tools/test/bin/stc-cell-repair

#!/bin/bash
#-------------------------------------------------------------------------------
# Utility to repair cell slave by destroying all cell-related containers and
# by cancelling all cell reservations hosted by that slave. Intended for use
# after the cell slave has been rebooted as a result of some H/W or OS issues.
#-------------------------------------------------------------------------------

slave=$1

[ -z "$slave" -o "$slave" = "-h" -o "$slave" = "-?" ] \
    && echo "usage: $(basename $0) slaveIp" && exit 1

# Destroy all cell-related containers on the specified slave; do not touch
# the base-* template images though.
ssh sdn@$slave "
    containers=\$(sudo lxc-ls --fancy | grep -E '[a-z]*-[0-9cn]' | cut -d\  -f1)
    for c in \$containers; do
        sudo lxc-stop -n \$c
        sudo lxc-destroy -n \$c
    done
"
# Cancel all reservations hosted on the specified slave
# Also, kill any warden spawned processes
ssh sdn@$CELL_WARDEN "
    cells=\$(grep -l $slave warden/cells/supported/* | xargs -n1 basename)
    for c in \$cells; do
        [ -f warden/cells/reserved/\$c ] && rm warden/cells/reserved/\$c
    done
    pids=\$(ps -ef | grep warden | grep -Ev 'grep|java' | cut -c10-15)
    [ -n \"\$pids\" ] && kill -9 \$pids
"