docker-entrypoint-initdb.d is nice, but how about docker-entrypoint-upgradedb.d?

[pluma]

When extending the ArangoDB docker image as a base image to dockerize a Foxx service (or anything else really) it's extremely useful to be able to just stuff /docker-entrypoint-initdb.d full of shell scripts doing things like

foxx install /myapp /myapp.zip -H $ARANGO_INIT_ENDPOINT

to get the database up and started. However this is only useful for a first-time run. Once the database is initialised, changes to these shell scripts or arangosh JS scripts won't have any effect.

If you want to deploy a new version of your Foxx service or run some migrations, there's no way to d that automatically as part of upgrading to a new version of your image (e.g. using docker-compose).

I ended up having to wrap the existing entrypoint script with my own like so:

#!/bin/sh
set -e
# This is the opposite of the check in entrypoint.sh:
# i.e. only proceed if the DB was already initialised
if [ -f /var/lib/arangodb3/SERVER ]; then
    export ARANGO_INIT_ENDPOINT=tcp://127.0.0.1:8999

    # Upgrade ArangoDB if the base image changed to a new semver minor
    # See #72 for thoughts on this
    arangod --config /etc/arangodb3/arangod.conf \
            --server.endpoint $ARANGO_INIT_ENDPOINT \
            --database.auto-upgrade true

    # Start up the DB in init mode, again
    arangod --config /etc/arangodb3/arangod.conf \
            --server.endpoint $ARANGO_INIT_ENDPOINT \
            --server.authentication false \
            --log.file /tmp/init-log \
		    --log.foreground-tty false &
    pid="$!"

    counter=0
    ARANGO_UP=0

    while [ "$ARANGO_UP" = "0" ]; do
        if [ $counter -gt 0 ]; then
            sleep 1
        fi

        if [ "$counter" -gt 100 ]; then
            echo "ArangoDB didn't start correctly during init"
            cat /tmp/init-log
            exit 1
        fi
        let counter=counter+1
        ARANGO_UP=1
            arangosh \
                --server.endpoint=$ARANGO_INIT_ENDPOINT \
                --server.authentication false \
                --javascript.execute-string "db._version()" \
                > /dev/null 2>&1 || ARANGO_UP=0
    done

    # Now we can upgrade the service
    foxx upgrade /myapp /myapp.zip -H $ARANGO_INIT_ENDPOINT

    # Wind the db down again
    if ! kill -s TERM "$pid" || ! wait "$pid"; then
        echo >&2 'ArangoDB Init failed.'
        exit 1
    fi
fi

# Proceed normally with the original entrypoint.sh of the base image
. /arango-entrypoint.sh $@

This is tedious and unwieldy. But I think we could generalise this.

Why not extend the entrypoint.sh so that on non-initial runs (i.e. if the SERVER file already exists) we check if a special directory like /docker-entrypoint-upgradedb.d (or whatever) is non-empty, and then just do what we're doing for the files in docker-entrypoint-initdb.d?

This would mean users no longer have to mess with the entrypoint and could just create a dockerfile like this for dockerizing a single Foxx service:

FROM arangodb:3.6.1

COPY myapp.zip /myapp.zip

RUN echo 'foxx install /myapp /myapp.zip -H http://127.0.0.1:$ARANGO_INIT_PORT' > /docker-entrypoint-initdb.d/install-myapp.sh

RUN echo 'foxx upgrade /myapp /myapp.zip -H http://127.0.0.1:$ARANGO_INIT_PORT' > /docker-entrypoint-upgradedb.d/upgrade-myapp.sh

And it would "just work" (TM).

[pluma]

/cc @joerg84

This is a papercut I ran into while using docker with Foxx myself. The workaround is clunky and fragile. I don't think it would be too complicated to extend the entrypoint.sh to simplify users' dockerfiles to what I show in the end with no other dependencies.

[michlschmid]

Adding the --database.auto-upgrade true option to the entrypoint script would be really helpful if you're interested in keeping your engine up-to-date and want to keep your existing database working in an easy way.

Thumbs up for this!

[ambs]

I am been bitten by this. And my lack of knowledge makes it harder to know the easiest way to run a upgrade...