Have connected a non-IBM storage device to AIX host via fibre channel. If the storage is rebooted or a raid controller fails over whilst connected to the host, the paths that drop do not come back online when the ports become active again.
I have tried enabling dynamic tracking and delayed_fail on the FC adapter and have played with numerous timeout setting but to no avail. AIX does not recognise when a failed path is available again unless I either reboot the host or initiate cfgmgr from the command line to force the FC adapter to rescan.
What needs to be changed to allow AIX to automatically recognise when a dead path is back online?
I am not sure what you mean by "I have tried enabling [...] delayed_fail on the FC adapter". I'd try enable fast_fail of the fscsi device. Put all children of all fscsi devices in the defined state
# rmdev -l fscsiX -R
change the policy from delayed_fail (i.e. off) to fast_fail (i.e. on) on every fscsi device
# chdev -l fscsiX -a fc_err_recov=fast_fail
Run cfgmgr to reconfigure all devices back to the available state.
# cfgmgr
You may also want to change the "dyntrk" to "yes" on the scsi device. I've changed the "fc_err_recov" (like shockneck suggested) and the dyntrk to yes and I've done a few tests where while writing to the disks, I would unplug one of the cards and the writing would continue, then I would plug it back in about two seconds later, unplug the other and the writing still continued. You can run this command to save the attribute changes to the ODM:
chdev -l fscsi0 -a dyntrk=yes -a fc_err_recov=fast_fail �P
Then you just need to remove the devices like shockneck said and use cfgmgr to bring them back.
It's a Nexsan Satabeast and the FC adapter is an IBM 1910
I have tried various configurations of the fscsi device with dynamic tracking enabled/disabled and fast_fail/delayed_fail but nothing so far has recognised when a failed path becomes active again.
I know multipathing does not work correctly without a device PCM so I'm wondering if this may be the same issue.