Schedule a scrub once the resilver is finished

[FransUrbo]

I'd like the option to schedule a scrub (zpool scrub [pool]) which should run as soon as the current resilver is finished. As it is now, it (the scrub request) will fail and I have to wait at the console until the resilver is done until I can run the scrub.

# zpool scrub share
cannot scrub share: currently resilvering
# zpool status
  pool: share
 state: ONLINE
status: One or more devices is currently being resilvered.  The pool will
        continue to function, possibly in a degraded state.
action: Wait for the resilver to complete.
  scan: resilver in progress since Wed May 18 11:11:32 2016
    130M scanned out of 29.2T at 9.30M/s, (scan is slow, no estimated time)
    5.69M resilvered, 0.00% done

This would probably be something that ZED should deal with.

[Lilolbet]

No it doesn't. Say you have two mirrors of two drives. You lose one drive in one of the mirrors and replace it. It resilvers from it's mirror companion, not touching the other drives in the pool.

[DeHackEd]

A resilver does do a metadata walk of the whole pool, but while a scrub will read from all disks (parity, mirrors and copies/ditto blocks) a resilver will only do so when it involves a disk marked as (resilvering). File data on non-resilvering disks is completely untouched except where needed to rebuild the resilvering disks.

[GregorKopka]

Wouldn't be a mode in which the resilver touches everything be better?

[rlaager]

Not necessarily. If you have a resilver that is half done, it'd still be better to let it finish before starting the scrub. The behavior requested here would be especially nice for the default system-wide scrubbing script (if your distro has such a thing).

[kingneutron]

--Why wait at the console? My philosophy is "Let the machine do the heavy lifting." ;-)

--You can accomplish what you are asking with a shell script (recommend running this as root, in a GNU "screen" session - and don't forget to chmod +x it):

BEGIN partial code for waitresilver-n-scrub.sh

#!/bin/bash

zp=$1 # Pass arg to script

    printf `date +%H:%M:%S`" ...waiting for $zp resilver to complete..."
    waitresilver=1
    while [ $waitresilver -gt 0 ];do 
      waitresilver=`zpool status -v $zp |grep -c resilvering`
      sleep 5
    done 
    echo  `date +%H:%M:%S`' - DONE - Syncing to be sure'; time sync;

do your next thing after this line

zpool scrub $zp

--Man, markup on this site is $bad, handy file attached ;-)
waitresilver.zip

[behlendorf]

@tonyhutter this might be a nice configurable way to handle some of drive replacement issues we talked about regarding when a pool should be automatically scrubbed.

[tonyhutter]

Yep, this would be very nice. You could kick off a zedlet on a resilver finish event to do it. Below is a completely untested script, using resilver_finish-notify.sh as inspiration. If I wrote it correctly, you should just be able to paste it in cmd/zed/zed.d/resilver_finish-start-scrub.sh and it will work:

# resilver_finish-start-scrub.sh
# Run a scrub after a resilver
#
# Exit codes:
# 9: internal error
#
[ -f "${ZED_ZEDLET_DIR}/zed.rc" ] && . "${ZED_ZEDLET_DIR}/zed.rc"
. "${ZED_ZEDLET_DIR}/zed-functions.sh"

[ -n "${ZEVENT_POOL}" ] || exit 9
[ -n "${ZEVENT_SUBCLASS}" ] || exit 9

zed_check_cmd "${ZPOOL}" || exit 9

if   [ "${ZEVENT_SUBCLASS}" = "resilver_finish" ]; then
    ${ZPOOL} scrub ${ZEVENT_POOL}
fi

[pashford]

If I understand the dRAID information I've been reading, this feature would be very important for pools containing dRAID VDEVs, as a dRAID resilver doesn't do checksum validation of the source data.

Has the above script from @tonyhutter been tested yet?

[behlendorf]

@pashford yes we're going to want exactly this for dRAID.

@tonyhutter what do you think about opening a PR which adds this script, then create a new test case under tests/zfs-tests/tests/functional/fault/ which verifies it works as intended.

[tonyhutter]

@behlendorf good idea, I'm on it.