commands/block/ls_-lanR_.dev | tail -n1
lrwxrwxrwx. 1 0 0 8 Sep 6 22:13 238:0 -> ../ipmi0~
[3]
$ head -n1 sos_commands/podman/containers/podman_inspect_9f3f11fa346f
time="2025-09-29T12:38:46+05:30" level=warning msg="Could not locate device 236:0 on host"
Error message: [2025-09-29 10:02:04] ipmi plugin: c_ipmi_read: I'm not active, returning false. [2025-09-29 10:02:04] read-function of plugin `ipmi/main' failed. Will suspend it for 86400.000 seconds.
Expected behavior
- "collectd" container should continue to bind the device "/dev/ipmi0" from host
Bug impact
- IPMI plugin within collectd conatiners stops working impacting production
Known workaround
- Recreation of "collectd" container binds back the ""/dev/ipmi0" device
Additional context
- major:minor number of "/dev/ipmi0" device is dynamically assigned and can get changed
- any running container that had "/dev/ipmi0" bind-mounted will continue referencing the old inode, which no longer exists post modification. Hence the container will not see the new "/dev/ipmi0" until it is recreated (not just restarted)
- instead of "- - device /dev/ipmi0", creating the "collectd" container with volume mount "/dev" along with "--privileged" flag will be able to prevent the issue