LVM cleanup executes on master instead storage node

[kvaps]

Description
I just found out that fs_lvm/delete operation is executes on the main controller, but not in the compute node or any other node from BRIDGE_LIST

To Reproduce

• Add echo "$0" "$@" >> /tmp/123 line to the fs_lvm/delete file
• Create fs_lvm datastore with BRIDGE_LIST
• Create VM
• Remove VM
• Check /tmp/123 on node and on container
• On controller you will see the next output:

/var/lib/one/remotes/tm/fs_lvm/delete opennebula:/var/lib/one//datastores/138/5826/disk.0 5826 100

Expected behavior

opennebula - it is my controller node.
I expected some other name instead, like name of compute node or name of node from BRIDGE_LIST.

/var/lib/one/remotes/tm/fs_lvm/delete <some_other_node>:/var/lib/one//datastores/138/5826/disk.0 5826 100

Details

• Affected Component: Storage drivers
• Hypervisor: KVM
• Version: 5.6.0

Additional context

There is also cleanup procedure before creating and resizing drive, so, it is up to discussion is that really needed here to cleanup it twice.

Progress Status

• Branch created
• Code committed to development branch
• Testing - QA
• Documentation
• Release notes - resolved issues, compatibility, known issues
• Code committed to upstream release/hotfix branches
• Documentation committed to upstream release/hotfix branches

[rsmontero]

All actions are executed locally. Each action then ssh to the target host (if needed, depending on the driver and/or specific action). For example in fs_lvm/delete, you have:

if [ -n "${DS_SYS_ID}" ]; then                                                        
    ssh_exec_and_log "$DST_HOST" "$ZERO_CMD" \                                        
        "Error cleaning $DST_PATH"                                                    
                                                                                      
    LOCK="tm-fs_lvm-${DS_SYS_ID}.lock"                                                
    exclusive "${LOCK}" 120 ssh_exec_and_log "$DST_HOST" "$DELETE_CMD" \              
        "Error deleting $DST_PATH"                                                    
else                                                                                  
    ssh_exec_and_log "$DST_HOST" "$DELETE_CMD" \                                      
        "Error deleting $DST_PATH"  

ssh_exec_and_log do the ssh part.

[kvaps]

I just found that this is happens because this action is actualy use not for system datastore but for images datastore, for cleanup non persistent drives, and took BRIDGE_LIST parameter from it.

Sorry wrong bug.

[kvaps]

I was wrong about wrong bug. Bug is still there:

To Reproduce

• Deploy new VM
• Undeploy this VM
• Destroy this VM

You will see that driver action will be called with wrong DST_HOST in the path (there will be controller hostname)

• Then try to recover --retry

You will see that driver action will be called with right DST_HOST in the path (there will be hostname of the last compute node)
You wil lse that driver action

Please reopen issue.

[xorel]

Ready to merge:

• one: F #2352 #2359 #2981 #3130 #3233: Fix fs_lvm cleanup and offline migration issues #3201
• docs: F OpenNebula/one#2352: Add LVM fixes to whats new docs#584
• tests (already merged): https://github.com/OpenNebula/infra/pull/228