linux and eva 5000

[unre4l]

Has anyone here successfully configured a RHEL4 host using an hp HBA to connect to a EVA 5000 using the hp linux driver kit?
We are having some issues with failover and failback and would like to know if someone here has this working the way it should be.

Thanks.

[Steven E. Protter]

Shalom,

Tell us more about your system.

HBA choice is important.

With HP servers and qlogic cards you should not be having any issues.

Details on your configuration would be very helpful in moving to resolution. What error is received in /var/log/messages when failover does not work?

SEP

[Rob Leadbeater]

Hi,

Yes I've had this working..
What issues are you seeing ?

Cheers,

Rob

[DanAm]

Hi there,
there is a change going on with multipath and linux. The older distros use "qlogic failover", but this relies on "/proc/scsi/scsi/qla2xxx/*".
This directory is no longer present in say RHEL4u4 or SLES10. The way to go seems to be Linux very own "dm-multipath" (see attached HP pdf)
HTH
Dan

[unre4l]

We are using the FC1242SR dual port card and the hp linux driver kit. After installing the driver when we use sansurfer to view the paths it does not show 4 paths to the EVA 5000. It shows only 3 paths. If we do a failover then the luns start using the alternate path. When the failed path is restored the luns to which data was being written stays on the failed over path and the other luns fail back. I guess this is part of static load balancing??
At this point if we reboot the system then it fails to see the luns that did fail over. At boot time only port1 is used to scan the luns and port 2 is inactive. So it errors out on the luns that failed over to port2. Once the system comes up since both paths are there when i do fdisk i can see all the devices.

The workaround was after i tried a fail over, say i pull out port1 and then restored it (at this point some luns are using port 1 and some on port2), then if i remove port 2 so that everything failed back to port1 and then i reboot then all the luns are scanned correctly.
Is there any setting that needs to be changed on the card itself. My /etc/hp_qla2x00.conf looks like this by default
qdepth = 16
port_down_retry_count = 30
login_retry_count = 30
failover = 1
load_balancing = 1
auto_restore = 0x80

If you have it working what drivers and kernel are you using and does san surfer see 4 paths to the eva? Does the failover behave the same way as above?

As for the dm-multipath the document says not to use it in a production environment??

[Uwe Zessin · Accepted solution]

If I remember correctly, the device mapper is only supported on active/active EVAs - the EVA3000/5000 would need to run VCS V4 firmware.

If you only see 4 paths, something is wrong. Maybe a zoning error or an error on the EVA port or the Linux driver could not properly talk to the EVA.

You need to choose "Linux" for the Host OS and, if I remember correctly, select Path-A or "Path-B failover only" on the virtual disk presentation tab.

[unre4l]

Well it happened to be an error on the port on the switch. Thanks for all the replies.

[unre4l]

-