Tuesday, November 27, 2012

EMC ~management storage group problems

I was recently working with a client that had an EMC CX4-120 and we were trying to setup a Cisco UCS blade cluster to boot from iSCSI on it.  Another vendor was doing the UCS portion so I was tasked with getting the EMC side setup.  Through Unisphere I had to go in and manually create the hosts and register their IQN's.  Next I created the LUN's and the storage groups.  When assigning the LUN's to the storage group, make sure that the "Host LUN ID" field gets changed if the initiator will eventually be in multiple groups where LUN ID 0 is already taken.  So in the end, each blade had a storage group that was mapped to a LUN with a unique LUN ID to be used for iSCSI booting ESXi. 

Great, but the hosts would not connect. 

We kept getting a warning "EV_TargetMapEntry::UpdateTargetMapEntry - different ports" and an information message saying iSCSI login failure.  I went ahead and opened up a case with EMC support because from what I could tell the configuration looked correct at this point.  We weren't really getting anywhere until after some clicking around I noticed the ~management storage group showed all the blades in it and so did the individual storage groups created for the blades.  I recalled a previous experience similar to this where we had to get the hosts out of the ~management group.  Simply removing them from that group doesn't work.  I ended up having to remove them from their individual groups, apply the settings, then re-add them to the individual groups to get them to disappear from the ~management group.  Once that was done, the blades were able to see their respective LUN's and life was good.

A couple of notes:

* Re-registering the hosts manually seems to have caused them to go back to the ~management group.
* I'm not positive is the Host LUN ID is able to stay 0 or not, but I did not want to risk a LUN ID conflict since production data was on a LUN that already had ID 0 for another storage group (one that the blades will eventually belong to).
* If you are using Active/Failover when registering a host connection, the UCS blades will show one success and one failure if setup correctly (at least what I think is correctly!).
* To deregister certain things required being in Engineer mode.  To get into that mode from Unisphere, press Ctrl+Shift+F12 and type in the password.  Appears to be either "SIR" or "messner" depending on how old your firmware is.  Don't use this mode unless you are on the phone with support and absolutely sure you know what you are doing.

Sorry if I am missing some details, just wanted to do a brain dump while it was still semi-fresh.