Windows Host Paths May Become Unavailable During Node or Node Canister Replacement Procedures

Content

If a Storwize V7000 node canister or SAN Volume Controller node is removed from its cluster then subsequently powered off, the paths to this node as seen from an attached Windows 2008 or Windows 2012 hosts may remain in a "dead" state and not automatically recover when the node is re-added to the cluster.

If the I/O group partner node is then removed, this will result in a loss of access to Volumes for these hosts.


Resolution

When performing Storwize V7000 or SAN Volume Controller node or node canister replacement procedures, paths to the node from Windows 2008 hosts should be monitored. If these paths do not automatically recover after the first replacement node has rejoined the cluster, these should be manually re-added by performing a LUN rescan on the host.

Alternatively

For Emulex configurations the node timeout setting in the Emulex driver may be decreased from the default 30 seconds to 3 seconds which will allow the paths to automatically recover.

For Brocade configurations the Path timeout setting in the Brocade driver may be decreased from the default 30 seconds to 3 seconds which will allow the paths to automatically recover.

Note the above should be done prior to replacing the nodes.


Windows Clustered Configurations


During a node hardware replacement with SAN Volume Controller or Storwize version 7.3 and higher Windows 2008 or Windows 2012 clusters may suffer permanent path loss when the second node of the IO group is removed from the cluster.

Resolution

This issue is resolved in SDDDSM Version 2.4.4.0 and higher.

With earlier versions of SDDDSM

Introduce the first replacement node to the cluster. When the node is powered up and has joined the cluster issue a warmstart on this node. The warmstart on the initial node should be performed prior to powering down or removing the second node from the cluster

svctask stopnode -warmstart

This will allow the paths to be recovered on the first replacement node prior to the second node being removed.

Alternatively re-scanning devices from the Windows server after the first node has been replaced will also ensure that paths are recovered prior to the second node being removed.