vinum raid 5

[Brian Szymanski]

>>>>>> "Brian" == Brian Szymanski <ski at indymedia.org> writes:
>
> Brian> Actually I experienced a number of bugs with vinum in 5.3 that
> Brian> proved fatal to a root vinum install (in fact, everything on
> Brian> the second ATA channel was marked down after every reboot if I
> Brian> recall correctly).
>
> I had some problems with a vinum machine (still back at 5.1)
> recently.  I removed a pair of drives and the remaining setup wouldn't
> come back up.  Turns out that multiple incantations of 'vinum
> setdaemon 0' and 'vinum saveconfig' finally fixed things.
>
> vinum is fragile.

Hmm, the vinum setdaemon 0 ; vinum saveconfig ; trick didn't work for me.
If anyone else finds themself in a similar situation and feels like living
dangerously, I've implemented a solution with the below shell script, to
be called from /etc/rc.local. Use at your own risk.

First more on the symptom: on reboot, vinum looses all config information.
You can load the vinum.cfg again, but this puts every subdisk in the Raid5
in the empty state (when in fact they should usually be in the up state).
You can manually set these subdisks to be up, and find all your
information as it was, a nice "feature".

The solution: write a crappy little shell script to remember vinum's Raid5
config information, and set its state accordingly on boot before
mounting/fscking, etc.

If anyone uses this, I'd be interested to hear their results. It works for
me with drive power pulls. Right now it depends on state being recorded at
boot time, so if vinum thinks a subdisk is down (and thus out of date),
then you reboot and the drive comes back to life for a moment, you're in
deep doo-doo. That is to say: flaky drives may introduce errors onto the
R5, but a drive that fails once and then stays down forever will be fine
with this script (this is the usual case from what I've seen, but I
haven't seen everything)... Also I'm not sure why bgfsck doesn't catch the
filesystem automatically, so I used atd to schedule a job to run at +7
minutes with priority 10 to do the background fsck. The idea here being
that if bgfsck does for some reason catch your Raid5, this fsck should lag
behind and not cause any race condition problems. A future improvement
would be to call the state saving portion of the script in rc.shutdown and
periodically from cron. There are many others, but it "works for me" (tm).

Of course the right solution is to fix {g}vinum so this hideousness isn't
necessary...

Cheers,
Brian Szymanski
ski at indymedia.org

step 1:
in /etc/rc.local, add:
sh /etc/rc.vinum.raid5

step 2:
create /etc/rc.vinum.raid5:
#!/bin/sh

### configuration
# ie, mount /dev/vinum/$VINUM_NAME $VINUM_MOUNTPT
VINUM_MOUNTPT="/home"
   VINUM_NAME="big"
# location of the configuration file with just the Raid5 info
    VINUM_CFG="/etc/vinum.big.cfg"
# email address to send notifications to
  ADMIN_EMAIL="ski at wjb"
# place to store state
 DEGRADE_FILE="/vinum.degraded"
# if n=# subdisks, SUBDISKS=0..n-1, e.g. the below works for a 4 drive raid5
     SUBDISKS="0 1 2 3"
### no config below this point

#if already mounted, abort now
mount | grep -q "^/dev/vinum/$VINUM_NAME on $VINUM_MOUNTPT" && exit 0

# mount R5 device
# (why on earth doesn't vinum info persist across reboot?!?)
/sbin/vinum create -f "$VINUM_CFG"

if [ -e "$DEGRADE_FILE" ] ; then
        # start subdisks based on content of DEGRADE_FILE
        for i in $SUBDISKS ; do
                #start all non-degraded subdisks
                grep -q $VINUM_NAME.p0.s${i} "$DEGRADE_FILE" && \
                        echo $VINUM_NAME.p0.s$i down || \
                                vinum start $VINUM_NAME.p0.s${i}
        done
        #this maintains degraded state even if new drive is inserted
        #this is important or else we will lose major information
else
        /sbin/vinum start $VINUM_NAME.p0
fi

#let vinum catch up...
sleep 1

#check the state, mail $ADMIN_EMAIL if things are horked...
state=`/sbin/vinum l | grep "^P.$VINUM_NAME\.p0" | awk '{ print $5 }'`
echo state: $state
if [ "z$state" = zup ] ; then
#       mount /dev/vinum/$VINUM_NAME "$VINUM_MOUNTPT"

        #fsck and mount
        fsck -F -p "/dev/vinum/$VINUM_NAME"
        mount "/dev/vinum/$VINUM_NAME" "$VINUM_MOUNTPT"
        #bgfsck should do this automagically, but it doesn't?
        at "+7 minutes" <<END
nice -n 10 fsck -B -p "/dev/vinum/$VINUM_NAME"
END
elif [ "z$state" = zdegraded ] ; then
        echo "VINUM DEGRADED!!! check $ADMIN_EMAIL's email..."
        mail -s "VINUM DEGRADED!!!" "$ADMIN_EMAIL" <<END
vinum is in degraded state! buy a new drive and replace the faulty one!
END
        #let the mail go out, something later in startup horks things...
        sleep 15
        /sbin/vinum l | grep "^S.$VINUM_NAME\.p0\.s" | \
                grep -v 'State:.up' >>"$DEGRADE_FILE"

        #fsck and mount
        fsck -F -p "/dev/vinum/$VINUM_NAME"
        mount "/dev/vinum/$VINUM_NAME" "$VINUM_MOUNTPT"
        #bgfsck should do this automagically, but it doesn't?
        at "+7 minutes" <<END
nice -n 10 fsck -B -p "/dev/vinum/$VINUM_NAME"
END

else
        echo "VINUM BROKEN!!! check $ADMIN_EMAIL's email..."
        mail -s "VINUM BROKEN!!!" "$ADMIN_EMAIL" <<END
vinum is in an unknown state: $state. look at what's going on!
END
        #let the mail go out, something later in startup horks things...
        sleep 15
fi

echo done.