Waiting for MySQL to be Ready in Docker Compose

Posted on by Christopher Davis

Had a … fun thing happen in GitHub Actions today where my tests were starting before the MySQL server they needed was up and running. MySQL in this case was running in Docker compose. So I started my process with a little shell script using mysqladmin:

#!/usr/bin/env bash

count=0
alive="no"
while [ "$count" -lt 5 ]; do
    docker-compose exec -T mysql mysqladmin ping --silent > /dev/null
    if [ $? -eq 0 ]; then
        alive="yes"
        break
    fi

    echo "waiting for mysql"
    sleep 5
    count=$((count+1))
done

if [ "$alive" = "yes" ]; then
    exit 0
else
    echo "MySQL did not start up in time"
    exit 1
fi

But for some reason this wait_for_mysql script would work, but the server would still go away. I thought it might be some sort of issue with docker itself, so I switched to using a healthcheck via docker compose:

mysql:
    image: mysql:8
    # ...
    healthcheck:
      test: ["CMD", "mysqladmin", "ping", "--silent"]
      retries: 1

And checking the container status in my wait_for_mysql script:

#!/usr/bin/env bash

count=0
alive="no"

while [ "$count" -lt 6 ]; do
    health=$(docker inspect --format "{{.State.Health.Status}}" "$(docker-compose ps -q mysql)")

    if [ "$health" == "healthy" ]; then
        alive="yes"
        break
    fi

    echo "waiting for mysql: $health"
    sleep 10
    count=$((count+1))
done

if [ "$alive" = "yes" ]; then
    exit 0
else
    echo "MySQL did not start up in time"
    exit 1
fi

I like this quite a bit better, honestly, since it accounts for the start up of the container itself.

However, the same issue kept up: either I’d get a MySQL server has gone away error or my initial setup creating the database would fail, despite the health check succeeding.

It turns out that MySQL server after its first start up goes down again before coming back up again. On a CI server, this happens every time. I ended up changing my healthcheck command to a simple mysql call with -e flag and this seemed to catch everything only when it was truly up.

  mysql:
    image: mysql:8
    # ...
    healthcheck:
      test: ["CMD", "mysql", "-u", "root", "-e", "USE changme_to_your_database_name;"]
      retries: 1