LVM2 Mirroring + Multipathing

[benoit Bruckert]

Hello,
I'm trying to configure a production box with OS=SLES 10 connected to a SAN.
the system will be stored on a SAN disk wich works very well.
I added then multipathing (dm-multipath) and all is correct, if I remove a link i can still see the disk through another path ! For information, the HBA is a qlogic with dual port.
The second step is to have a mirror on the system (or any other logical volume).
I then used lvconvert on the system, and stored the copy in another disk (LUN). The goal is to have high availability and continue to work if a disk (LUN in this case) failed.
lvconvert worked great and the system is then mirrored .
To test the case, I disconnected 1 of the 2 LUN in way that even multipathing cannot access this lost LUN, and then, the system froze, I can still ping the server but that's all. I have no log of the failure.
I thought the trouble was on this mirror LVM, then instead of lvm mirroring, I tried the md RAID device. Configuration is a little bit harder, but after all, I had a mirrored disk with the system on it and multipathing under it. I did the same test as for LVM mirroring, and I had the same trouble ! The system is frozen !
Does anyone already built such a configuration . I E : SAN - multipathing - LVM2/mirroring and Linux ?
thanks in advance...

[Ivan Ferreira]

I did that configuration, for multipath, I used the qlogic multipath enabled driver. I suggest you to stay with mdadm for RAID configuration and use LVM on top of it.

[Brem Belguebli]

Hi,
When you say no log of the failure, you should, at least, see timeout messages in your syslog if the disappeared LUN is being used.

When you disconnect 1 of the 2 LUNs, are you able to remain connected and type multipath -ll -v2 .

Do you get a result or the command just freezes and never returns ?

I did the same tests, and no such behavior happened.

I use DM-MP / md / LVM .

LVM mirroring is "bizarre", ie need for a third device for the logs, no way to resize a mirrored LV, etc...

Brem

[benoit Bruckert]

Thanks for these quick answers...
I'm OK to play with md, which is what i try first (before lvm mirroring). But the trouble I have is the same in both cases.
The system hangs , I can play with commands in memory but that's all, after a few seconds it's frozen (i think in case of any disk access)!! And after 2 hours I don't think there's a timeout. And of course no logs ! May be I should try this crash test on another disk than the system one , in order to have something in the log this time. But this will take me some time :). I'll check after the W.E..

Concerning the driver, i'm using the qla2xxx, but I've just check the documentation and there are some parameters in the BIOS setting to check. May be I have something wrong ?
Is there something specific in the driver to configure in order to manage multipathing/failover ?

If you have any other ideas , i accept all of them..

[benoit Bruckert]

Hi,
I tested this time a disk failure on a non system disk !
I had the system log which say that the disk is out of order, a message from multipath to inform the failing of the path to these disks !
I did some disks accesses to the other part of the mirror and the filesystem is still alive (lvm commands hang for a few minutes - but i have always accesses to the system). I made some cp on this filesystems, bring back the failed LUN and magically all is OK !
My question now is : What is the trouble with system volume group mirroring ?