Shepherd

A Docker swarm service for automatically updating your services whenever their base image is refreshed.

Usage

docker service create --name shepherd \
                      --constraint "node.role==manager" \
                      --mount type=bind,source=/var/run/docker.sock,target=/var/run/docker.sock,ro \
                      mazzolino/shepherd

Or with Docker Compose

version: "3"
services:
  ...
  shepherd:
    build: .
    image: mazzolino/shepherd
    volumes:
      - /var/run/docker.sock:/var/run/docker.sock
    deploy:
      placement:
        constraints:
        - node.role == manager

See the example docker-compose.yml and some more compose configs in the examples folder.

Configuration

Shepherd will try to update your services every 5 minutes by default. You can adjust this value using the SLEEP_TIME variable.

You can prevent services from being updated by appending them to the IGNORELIST_SERVICES variable. This should be a space-separated list of service names.

Alternatively you can specify a filter for the services you want updated using the FILTER_SERVICES variable. This can be anything accepted by the filtering flag in docker service ls.

You can set Shepherd to roll back a service to the previous version if the update fails by setting the ROLLBACK_ON_FAILURE variable.

You can control additional parameters for the docker service update and docker service update --rollback calls using the variables UPDATE_OPTIONS and ROLLBACK_OPTIONS.

If the docker service update takes too long then it will be killed after 5 minutes by default. You can adjust this value using the TIMEOUT variable.

You can enable private registry authentication by setting the WITH_REGISTRY_AUTH variable.

If you need to authenticate to a registry (for example in order to get around the Docker Hub rate limits), you can set the variable REGISTRY_USER and store the password either in a docker secret named shepherd_registry_password or in the environment variable REGISTRY_PASSWORD. If you are not using Docker Hub but a private registry, set REGISTRY_HOST to the hostname of your registry.

It is also possible to put all authentication information in a secret file. This approach is required if you need to authenticate with multiple accounts to the same registry (eg authenticate to the Gitlab registry for images from different projects). The content of the file is <TAB> separated and has 4 columns:

• id: an identifier for the account. This should be an acceptable Docker config name.
• registry: the registry to authenticate against, eg registry.gitlab.com
• login: the user to authenticate as
• password: the password to authenticate with Lines starting with # are comments, and are ignored, as are invalid lines. Here is an example:

blog	registry.gitlab.com	gitlab+deploy-token-5123674	ssw2Nrd2

Create and edit that file locally, eg at the location private/shepherd-registries-auth, and create the docker secret with

docker secret create shepherd-registries-auth private/shepherd-registries-auth

You need to make the secret available to shepherd with the secrets key, and pass the secret file to the REGISTRIES_FILE variable:

services:
  app:
    image: mazzolino/shepherd
    environment:
      REGISTRIES_FILE: /var/run/secrets/shepherd-registries-auth
    secrets:
      - shepherd-registries-auth
secrets:
    shepherd-registries-auth:
      external: true

You also need to add a label shepherd.auth.config to the container to be updated specifying which line of the secret file should be used. The value of that label should be the id in the secret file:

    deploy:
        labels:
            - shepherd.enable=true
            - shepherd.auth.config=blog

You can enable connection to insecure private registry by setting the WITH_INSECURE_REGISTRY variable to true.

You can prevent pulling images from the registry by setting the WITH_NO_RESOLVE_IMAGE variable to true.

You can enable notifications on service update with apprise, using the apprise microservice and the APPRISE_SIDECAR_URL variable. See the file docker-compose.apprise.yml for an example.

You can enable old image autocleaning on service update by setting the IMAGE_AUTOCLEAN_LIMIT variable.

You can enable one shot running with RUN_ONCE_AND_EXIT variable.

If you care about log entries having the right timezone, you can set the TZ variable to the correct value (make sure to not include quotation marks in the variable value).

Example:

docker service create --name shepherd \
                    --constraint "node.role==manager" \
                    --env SLEEP_TIME="5m" \
                    --env IGNORELIST_SERVICES="shepherd my-other-service" \
                    --env WITH_REGISTRY_AUTH="true" \
                    --env WITH_INSECURE_REGISTRY="true" \
                    --env WITH_NO_RESOLVE_IMAGE="true" \
                    --env FILTER_SERVICES="label=com.mydomain.autodeploy" \
                    --env APPRISE_SIDECAR_URL="apprise-microservice:5000" \
                    --env IMAGE_AUTOCLEAN_LIMIT="5" \
                    --env RUN_ONCE_AND_EXIT="true" \
                    --env ROLLBACK_ON_FAILURE="true" \
                    --env UPDATE_OPTIONS="--update-delay=30s" \
                    --env TZ=Europe/Berlin \
                    --mount type=bind,source=/var/run/docker.sock,target=/var/run/docker.sock,ro \
                    --mount type=bind,source=/root/.docker/config.json,target=/root/.docker/config.json,ro \
                    mazzolino/shepherd

Running on a cron schedule

When running shepherd as described with a SLEEP_TIME, the de facto running times will drift the longer the container is running. If you want to run shepherd on a fixed schedule instead, it is recommended to pair it with swarm-cronjob:

1. Create a swarm-cronjob service as described in its documentation.
2. Set RUN_ONCE_AND_EXIT to true, replicas to 0 and restart_policy to condition: none. Add docker labels for the schedule.

See docker-compose.swarm-cronjob.yml for a full stack example which includes both shepherd as well as swarm-cronjob.

How does it work?

Shepherd just triggers updates by updating the image specification for each service, removing the current digest.

Most of the work is thankfully done by Docker which resolves the image tag, checks the registry for a newer version and updates running container tasks as needed.

Also, Docker handles all the work of applying rolling updates. So at least with replicated services, there should be no noticeable downtime.