Fibre attached VG not activating

[CGC Unix Team]

Setup:
EMC 8530 / Brocade 3800's / HPUX 10.20 on K420 @ current patch levels w/ A3404A fibre cards (arbitrated loop)

Problem:
At boot time, when the volume group vg01 is trying to activate, it complains about not being able to see the disk in the vg, so the vg doesn't get activated. After the box is up (run level 4), the vg can be activated and the file systems mounted without a problem.
It would appear that the fibre channel drivers are not loading early enough.

Thanks for any help,

Steve.

[Bill McNAMARA_1]

What does /etc/rc.log say..

Later,
Bill

[Eugeny Brychkov · Accepted solution]

Steve,
and what 'dmesg' says? Attach here
Eugeny

[CGC Unix Team]

/etc/rc.log doesn't say anything because the first entry there is for 'localmount'. Volume groups are activated by /sbin/bcheckrc which calls /sbin/lvmrc to do the work.

I have attached my dmesg ouput. Strange, it looks like it see's the disks?

[Eugeny Brychkov]

Private loop, two HBAs (looks like direct connect), EMC alpas are 66 and 77, LUNs configured: 0, 16, 17, 18, 19, 122. All these LUNs are seen at startup. No errors in dmesg.
Check if there are any missing LUNs configured to be seen from host
Eugeny

[Bill McNAMARA_1]

seems your system had crashed::
check
/var/adm/crash
and
/var/adm/tombstones

WARNING: Insufficient space on dump device to save full crashdump.
Only 809500672 of 2147484672 bytes will be saved.
Dump device table: (start & size given in 1-Kbyte blocks)
entry 0 - major is 31, minor is 0x5000; start = 52064, size = 790528


What do your /etc/rc.config.d/fc*
and /etc/rc.config.d/hpfc*

fcms startup scripts configure.

[CGC Unix Team]

/etc/rc.config.d/hpfcmsconf has FCMS_START=1 only.
Just to be clear, the HBA's are going through the 3800's using quickloop.

I don't see the LUNS you say are configured (0,16,17,18,19,122), even in the dmesg, I may be reading it wrong. vg01 only has t2d0, t2d1, t2d2, t2d3.
I should mention that this is a server recently created from a current Ignite tape, with the vg's imported. The old hardware used SCSI to connect to disk, so the fibre is new to this config, and maybe the Ignite didn't set something up correctly?

[Eugeny Brychkov]

Steve, according to FC volume set addressing rules LUN#='d'+8*'t'+bus*128, where d and t are from device file (and ioscan), bus is bus from ioscan. So:
t2d0 = LUN16, t2d1 = LUN17, t2d2 = LUN18, t2d3 = LUN19.
All these LUNs are seen on server's startup time. Did you install latest FC driver? Did you patch system (GR and HW enablement)?
Eugeny

[CGC Unix Team]

Sorry for the delay, thanks for the info on LUN numbering. Yes the system is at the current patch levels (Custom Patch Manager from this site) The J3630BA driver is also installed.

[Eugeny Brychkov]

Is this problem still exists? Try to get more info from this erroneous output when system starts. Is there disk path, device file or anything else pointing to device displayed when vg01 does not activate? I see nothing in dmesg... maybe anything in syslog.log?
You can to the following test:
- boot to single-user mode;
- mount all '/etc/mount -a';
- issue ioscan to see that all these LUNs are seen;
- issue pvdisplay for ALL these disks within vg01 and check all outputs to see if there's anything wrong
Eugeny

[ANDRIEU David]

Hi Steve,

It seems I have the same problem with EMC 8530 / EMC switch / HP-UX 11.0.

Do you resolve "VG not activating" problem?

Thanks in Advance,

Dave

[Ramesh Donti]

Steve,

We are seeing the same problem connected to HITACHI (HDS) 9910 (ie HP XP48). We have K360 with 2 x A3403A, latest patch/driver level, connected to Brocade 3800 with fw 3.02p in Quickloop. When you reboot, we see inconsistant behaviour, sometimes we see all the luns down one HBA, then the other, then none at all. Once the system is up, if you reset the cards using fcmsutil and do an ioscan -fn, you will see all the luns. However, if you attempt to use the luns down a path with has failed, the disks will constantly sit at 100% util, with very large times for avwait and avserv with "sar -d".

Also, on the 3800, when the server has failed to see the luns, the port is set to "loopback". So, we believe the problem is between the HBA and the 3800 switch port - not between the Disk Array and the 3800.

HP currently only support brocade 3800 fw 3.02K which we are about the load and test. Unfortunately, we did not purchase our 3800 via HP, but Hitachi, who said that 3.02p is fine. Prior to 3.02p, we had 3.02j, and Quickloop didn't work at all with other HBAs - A5158A, A6795A all set to quickloop on the switch. A mix of Fabric and QL didn't work either on this 3.02j, and only after updating to 3.02p did it work - With/without Zoning. However, now we have a problem with A3404A only in QL.

I'm counting on 3.02K fixing this problem, since HP support 3.02F and 3.02K ONLY on the 3800.

.....to be continued.

[Philip P. Hartl]

I am seeing the same problem intermittenly using a relatively simple configuration, 2-A3404A cards providing primary/alternate paths to a fc60/sc10 array. Sometimes on boot, some luns are not seen (not all), sometimes all are seen. HP has not been able to resolve. Suggests it may be a timming issue. My solution has simply been to allow the boot to continue and activate the vg's after.