[dm-devel] dm-emc: failback does not seem to work correctly

[egoggin at emc.com]

On Tuesday, May 02, 2006 8:33 AM, Eddie Williams wrote:
> 
> Is this behavior with inactive LUNs a problem with my setup or is the
> expected behavior with the current state of affairs with SLES 9 SP3?
> While this is not causing an observable failure, aka all IO's 
> and tests
> are running fine, the noise level in the log file is going to be one
> that customers are probably not going to like.
>

This is a benign but noisy nuisance.  It is a result of older versions
of the multipathd/main.c:need_switch_pathgroup() requiring that
the active path group be in an active state in the kernel while the
lazy activation of a path group by dm-mpath.c in the kernel prevents
this from the designated highest priority path group from being
initialized (and therefore being viewed as in an active state) until
the first io is dispatched.

The problem is fixed in upstream multipathd code.

> Eddie
> On Fri, 2006-04-28 at 17:29 -0400, Eddie Williams wrote:
> > I am testing SLES 9 SP3 with an EMC CLARiiON array (CX300). 
>  When I pull
> > cables (on the storage side of the switches) the IO's are 
> successfully
> > switched to the good paths (all the way down to a single good path).
> > However when I replace the cables the switchback does not 
> seem to work
> > successfully.  Well, for the active devices, e.g. where 
> there are IO's
> > being issued, everything seems to work fine.  For devices 
> that are not
> > being used there lots of messages (every loop of multipathd).
> > 
> > I see the message:
> > multipathd: 360060160cfd0150045435fc76f20da11: switch to 
> path group #1
> > 
> > and then
> > device-mapper: dm-emc: emc_pg_init: sending switch-over command
> > 
> > Then it repeats the switch to path group messages (each 
> iteration of the
> > multipathd daemon).  This seems to repeat indefinitely.  
> There are no
> > other messages that would indicate a failure or problem.  
> > 
> > The output from multipath -ll shows everything is now active/ready:
> > 360060160cfd0150045435fc76f20da11
> > [size=5 GB][features="1 queue_if_no_path"][hwhandler="1 emc"]
> > \_ round-robin 0 [prio=2][enabled]
> >  \_ 2:0:1:16 sdbz 68:208 [active][ready]
> >  \_ 1:0:0:16 sdr  65:16  [active][ready]
> > \_ round-robin 0 [enabled]
> >  \_ 1:0:1:16 sdal 66:80  [active][ready]
> >  \_ 2:0:0:16 sdbf 67:144 [active][ready]
> > 
> > The multipath.conf entry is:
> >   device {
> >           vendor                  "DGC"
> >           product                 "*"
> >           path_grouping_policy    group_by_prio
> >           getuid_callout          "/sbin/scsi_id -g -u -s /block/%n"
> >           prio_callout            "/sbin/mpath_prio_emc /dev/%n"
> >           hardware_handler        "1 emc"
> >           features                "1 queue_if_no_path"
> >           path_checker            emc_clariion
> >           failback                immediate
> >   }
> > 
> > If I startup IO to the luns that are being complained about 
> the messages
> > stop.  It does not seem the switchover is performed unless 
> there is IO
> > happening at the time the switchover is attempted.