Skip to content

Docker

Jump to bottom
Baud Rémy edited this page May 12, 2021 · 1 revision

Docker

The Flask project is easiest run within docker and very easiest with docker-compose, as described in the Quick Start section and the docker installation guide.

Most important is that the docker-compose.yml is not altered unless you know what you're doing. This might seem strange and comes from the fact that the app spins up containers to generate the routing graphs in-code. That makes supporting host-native installations as well as docker installations pretty painful.

Environment Variables

The default environment file is ./.docker_env which is referenced in the docker-compose.yml of this project. It already comes with a preconfigured set of variables, but can be altered/extended as you like.

Generally, the same list applies here as in the general Configuration list. The only important thing for docker is that the database hosts are named the exact same as the corresponding service in docker-compose.yml.

Two notable exceptions are:

  • COMPOSE_PROJECT_NAME: determines the internal docker-compose project name (usually defaults to the project's parent directory name). This is specified in .env and is important for docker-compose to generate a docker volume with the correct name (s. below).
  • DATA_DIR: needs to be set by both, the .env and .docker_env files. The .env file will tell docker-compose what DATA_DIR maps to when creating the docker volume (see below) and the .docker_env file will configure the containers.

All containers need access to the ./.docker_env environment file.

Volumes

app container

The container running the app only needs one volume which maps the host's DATA_DIR directory to the container's /app/data directory.

However, that volume needs to have the name routing-packager_packages because this volume will also need to be fetched by the app in-code. It's autogenerated by docker-compose from the COMPOSE_PROJECT_NAME and the volume name in docker-compose.yml and mapped to the DATA_DIR directory found in .env.

If that is failing for some reason, you can always generate the volume manually:

docker volume create routing-packager_packages \
    --driver local \
    --opt type=none \
    --opt device=/my/absolute/path
    --opt o=bind

You'll also notice that the containers running this image map the /var/run/docker.sock socket to the container's docker socket. This is important to be able to use docker-in-docker.

worker container

The worker container running the graph generation and graph updates, needs the same volume(s) mapped as the app container.

Additionally, it needs access to the .docker_env file within the container and must map to /app/.env. The reason is, that the worker container will fire up a flask process to update the graph packages and flask will only look .flaskenv and .env files. When you change .docker_env variables the changes will be propagated to the container only if it's mapped from the host into the container.

Containers

This project needs 2 containers from this image:

  • app: runs the HTTP API, i.e. Flask app. Also is responsible for updating the OSM PBF files with cron.
  • worker: runs the RQ worker instance, which works off jobs from the queue. Also is responsible for updating the graph packages with cron.

The ENTRYPOINT for the image is ./docker-entrypoint.sh and the CMD can be:

  • app (default): runs the Flask app
  • worker: runs the RQ worker instance
Clone this wiki locally