SCSI reservation is not cleaned always during cluster shutdown

[Anbu]

I am facing the below listed issue with my RAC cluster on XP storage array.  But I am not able to access hot fix 7 .
The system is hanging and to recover I have to power cycle every time when I do the hastop -all.


SCSI PGR reservations are not fully removed with I/O fencing on 5.0MP1 RP3 or earlier (5.0) on Hitachi arrays. Service groups will fail to move from one node to another intermittently and will have SCSI reservation errors in the /etc/vx/dmpevents.log.

Solution:
This is a know issue noted in etrack 1200083 which is the super incident. It is also noted in etrack 1165353 with specifics.  This is fixed with hot fix 7 HF7.  It does not seem to be in RP4.  The noted incident in the readme file of the HF7 is 1202403.

[Pam_Pledger_2]

Those etracks apply are when not all paths in an A/A array do not get the keys written -- not removed.   So, we'd need further details to determine if they were in fact the issue you are having.

Hitachi arrays need to be configured correctly.  The Hitachi Array's  support multiple PGR modes, and if the Hitachi array is setup for another cluster's pgr mode, it will have this problem.  Please check the Hitachi Array guide and set it up for Veritas RAC Cluster  (versus the Sun Cluster.)   In technote 275481, it describes some of the settings.  As the settings  are based on the array, the Hitachi Array Guide or Hitachi Technical Support is the best source for the correct settings.

Also, are you using scsi or dmp mode in /etc/vxfenmode?  If you're having problems, it might help to switch vxfenmode to another mode.

Lastly, running "hastop" should not lead to a power cycle.  If running Rac 10 or 11, please verify that a cssd resource is setup to shutdown cssd before the PrivNIC, Ocr and/or Vote disks are taken offline by VCS.    If CRS looses any of  these resources, CRS  will attempt to panic the system.     A power cycle should not be needed.   Please check to see that the eeprom settings will reboot on reset or panic.

Please contact technical support if you need further details or these do not resolve the problems.