[Veritas-ha] Issues with vxfentsthdw script and IO fencing

Wed Jul 9 01:50:01 CDT 2008

Thx Venu

So, taking into consideration that the new storage system (Quatrio) is
SCSI-3 compliant and is tested for SCSI-3 reservations [vendor claim]
what could be the cause of this issue. 

Rgds
Abhijit

________________________________

From: Venu Gadiraju [mailto:Venu_Gadiraju at symantec.com] 
Sent: Tuesday, July 08, 2008 9:46 PM
To: Abhijit Das; Sumit Sharma
Cc: Nathan Dietsch; Veritas-ha at mailman.eng.auburn.edu; Shirish
Vijayvargiya
Subject: RE: [Veritas-ha] Issues with vxfentsthdw script and IO fencing

Hi Abhijit,

>>> But, can you explain me why the preempt and abort step of the fence
test script is failing ? I tried all the steps manually in the script
and at this particular step is where the test fails. Is that an issue
going ahead. 

According to your info, Vxfentsthdw is failing in the preempt-abort
operation. The step 16 in the
[ftp://exftpp.symantec.com/pub/support/products/Database_Edition_AC/2427
23.pdf
<ftp://exftpp.symantec.com/pub/support/products/Database_Edition_AC/2427
23.pdf> ] also tries to do the same operation using vxfenadm commands. 

This might be an issue later in a split-brain situation (e.g. when both
the llt links are lost) or when a node leaves ungracefully. For example,
when there is a race following the detection of split-brain, nodes will
fail in preempting the keys of leaving nodes from the coordinator disks.
As they fail to do so, all the nodes in the cluster may unnecessarily
panic. It is hence recommended that vxfentsthdw is successful in
preempting keys on the disks before using them as coordinator disks.

Thanks,

-Venu

________________________________

From: veritas-ha-bounces at mailman.eng.auburn.edu
[mailto:veritas-ha-bounces at mailman.eng.auburn.edu] On Behalf Of Abhijit
Das
Sent: Monday, July 07, 2008 11:35 AM
To: Sumit Sharma
Cc: Nathan Dietsch; Veritas-ha at mailman.eng.auburn.edu
Subject: Re: [Veritas-ha] Issues with vxfentsthdw script and IO fencing

Thanks Sumit

Well, i can change the scripts to run at a startup level with different
numbers so that vcsmm starts before vxfen. But, thats how the install
placed the scripts in /etc/init.d ... Why is vcsmm starting before vxfen
? Any other solution ?

And, i have solved my 2nd problem. But, can you explain me why the
preempt and abort step of the fence test script is failing ? I tried all
the steps manually in the script and at this particular step is where
the test fails. Is that an issue going ahead. 

Rgds

Abhijit

________________________________

From: Sumit Sharma [mailto:sumit_sharma at symantec.com] 
Sent: Sunday, July 06, 2008 10:01 PM
To: Abhijit Da s  
Cc: Veritas-ha at mailman.eng.auburn.edu; Nathan Dietsch
Subject: RE: [Veritas-ha] Issues with vxfentsthdw script and IO fencing

Hi Abhijit,

>>Second, even though the fencing drivers startsup, I see this error at
system log. "" vcsmm: VCS RAC INFO V-10-1-15046 mmpl_reconfig_ioctl:
dev_open failed, vxfen may not be configured "" .. I searched but could
not find what this error means. As you see at last line the driver goes
into running state.

VxFEN driver should start before VCSMM. In your case, it's not happening
and that's why you are seeing this message. If you try to start VCSMM
driver after starting VxFEN, this message should not come.

Thanks.

Sumit.

________________________________

From: veritas-ha-bounces at mailman.eng.auburn.edu
[mailto:veritas-ha-bounces at mailman.eng.auburn.edu] On Behalf Of Nathan
Dietsch
Sent: Monday, July 07, 2008 7:18 AM
To: Abhijit Das
Cc: Veritas-ha at mailman.eng.auburn.edu
Subject: Re: [Veritas-ha] Issues with vxfentsthdw script and IO fencing

Hello Abjit, 

	I am trying to setup SFRAC 5.0MP1 on a 2-node T2000 cluster. OS
is Solaris 10 U5. Installation of OS, SFRAC 5.0 and 5.0MP1 has been
completed. I have also presented the desired mount points from a Quatrio
Storage System. Am now trying to test SCSI-3 persistent reservations and
I/O fencing.

IS the Quatrio Storage System supported for use with SCSI-3
reservations, not all storage systems support SCSI-3 reservations out of
the box and some require changing of settings for it to work.

I followed the settings in the VCS 5.0 Installation Guide and it worked
for me (using a Clariion), just add all disks into the fencing group in
one go, not separately as indicated by the document.  

Running the script vxfentsthdw it fails at the following step 
Preempt and abort key KeyA using key KeyB on node SystemB ..... Failed 

On the other system, I notice this error :- 
SystemA undecodable sense information: 0x0 0x0 0x0 0x0 0x0 0x0 0x0 0x0
0x0 0x0 0x0 0x0 0x0 0x0 0x0 0x0 0x0 0x0-(assumed fatal)

All other steps passes. Even, if I try manually running the vxfenadm
commands it passes, except Step 16 [from doc :
ftp://exftpp.symantec.com/pub/support/products/Database_Edition_AC/24272
3.pdf
<ftp://exftpp.symantec.com/pub/support/products/Database_Edition_AC/2427
23.pdf> ]. This is where the long read Step could not be aborted, as the
KeyA is still present. 

Please advise on this issue. 

Second, even though the fencing drivers startsup, I see this error at
system log. "" vcsmm: VCS RAC INFO V-10-1-15046 mmpl_reconfig_ioctl:
dev_open failed, vxfen may not be configured "" .. I searched but could
not find what this error means. As you see at last line the driver goes
into running state.

Jul  5 19:04:42 SystemB vcsmm: VCS RAC INFO V-10-1-15046
mmpl_reconfig_ioctl: dev_open failed, vxfen may not be configured

Jul  5 19:11:59 SystemB gab: GAB INFO V-15-1-20036 Port b gen   7c3807
membership ;1 
Jul  5 19:11:59 SystemB gab: GAB INFO V-15-1-20038 Port b gen   7c3807
k_jeopardy 0 
Jul  5 19:11:59 SystemB gab: GAB INFO V-15-1-20040 Port b gen   7c3807
visible 0 
Jul  5 19:11:59 SystemB vxfen: NOTICE: VXFEN INFO V-11-1-35 Fencing
driver going into RUNNING state 

Also, vxfenadm -d ... shows the following :- 

I/O Fencing Cluster Information: 
================================ 

 Fencing Protocol Version: 201 
 Fencing Mode: Disabled 
 Cluster Members:  

        * 0 (SystemA) 
          1 (SystemB) 

 RFSM State Information: 
        node   0 in state  8 (running) 
        node   1 in state  8 (running) 

Why is the Fencing Mode showing as Disabled ? I checked /etc/vxfenmode
and it says the parameter vxfen_mode=disabled. I changed it to enabled
and rebooted the machine, and now, fencing driver does not start at all.
I manually run vxfenconfig -c .. But get the following error message

There are three modes; SCSI3, Customized and  Disabled. You should set
it to SCSI3, this is all detailed in the install guide. 

http://sfdoccentral.symantec.com/sf/5.0/solaris/manpages/vcs/vxfenmode_4
.html

Kind Regards,

Nathan Dietsch  

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.eng.auburn.edu/pipermail/veritas-ha/attachments/20080708/6a0980cd/attachment.htm