The Agile View and the renaming of disk DSFs - success ?

[Francis Noël]

Hello ITRC Forums

I have been poking around in the Agile View for a few days and I pretty much 'get' the legacy path, LUN path and persistent DSF jibba-jabba.

scsimgr is nice too but one feature that I was expecting from it was the ability to arbitrarily rename the persistent DSF's in order, for example, to sync up the device names on hosts that are part of a RAC cluster.

I was a bit surprised to discover that this functionality is available thanks to an enhancement (is this even documented?) to ioinit rather than a scsimgr command. It does make sense since it seems that the device name is directly bound to the instance number.

It was actually pretty easy to do....too easy.. I fear I might be doing something wrong and/or dangerous/risky.

The renaming of disk DSF's seems to be a recurring topic on the forums and so far all the posts I have found seem to say that it is doable but those threads rarely have a conclusion.

Please read the attached screen dump to see how I did it. It is so simple i fear this just cant be right ...

Note that the DSF rename was effective immediately, no reboot required, as demonstrated by the diskinfo command.

I rebooted the host after setting this up and the new name survived the reboot.

Running rmsf -H on the lunpath followed by an insf -e restores the original DSF.

All comments are welcome.

[melvyn burnard]

Maybe you shoul dtake a read of
http://docs.hp.com/en/dsfmigration/persistent_dsf_migration.pdf
and also visit
http://docs.hp.com/en/LVMmigration1/LVM_Migration_to_Agile.pdf

[Francis Noël]

Hello Melvyn and thank you for your input.

I did go through those documents a few weeks back and they are pretty good at presenting the technology but I'm pretty sure that renaming DSF's was not among the topics covered.

io_redirec_dsf is in there and looks similar but it is only usable in a context where a disk is replaced or moved as it expects the old and new DSF to exist.

Would the correct way of doing this imply the manual creation of a device node with the desired minor number folowed by a call to io_redirect_dsf rather than using ioinit ? It is a possibility but I have not explored it.

[Ron Lawson_1]

You are correct.

ioinit -f is the right method to do this.

[Francis Noël]

Thanks Ron your answer is very encouraging.

Do you know if modifying only the persistent DSF on the LUNpath, without rebooting, is supported/allowed by HP or am I taking chances here ?

I.E. is this method viable in a production environment ?

[Duncan Edmonstone]

Francis,

Well the man page:

http://docs.hp.com/en/B2355-60130/ioinit.1M.html

Wouldn't give an example of exactly this if it wasn't supported!

Note that you can't reassign more than 14 devices at a time using this method without a reboot.

HTH

Duncan

[Francis Noël]

Hello Duncan and thank you for joining in.

I agree with you completely but I saw so many threads stating that you had to reboot (ioinit -r) or just plain saying it could not be done that it just looked to simple, and too good, to be true.

Do you happen to know if the 14 device limit is cummulative ? I.E. if after modifying 14 devices, one-at-a-time, one would be forced to reboot once device 15 is reached ?

Sorry for all the questions guys but this is a great new functionality and a big selling point for v3 in our organization, gotta make sure I get it right.

[Duncan Edmonstone · Accepted solution]

Francis,

The internal documenets I've seen on this indicate 14 entries *per invocation*

However, a quick dig on the patch database turned up a couple of patches that seem to remove even this limitation.

See PHCO_37481 and PHKL_37459

HTH

Duncan

[Duncan Edmonstone]

Incidentally, any references to "can't be done" or "requires reboot" are talking about 11iv2 and 11iv1 systems. The IO stack on 11iv3 is pretty much a complete rewrite. It can be done and its pretty reliable too.

HTH

Duncan

[Francis Noël]

Thank you Duncan
10 points and many Internets for you sir.

That pretty much des it for me, I'll leave the thread open until quitting time just in case someone has anything to add.

[Francis Noël]

Looks like I'm all set.

Pretty good thread here, hope this helps others use the feature.