SCSI3 PGR operations on a VxDMP dmpnode result in dmp path disablement with RHEL5U8 and later kernels.

Article: 100008814
Last Published: 2012-08-31
Ratings: 0 0
Product(s): InfoScale & Storage Foundation

Problem

SCSI3 PGR (Persistent Group Registrations) operations on a VxDMP (VERITAS Dynamic Multi-Pathing) dmpnode result in VxDMP path disablement.


Side effects of the issue include the following:
• Reservation conflict immediately followed by VxDMP error V-5-0-112, whenever SCSI3 PGR operations are executed on a dmpnode. 

         Operations that will issue SCSI3 PGR operation on a dmpnode include: 
          - Stopping the cluster with hastop.
          - Issuing `/etc/init.d/vxfen start | stop`.  This command is automatically executed at server boot and shutdown.
          - Deporting a diskgroup that contain disk(s) with registrations or importing a diskgroup with groupreserve option.

• When a node is rebooted. The surviving node(s) may see diskgroups go into dgdisabled state and file system(s) get disabled. If VCS is managing resources you will see resource faults. This occurs because dmp paths and dmpnodes are being disabled as a result of multiple SCSI3 PGR operations. 

• When a node is rebooted. The surviving node(s) may get paniced. The panic will be initiated by VxFEN as it is trying to avoid a split brain condition. This occurs because dmp paths and dmpnodes are being disabled as a result of multiple SCSI3 PGR operations, then reenabled by the VxDMP recovery daemon.

 

Error Message

kernel: sd 2:0:0:14: reservation conflict
kernel: VxVM vxdmp V-5-0-112 disabled path 66/0x0 belonging to the dmpnode 201/0xd0 due to path failure

Cause

RHEL5U8 kernel SCSI layer error handling routine introduced a new error type:

DID_NEXUS_FAILURE

This new error type is not handled properly by VxDMP, resulting in dmp paths getting disabled during SCSI3 PGR operations.

Solution

If you are planning to upgrade to RHEL5U8 or are currently running RHEL5U8 kernel 2.6.18-308.el5 or later kernels and are running VxVM 5.1 and later. Veritas recommends installing VxVM5.1SP1RP2P3HF5

If you are planning to upgrade to RHEL5U8 or are currently running RHEL5U8 kernel 2.6.18-308.el5 or later kernels and are running VxVM 6.0 or 6.0RP1. Veritas recommends installing VxVM6.0RP1HF1

Workarounds:

Downgrade the kernel to pre RHEL5U8.
Configure vxfenmode in disabled mode.

Note: VERITAS Storage Foundation 6.0.1 will contain the fix and is slated for public release in early September. Rolling patch 5.1SP1RP3 will contain the fix and is slated for public release in early October. Rolling Patch 5.1SP1RP3 will also contain the vxfen patch noted in related articles.

 

Applies To

RHEL5U8 running kernel 2.6.18-308.el5 and later
Storage Foundation 5.1 and later or Storage Foundation 6.0, 6.0RP1
VxFEN is configured and enabled with SCSI3 disk based fencing in either raw or dmp mode.
VxDMP is configured.

 

References

Etrack : 2866527

Was this content helpful?