Tricks to keep Linux PowerPath devices in sequence

I don't know of any plans to implement emcpadm on AIX at the moment

Hi All, some good news:

Starting with PowerPath 5.3 on AIX the emcpadm utility is now supported, please see emc166911 for full steps on how to use emcpadm utility

power devices will remain persistent, no need to change them after reboot. That's why you want to use power devices in you logical volumes instead of native devices.

Am I correct in understanding that, even using the mapping file, you might have to run the command to rename the devices every time you boot the server? Is there no way to keep this device mapping permanently across reboots?

very strange indeed, your steps look good to me. This is exactly what i do on my Redhat RAC nodes. Is it just this one node that scrambles its power device names upon reboot ?

Dynamox,

Thanks for the quick reply. I have a SUSE10.2 server where the devices are not remaining consistent. It is part of a 5 node Clusterware cluster being built. The disks assigned to it are exactly the same as the other servers in the cluster. I have seen this server rebooted and keep the device names but most times it does not.

I exported the cluster disk names using the emcpadm utility and imported the names just fine. I have also manually reassigned these names using the emcpadm utility. I have removed the /etc/powermt.custom file and then performed a 'powermt save' command to rebuild the file. It looks good but after a reboot on Friday the disks came in as differnt emcpower* names.

Am I missing something?

if it's just this one node, have you tried the "brute force windows" approach of re-installing PowerPath ?

Yes. What's even more kooky is that it doesn't necessarily do it every time. The last time I rearranged the devices it kept over two reboots. Then the DBA setting up the cluster reboots the server and they change...

i know ..don't tell your Windows guys if it fixes your issue.

As a workaround, create your own devices:
#mknod /u02/oradata/css c 162 1
#mknod /u02/oradata/ocr c 162 2
#mknod /u02/oradata/asm/lun1 c 162 3
#mknod /u02/oradata/asm/lun2 c 162 4
#mknod /u02/oradata/asm/lun3 c 162 5
#mknod /u02/oradata/asm/lun4 c 162 6
where the major/minor numbers are the same as the power devices.

A reinstall of PowerPath did not help this issue. The /dev/emcpower* names stayed consistent over one reboot but changed around after a second test reboot. This is also overwriting the /etc/powermt.custom file as well as the information there changes.

Any ideas?

Just wanted to give an update before going home. I rebooted this server 12 different times and only twice did the logical names stay the same over a reboot. I definitely see the /etc/powermt.custom file being changed somehow between the time I issue the shutdown command until the time the server is back up. I have opened a case with EMC about this and we'll see where it goes.

Just for closure's sake...

I worked with EMC techs on this issue for a number of weeks. I installed a number of debug versions of PowerPath to try to capture the issue but we had no luck. Eventually after removing the debug version and reinstalling the original version of PowerPath on the server the disks aren't changing logical device names anymore.

EMC's response is that this is a known but rare issue that seems to have something to do with Oracle or Clusterware and that it can be resolved by uninstalling and reinstalling the software. I will update this thread if/when I learn anything else.