EMC CX400 and powerpath software on HP-UX 11.00

[THEUNISSEN, J]

I am using an EMC CX400 RAID system (with powerpath software) on a HP-UX11.00 MC/SG environment. This is working just fine, but I would like to test the functionality of the MC/SG as well. So I disconnect both fibercables (we have two to have a primary and secondary diskpath via a switch) on one of the HP9000 systems and I thought this would cause a switchover to the other system. The command "powermt display dev=all" gives me all paths are "dead", but the mount was still there (as I saw with the commands bdf and mount). Of cource I could not write anything to the mounted volumegroup anymore, even worse this session hung and I couldn't kill this process either. The only thing I could do was to poweroff the system with the disconnected fibercables.

My question is: is powerpath not working properly in this case (in other words, HP-UX doesn't know the paths aren't there anymore?). What will cause a switchover then? Should we install extra patches on HP-UX in order to be sure this will work?

[Jeff Schussele]

Hi,

What MC/SG version are you running?

Jeff

[david5337_1]

Power path can onlu provide the path transfer,but can't transfer application on server.
So if you only disconnect one fibre cable,other patch can work fine.
Now he disconnect one server all path,but the CLuster software didn't transfer application.
You need to check the Cluster sofware setting.
David

[Bernhard Mueller · Accepted solution]

Hi,

this is not a fault of PowerPath (which is simply geared to provide load balancing), nor is it a MC/SG fault (because it is not a *single* failure, if you pull *two* FC cables...)

If you have *no* path to a disk anymore, any IO to it cannot be processed. Therefore your DB or application will just hang and nothing happens.

So if you think you have to make sure a failover occurs in that situation (e.g. if you use MirrorDisk to mirror your lvols from one EMC to another and want to make sure you can survive the total breakdown of one of the EMCs), you need to add a service to your package configuration and control scripts that monitors disk accessability -e.g. through ioscan- and TOCs that machine, in case the monitor indicates the failure. The TOC is initiated by setting FAIL_FAST_ENABLED to YES. Note you *have* to do it this way and force a TOC, because otherwise the machine would forever try to complete IOs to the disks which are no longer accessible.

Regards,
Bernhard

[THEUNISSEN, J]

The ioscan still gives me CLAIMED (instead of NO_HW), so a test with ioscan will not provide me an error code I guess. Would a touch command give me the failover I need?

[Bernhard Mueller]

Hi,

you need to run e.g.
ioscan -fnCdisk

do *not* use -k option.

if you have no path to a disk, ioscan will eventually show NO_HW, however if you just ask what the kernel "thinks" there is, it may still be CLAIMED.

Regards,
Bernhard

[Tim Sanko]

OK Here are a few Statements that are true.

1. You are on a SAN. not direct connected. If you'd have been direct connected the EMC would have been generating an EMC call home.

2. You only have two paths to the symm defined.

3. You may have been able to just replace the fc connections and run symcfg discover and been fat dumb and happy.

Tim

[THEUNISSEN, J]

The suggestion of Bernard was the solution for this problem (FAIL_FAST_ENABLED set to YES will invoke a TOC and the cluster will take over the package from the host that does not have any I/O paths). Thanks.

[THEUNISSEN, J]

I found the solution