[dm-devel] dm-multipath fails when 1 path is taken offline

[Jason Price]

I've apparently got multipath configured 'correctly', however it seems to
favor one path over the other.  I can tell this by watching the switch
traffic... traffic only flows down one path.  When I take the 'non-prefered'
path offline the logs notice, and take down the path.  I can bring that path
back online and the path reconnects.

However, when I take the 'preferred' path offline, the volume shows
Input/Output errors, and forces the volume into read-only mode.

Background:
Running v0.4.9 of device-mapper-multipath
Hitachi AMS2500 array, using ports 0F and 1F
Brocade 5300 switches, split into two fabrics.  (fabric A, port 30,
fabric B port 30)
Host has a qlogic 2462 card with two ports in use.

Current multipath.conf:

blacklist {
       devnode "^sda$"
}

defaults {
               checker_timeout         5
               polling_interval        5
}

multipaths {
       multipath {
               wwid 360060e8010053b90052fb06900000190
               alias                   vrp
#               path_selector           "round-robin 0"
       }
}

devices {
        device {
                vendor                  "HITACHI"
                product                 "DF.*"
                getuid_callout          "/sbin/scsi_id -g -u -s /block/%n"
                features                "1 queue_if_no_path"
                hardware_handler        "0"
                path_selector           "round-robin 0"
#               path_grouping_policy    group_by_prio
                path_grouping_policy    multipath
# Hitachi recommendation.
                failback                immediate
                rr_weight               uniform
                rr_min_io               1000
                path_checker            tur
                prio                    hds
        }
}

results of 'multipath -ll'

# multipath -ll
vrp (360060e8010053b90052fb06900000190) dm-8 HITACHI,DF600F
size=70G features='1 queue_if_no_path' hwhandler='0' wp=rw
|-+- policy='round-robin 0' prio=1 status=enabled
| `- 6:0:0:0 sdc 8:32 active ready running
`-+- policy='round-robin 0' prio=0 status=enabled
 `- 5:0:0:0 sdb 8:16 active ready running

I've also worked with several revisions of the multipath.conf file.  If I
remember correctly, with some device stanza revisions, I've had multipath -ll
returning this result instead:

# multipath -ll
vrp (360060e8010053b90052fb06900000190) dm-8 HITACHI,DF600F
size=70G features='0' hwhandler='0' wp=rw
`-+- policy='round-robin 0' prio=1 status=active
|- 6:0:0:0 sdc 8:32 active ready running
`- 5:0:0:0 sdb 8:16 active ready running

However, both are affected by the same problem.

Here's what multipath -ll shows in a path failure:

[root at prtdb02 ~]# multipath -ll
May 26 18:50:12 | sdc: hds prio: SCSI error
vrp (360060e8010053b90052fb06900000190) dm-8 HITACHI,DF600F
size=70G features='1 queue_if_no_path' hwhandler='0' wp=rw
|-+- policy='round-robin 0' prio=0 status=active
| `- 6:0:0:0 sdc 8:32 failed faulty running
`-+- policy='round-robin 0' prio=0 status=enabled
  `- 5:0:0:0 sdb 8:16 active ready  running

Excerpt from /etc/fstab:
LABEL=vrp-db            /vrp-db                 ext3    defaults        0 2

relevant line from pvscan:
 PV /dev/mpath/vrp   VG vrpdg     lvm2 [70.00 GB / 0    free]

relevant line from vgscan:
 Found volume group "vrpdg" using metadata type lvm2

While I'd prefer an 'active-active' setup, I'd accept an active/passive
setup, provided it failed over correctly... preferably with a fast failback.

I'm more than happy to provide any other information.

--Jason