[Linux-cluster] dm_multipath and failover

[Matthew Geier]

  Not strictly a clustering question, but

  I have for RHEL CS servers connected to an EMC CX300. QLA2432 fibre 
channel controllers.

  The cards have two FC channels. I've used them to run a redundant 
fabric - there are two fibre channel switches, that each connect to one 
of the EMC's FC ports - so there are two paths to the storage. A link or 
switch can fail, but the hosts still can see the storage.

  The redundancy works fine - I can pull a fibre out and the hosts can 
see the storage fine. Syslog gets spammed with lots of scsi errors, but 
dm_multipath does its stuff and the server keeps running.

  However if I plug that fibre back in, the path doesn't come back. So 
far the only way to get the path back i've found is to reboot the server.
  Is there any way to tell the controller the path is back ?.

  I have noticed that if i'm in this state, rebooting the McData FC 
switch will also restore the channel - it's like the switch sends some 
sort of re-initialise channel signal as it reboots.