CiaB bug: shutdown-instances.sh fails

[garloff]

The shutdown-instances.sh script does not do its job, as the loop does not loop in
for domain in "$DOMAINS"; do (l.25) as it treats all domains as one string.
See #262 (comment)

[garloff]

#!/usr/bin/env bash
# (c) Kurt Garloff <s7n@garloff.de>, 2/2024 - 6/2025
# SPDX-License-Identifier: Apache-2.0

set -ex

CONTAINER=nova_libvirt

# Send SysRq-$1 to all domains
# Returns the number of domains treated
# BEWARE: Do not treat non-zero return as error (set -e)
# So always use `send_sysrq S || true` or `send_sysrq S || sleep 1`
send_sysrq()
{
        declare -i ctr=0
        DOMAINS=$(docker exec $CONTAINER virsh list --name | tr -d '\r')
        for dom in $DOMAINS; do
                docker exec $CONTAINER virsh send-key $dom KEY_LEFTALT KEY_SYSRQ KEY_$1 || continue
                let ctr+=1
        done
        return $ctr
}

# Wait for libvirt reporting all domains are gone
# $1: max waiting time (in s)
# Returns number of still running domains
wait_for_shutdown()
{
        set +x
        declare -i ctr=0
        MAXWAIT=$1
        shift
        while [ $ctr -lt $MAXWAIT ]; do
                declare -i running=0
                DOMAINS=$(docker exec $CONTAINER virsh list --name | tr -d '\r')
                for dom in $DOMAINS; do
                        domstate=$(docker exec $CONTAINER virsh domstate $dom 2>/dev/null | tr -d '\r' || true)
                        if [ "$domstate" != "shut off" ]; then let running+=1; fi
                done
                if [ $running = 0 ]; then break; fi
                let ctr+=1
                sleep 1
        done
        echo "Still running after $ctr seconds: " $DOMAINS
        set -x
        return $running
}

# Do virsh action $1 (shutdown, destroy, ...)
virsh_action()
{
        DOMAINS=$(docker exec $CONTAINER virsh list --name | tr -d '\r')
        for dom in $DOMAINS; do
                docker exec $CONTAINER virsh $1 $dom 2>/dev/null || true
        done
}

# Overall procedure should be < 60s
# Within that time, try to achieve maximum chance for data to survive
# Sync FS: Give large instances maximum time to have date written back
send_sysrq S || true
# Preferred method: proper shutdown (via qemu-agent)
virsh_action shutdown
# wait max 42s for proper shutdown to succeed on all instances
wait_for_shutdown 42 || true
# Remaining instances: Send a termination signal to apps (they may write back data)
send_sysrq E || sleep 1
# Extra Sync (in case later 'U' is not allowed)
send_sysrq S || sleep 1
# Emergency read-only FS (flushes pending writes also)
send_sysrq U || sleep 2
# And power off (give guest kernel a chance to return pass-through HW into a good state)
send_sysrq O || sleep 1
# Last action: Stop emulation (destroy anything that's remaining)
virsh_action destroy

[garloff]

Happy to submit a PR ...

[jklare]

Hi @garloff , I am currently going through the open issues in our repositories and found this one. Since you linked another issue and that one was closed, i was wondering if this is still relevant? If so, and since you said you would be happy to submit a PR, could you do that and link this issue in it? Cheers!