Run riak in a docker container for CI

[arcusfelis]

This PR allows to run Riak in a container.

Also adds:

• Adds healthcheck for Docker MySQL container.

See previous work:

• SSL for Riak driver #1564 - SSL for Riak driver

[kzemek]

Is env is more likely to be in /usr/bin than bash?

[arcusfelis]

@kzemek we use /usr/bin/env in many scripts, it's just another way to do the same thing.
Never seen a system without /bin/bash or without /usr/bin/env.

[kzemek]

/usr/bin/env <sth> is a trick to find <sth> in caller's $PATH.
I'm not saying that the change is bad. I'm not saying it's good either.

[arcusfelis]

Feeling the same.
But there was some conversation on hipchat about using env everywhere. It was more about "one way to do this", and env was chosen.
Maybe we should document it.

[kzemek]

You're really going overboard with the comments here.

[kzemek]

What if instead of doing create + fileswap + start trick we did the following?:

RIAK_CONF=$(mktemp)
docker run --rm --entrypoint cat michalwski/docker-riak:1.0.6 /etc/riak/riak.conf \
    sed 's/^search = \(.*\)/search = on' \
    > "${RIAK_CONF}"
cat "${DB_CONF_DIR}/riak.conf.ssl" >> "${RIAK_CONF}"
docker create -p 8087:8087 -p 8098:8098 \
    -e DOCKER_RIAK_BACKEND=leveldb \
    -e DOCKER_RIAK_CLUSTER_SIZE=1 \
    --name=mongooseim-riak \
    -v "${DB_CONF_DIR}/advanced.config:/etc/riak/advanced.config:ro" \
    -v "${SSLDIR}/fake_cert.pem:/etc/riak/cert.pem:ro" \
    -v "${SSLDIR}/fake_key.pem:/etc/riak/key.pem:ro" \
    -v "${SSLDIR}/ca/cacert.pem:/etc/riak/ca/cacertfile.pem:ro" \
    -v "${RIAK_CONF}:/etc/riak/riak.conf:ro"
    --health-cmd='riak-admin status' \
    "michalwski/docker-riak:1.0.6"

Thoughts?

[arcusfelis]

Two times container creation in your case vs one time container creation in mine.
I want to measure both, but yours version would crash.
Because of the reason, described in a comment here in the code:

+    # We can't use volumes for riak.conf, because:
+    # - we want to change it
+    # - container starting code runs sed on it and gets IO error,
+    #   if it's a volume (THIS)

[kzemek]

Two times container creation in your case vs one time container creation in mine.

I don't think this is a problem

I want to measure both, but yours version would crash.

Yep, that's a problem

[kzemek]

Why go through the complications of setting up a separate timeout process? We can achieve less fidelity but much more clarity with simply manipulating the loop:

for i in $(seq ${TIMEOUT}); do
    if [ $(health_status "$CONTAINER")"" = "\"healthy\"" ]; then
        echo "\nWaiting is done"
        exit 0
    fi
    echo -n "."
    sleep 1
done
echo "\nKilled by timeout"
exit 1

[arcusfelis]

ok, I like your code more.
We can use also timeout utility.

[kzemek]

There are two main problems with timeout:

• still requires trapping exits, otherwise SIGINT from user will be lost
• not present on macOS unless you install coreutils (and then present as gtimeout

[arcusfelis]

echo -e "\n"
heh, it's tricky

[arcusfelis]

Will not kill or print progress, if docker call does not return. But we can live with it I guess.

[kzemek]

I definitely can.
It's also possible that kill wouldn't kill the process if it hanged in docker call, anyway.

[arcusfelis]

OK, rewrote using cycle.
blocking docker is super rare, I would care, if it would cause problems for me.
Until then it should be enough.

[michalwski]

Looks good to me.

I like the idea with docker health checks.