Red Hat Bugzilla – Bug 1394138
Shared storage mount point should be unmounted as part of peer detach
Last modified: 2018-10-31 00:29:09 EDT
Document URL: Section Number and Name: Describe the issue: When a node is to be deleted from an HA cluster in ganesha and from gluster ,then first umount needs to be performed. If umount is not performed and a node is to be deleted from HA Cluster ganesha and then doing peer detach of that node, that volume is still accessible in /var/run/gluster/shared_storage/ location even after removing node in HA-Cluster /usr/libexec/ganesha/ganesha-ha.sh --delete /var/run/gluster/shared_storage/nfs-ganesha/ dhcp37-169.lab.eng.blr.redhat.com gluster peer detach dhcp37-169.lab.eng.blr.redhat.com Suggestions for improvement: Additional information:
Hi Anjana, The document will be updated with the mentioned chnage for the 3.2 release. Assigning it to myself.
Based on the discussion with the engineering (Soumya, Avra, Surabhi, Manisha), it was decided to mark this as a bug fix for any upcoming release. And, wrt doc this has to be changed to "Known Issue" for 3.2.
From the discussions with QE, looks like when a new node is added to the gluster cluster, shared storage volume gets auto-mounted but in case of peer detach, it doesn't get unmounted. We planned to document this as known_issue until it gets fixed. Assuming this needs to be taken care as part of hook-scripts of gluster peer detach CLI, changing the component to glusterd. Kindly re-assign/correct the components if that is not the case.
As fas as shared storage is concerned, the doc text looks good to me. I would like Soumya to have a look at the NFS-Ganesha bits of it.
The doc text looks good, but since this issue is not just specific to NFS-Ganesha, have made few corrections. Please check the same- If a node is detached from the Gluster storage pool with cluster.shared_storage enabled, the shared volume shall still be accessible in /var/run/gluster/shared_storage/ location even after removing the node. Workaround: Before a peer is detached from the cluster, manually unmount the shared storage on that peer.
This looks much better and simple.
Sunny - What's the plan on this BZ? Are we planning to address this in any of the upcoming release?
Sunny - Did you get a chance to estimate the size of the fix?