Dagster with celery and docker

This is an example of a simple parallel pipeline packaged in docker that runs with celery through a simple docker-compose up command. It uses:

docker to deploy database, rabbitmq, dagster-dagit and dagster-celery workers
celery as execution backend
s3 as intermediate storage backend

This example is meant to illustrate how the components in a simple scalable dagster deployment are connected and configured. For instance, one particular aspect that is a little untouched in the documentation is how to use a custom broker connection string and what could be the contents of the yaml passed to the dagster-celery worker initialization.

Run example

Prerequisites:

s3 bucket and an account with write and read permissions on the bucket

Steps:

setup the aws s3 environment variables of the account, these are used in the docker-compose and passed to the dagster instances
- AWS_ACCESS_KEY_ID
- AWS_SECRET_ACCESS_KEY
- AWS_DEFAULT_REGION
change the configuration file in app/celery_execution.yaml to the appropriate bucket name

build the image with

$ docker build . -t dagster_test

The code is copied directly into the image

then deploy the nodes with docker-compose

$ docker-compose up

There could be some issues with rabbitmq. In theory, the configuration inside .docker/rabbitmq/etc/rabbitmq.conf should create a vhost called dagster that could be accesed with credentials rabbitmq:rabbitmq(user:pass). If this is not the case access the web management server in localhost:15672 and create the vhost.

The compose spins up a dagit server and a dagster worker. they use the same image but different entrypoints. Test the pipeline by going to localhost:2001 select the parallel_pipeline and the preset celery in the playground

nikkij/dagster-celery-test

Dagster with celery and docker

Run example