MSA500 G2 and multipath

[cristian mazza]

hi all,

i have a msa500 g2 with a 4 port scsi i/o module and 2 proliant dl 385 with 2 smartarray 642.

i try to use the md multipath to install service guard but i have problem when i try to mount the device /dev/md0.

if i use the phisical device /dev/cciss/c1d1p1 or /dev/cciss/c2d1p1 it works fine and i can write my data.

when i try to mount the device /dev/md0 i have a lot of error in /var/log/messages about the io paths and the mount command freeze.

i use this command to create the md multipat mdadm -C /dev/md0 --level=multipath --raid-disks=2 /dev/cciss/c1d1p1 /dev/cciss/c2d1p1, after i create the fs raiuser or ext3 without problems. but when i try to mount it fails.

can anyone help me?

regards, Cristian Mazza

[cristian mazza]

the operating system is SLES 9 sp3 x86_64

[Serviceguard for Linux]

If I am remembering correctly, what you are doing is correct. Is there any detail in the system log when the mount fails?

[cristian mazza]

Dec 29 15:39:55 cssrvfe01 kernel: multipath: IO failure on cciss/c1d2p1, disabling IO path.
Dec 29 15:39:55 cssrvfe01 kernel: Operation continuing on 1 IO paths.
Dec 29 15:39:55 cssrvfe01 kernel: multipath: cciss/c1d2p1: rescheduling sector 66104
Dec 29 15:39:55 cssrvfe01 kernel: MULTIPATH conf printout:
Dec 29 15:39:55 cssrvfe01 kernel: --- wd:1 rd:2
Dec 29 15:39:55 cssrvfe01 kernel: disk0, o:0, dev:cciss/c1d2p1
Dec 29 15:39:55 cssrvfe01 kernel: disk1, o:1, dev:cciss/c2d2p1
Dec 29 15:39:55 cssrvfe01 kernel: MULTIPATH conf printout:
Dec 29 15:39:55 cssrvfe01 kernel: --- wd:1 rd:2
Dec 29 15:39:55 cssrvfe01 kernel: disk1, o:1, dev:cciss/c2d2p1
Dec 29 15:39:55 cssrvfe01 kernel: multipath: cciss/c1d2: redirecting sector 66064 to another IO path
Dec 29 15:39:55 cssrvfe01 kernel: multipath: only one IO path left and IO error.
Dec 29 15:39:55 cssrvfe01 kernel: multipath: cciss/c2d2p1: rescheduling sector 66104
Dec 29 15:39:55 cssrvfe01 kernel: multipath: cciss/c2d2: redirecting sector 66064 to another IO path
Dec 29 15:39:55 cssrvfe01 kernel: multipath: only one IO path left and IO error.
Dec 29 15:39:55 cssrvfe01 kernel: multipath: cciss/c2d2p1: rescheduling sector 66104
Dec 29 15:39:55 cssrvfe01 kernel: multipath: cciss/c2d2: redirecting sector 66064 to another IO path
Dec 29 15:39:55 cssrvfe01 kernel: multipath: only one IO path left and IO error.
Dec 29 15:39:55 cssrvfe01 kernel: multipath: cciss/c2d2p1: rescheduling sector 66104
Dec 29 15:39:55 cssrvfe01 kernel: multipath: cciss/c2d2: redirecting sector 66064 to another IO path
Dec 29 15:39:55 cssrvfe01 kernel: multipath: only one IO path left and IO error.
Dec 29 15:39:55 cssrvfe01 kernel: multipath: cciss/c2d2p1: rescheduling sector 66104
Dec 29 15:39:55 cssrvfe01 kernel: multipath: cciss/c2d2: redirecting sector 66064 to another IO path
Dec 29 15:39:55 cssrvfe01 kernel: multipath: only one IO path left and IO error.
Dec 29 15:39:55 cssrvfe01 kernel: multipath: cciss/c2d2p1: rescheduling sector 66104
Dec 29 15:39:55 cssrvfe01 kernel: multipath: cciss/c2d2: redirecting sector 66064 to another IO path
Dec 29 15:39:55 cssrvfe01 kernel: multipath: only one IO path left and IO error.
Dec 29 15:39:55 cssrvfe01 kernel: multipath: cciss/c2d2p1: rescheduling sector 66104

[cristian mazza]

the log report c2d2 infact i have 3 logical drive defined on msa500 and it report the errors when i try to mount /dev/md3.

it is just a log.

regards, Cristian

[Serviceguard for Linux]

Rather than partitioning first and then doing multipath, can you try doing multipath and then doing partioning.

mdadm -C /dev/md0 --level=multipath --raid-disks=2 /dev/cciss/c1d1 /dev/cciss/c2d1

Then run fdisk to partition /dev/md0.

[cristian mazza]

hi,

i try with /dev/cciss/c1d3 and /dev/cciss/c2d1 to create /dev/md3.
and it fails when i try to mount /dev/md3

instead if i try to use the dd command to write on /dev/md3 it works without errors.

regards, Cristian

[cristian mazza]

if i use ext2 it works....

[cristian mazza]

now, i upgrade the kernel and it works fine.

the last problem is with the lock lun on md device.

any ideas?

[Serviceguard for Linux · Accepted solution]

The MD driver does not support some of the functions required by LockLUN. So for MSA500 you must use single path for LockLUN.

This does not significantly affect overall availability. Assume you have a cluster using MD for other devices and single path for LockLUN. If the path to LockLUN fails from one of the servers (call is ServerA), you will get messages in the system log but there is no other affect in the cluster. ONLY if the other server (ServerB) fails before the repair is made to ServerA will you have a problem. There is no guarantee to survive dual failures - although we try. This is dual failure case.

More explaination may be available in this white paper http://docs.hp.com/en/B3936-90078/B3936-90078.pdf

[cristian mazza]

I could use for Server A the path /dev/cciss/c1d0p1 (first smartarray) and for Server B /dev/cciss/c2d0p1 (second smartarray).

do you think that this may be a good solution? or not?

do you think that the quorum server may be a preferred solution?

regards, Cristian

[Serviceguard for Linux]

Your LockLUN solution is fine.

On Quorum Service vs LockLUN. Quorum service is easier to manage in some ways, but it must reside on a computer outside of the cluster. Some customers do not like that. In some cases, failover can be a little faster with Quorum Service. If the server running QS fails, the cluster keeps running (if there are no there failures within the cluster). if that computer does fail, it can be reapired without any impact at all to the cluster.

Since LockLUN is entirely within the cluster there is no extra hardware. If there is a failure on the single path to the LockLUN because of an HBA failure, the repair should be scheduled. In order to do that repair, the node needs to come down so there is some impact since the packages need to be moved over to the other node.

With that information, you should be able to make a decision appropriate for your environment.

Regards