Yes, we noticed that issue on hard node failure. It shouldn't happen on Pod restart or if you drain a node.
If we don't go through the normal Pod lifecycle the driver doesn't receive the NodeUnpublish call and therefore doesn't clear the client IP from the export.
Flo_csI
2 Intern
•
167 Posts
0
May 5th, 2021 00:00
Hi @thomas.menard,
Yes, we noticed that issue on hard node failure. It shouldn't happen on Pod restart or if you drain a node.
If we don't go through the normal Pod lifecycle the driver doesn't receive the NodeUnpublish call and therefore doesn't clear the client IP from the export.
Also, we developed a module named CSM resiliency to improve the behavior on node failure : https://www.youtube.com/watch?v=veFRQ7xvoT8
Once ready for PowerScale, it will clean the export in such a case and there will be no need for manual cleanup ; even in such a case.
Hope this helps.
thomas.menard
3 Posts
0
May 5th, 2021 14:00
Well we have encountered this issue on node drain last time we did maintenance on one of our cluster (about 50 pvc).
It's quite annoying to clean all exports on cli....
But this issue is also occuring on pod restart but not so often, I'll try to gather some logs on the operator part just in case you find something.