I'm setting up a lab with oVirt 4.2. All hosts are disk-less and boot from a NetApp using iSCSI. All storage domains are also iSCSI, to the same NetApp as BFS.
Whenever I put a host into maintenance vdsm seems to try to un-mount all iSCSI partitions including the OS partition, causing the host fail.
Is this a supported configuration?
This works (with some issues, see bellow) for FC, when all luns are always connected.
For iSCSI we don't have a way to prevent disconnect since we are not aware thst you boot from one of the luns. I guess we could detect that and avoid the disconnect but nobody sent a patch to implement it.
It can work if you serve the boot luns from a different portal on same server. The system will create additional iscsi connection for the ovirt storage domains and disconnecting from storage will not affect your boot luns connection.
It can also work if your luns look like FC devices - to check this option, can you share output of:
vdsm-client Host getDeviceList
On one of the hosts?
Elad, did we test such setup?
You also need to blacklist the boot lun in vdsm config - requires this patch for 4.2:
And add multipath configuration for the boot lun with "no_path_retry queue" to avoid readonly file system if you loose all paths to storage.
Nir