Logging Driver: Allow driver to use docker DNS by joining docker network

[hamiltont]

Is your feature request related to a problem? Please describe.

We run a docker swarm and are looking into the Loki driver. One snag - all examples I can find for loki-url use either a static domain (e.g. loki.foo.com) or they use the docker host IP address (host.docker.internal).

Neither of these work well a swarm with multiple nodes that is running a loki instance inside the swarm. The docker host address will not work at all on nodes other than the one node running loki

Describe the solution you'd like

Ideally, we could specify which network to attach the loki driver to (currently it runs with the host's network). That would allow the common docker pattern of using a docker-provided DNS name that round-robins to the matching containers. This would also work well in swarm mode using an overlay network. For example:

loki-url: http://loki:3100/loki/api/v1/push
loki-network: loki

Describe alternatives you've considered

We considered HTTP redirect sidecars to make the host.docker.internal approach work on all nodes, even those not running loki. Problem is you need to run the sidecar on all nodes except the node running loki, so you now have to label your nodes specifically to work around this issue.

We also considering using a static name (e.g. loki.foo.com) and only making it available from inside the swarm. While this would work for our production stack, anyone trying to do local development would have a problem (we often spin up a vagrant-powered swarm cluster locally) and have to reconfigure their local machine to affix a permanent IP address to the loki container, which can change location. In short, they would have exactly the type of issue the docker dns system addresses ;-)

[stale]

This issue has been automatically marked as stale because it has not had any activity in the past 30 days. It will be closed in 7 days if no further activity occurs. Thank you for your contributions.

[hamiltont]

Bump

[stale]

This issue has been automatically marked as stale because it has not had any activity in the past 30 days. It will be closed in 7 days if no further activity occurs. Thank you for your contributions.

[hamiltont]

bump

[caleb-adams]

This is highly desirable for my team.

[rosconap]

Would like some more info on this as well.

[stale]

This issue has been automatically marked as stale because it has not had any activity in the past 30 days. It will be closed in 7 days if no further activity occurs. Thank you for your contributions.

[batazor]

up

[stale]

This issue has been automatically marked as stale because it has not had any activity in the past 30 days. It will be closed in 7 days if no further activity occurs. Thank you for your contributions.

[hamiltont]

🔼

[stale]

This issue has been automatically marked as stale because it has not had any activity in the past 30 days. It will be closed in 7 days if no further activity occurs. Thank you for your contributions.

[batazor]

🆙

[cyriltovena]

There's nothing we can do, by default the plugin runs on the host network.

see #1368 (which has workaround and a lot of explanation) and #1962

[KES777]

@cyriltovena How to run plugin on different network?

My service is ran on its own network:

networks:
  mynet:

It would be nice that logging driver loki will attach to service's network. And this issue should be mentioned on documentation page: https://grafana.com/docs/loki/latest/clients/docker-driver/

I dig whole the day, to understand what is happening... =(