Chaining the command or using and entrypoint is less optimal in case you want to horizontally scale your application.
Then all replicas will do the migration, at the same time. It's likely to not cause real problems, but It's still not perfect IMO.
Instead, this should be handled separately, as a one shot command when actually required. For example, in Kubernetes in would be good to run a dedicated migration job along with your application release, if the database schema has actually changed.
With compose, there are no jobs, but you can achieve similar behaviour.
services:
migration:
image: busybox
command: sh -c 'echo "running migration..."; sleep 20; echo "migration completed"'
app:
image: busybox
command: echo "app started"
depends_on:
migration:
condition: service_completed_successfully
deploy:
replicas: 3
Now you row the migration only once and all 3 app replicas wait for the migration to complete before they start up.
$ docker compose up
Attaching to app_1, app_2, app_3, migration_1
migration_1 | running migration...
migration_1 | migration completed
migration_1 exited with code 0
app_2 | app started
app_3 | app started
app_1 | app started
In your case, you would use the same image you build from the Dockerfile for both migration and app service. In the migration service you use knex migrate
and in the app service you use npm run start
.
If you need the migration to even wait for the DB, depends_on might not be sufficient, unless you build in a health check that reflects if the database is actually ready to accept a connection. If you have a health check, then you can use the condition service_healthy
.
For example, you could dome something like this.
services:
db:
image: mysql
environment:
MYSQL_ROOT_PASSWORD: "root"
MYSQL_DATABASE: "wordpress"
MYSQL_USER: "wordpressuser"
MYSQL_PASSWORD: "wordpresspassword"
healthcheck:
test: mysqladmin -u root --password=$$MYSQL_ROOT_PASSWORD ping
interval: 30s
timeout: 10s
retries: 10
migration:
image: busybox
command: sh -c 'echo "running migration..."; sleep 20; echo "migration completed"'
depends_on:
db:
condition: service_healthy
app:
image: busybox
command: echo "app started"
depends_on:
migration:
condition: service_completed_successfully
deploy:
replicas: 3
You can health check logs by doing a container inspect.
$ docker inspect sample_db_1 --format \
'{{range .State.Health.Log}}{{.End}} | Exit Code: {{.ExitCode}} | {{.Output}}{{end}}'
2022-01-30 12:53:43.749365 +0000 UTC | Exit Code: 0 | mysqladmin: [Warning] Using a password on the command line interface can be insecure.
mysqld is alive
If you don't want to use a health check, you can also use third party solutions like https://github.com/Eficode/wait-for.