Edge Agent makes assumptions about swarm mode

[SvenDowideit]

I have a system in my network that has docker swarm mode enabled, but is not the system that the portainer server is running on.

when I use docker run to start an edge agent on this host, it complains that it needs AGENT_CLUSTER_ADDR set.

In this case, I want to have a non-swarm cluster agent...

[deviantony]

@SvenDowideit The agent automatically detects if it is running inside a Swarm cluster and if so, requires the AGENT_CLUSTER_ADDR to be configured to be able to form an agent cluster (even for a single node cluster).

Deploy the agent as a service and it will be fine.

See #70 for an enhancement that will allow you to avoid the usage of the AGENT_CLUSTER_ADDR env var.

[SvenDowideit]

@deviantony I think you missunderstood - I need this agent to not insist that its in swarm mode. The fact this node is for other purposes set to enable swarm has nothing to do with how this node should be used by portainer.

I use the same nodes for multiple projects - this notebook is used for testing $work related swarm things, and for non-$ volunteer things for working on very non-swarm related things. TBH, one of them will likely run 2 agents, talking to different portainers.

[deviantony]

@SvenDowideit what prevents you from simply deploying the agent as a service on that node?

EDIT: just saw your comment in #70 (comment)

#70 is unrelated to your question then and is only related to the current usage of the automatic Swarm mode.

If you want to treat a node of your cluster as a standalone node I'm afraid that the agent won't be able to help here. You can see the agent as a wrapper around the Docker API and Portainer will actually query the agent to retrieve information about the node and enable Swarm management features if needed.

I also don't see the problem about having two agents running on the same node (cluster mode enabled, with Swarm awareness) and use a Portainer instance to manage node scoped resources and another to manage cluster scoped resources.

[SvenDowideit]

this issue is closed - so as a user, I'd prefer to discuss my use case in an open issue....

but basically automation - in the edge agent setup, the dual homed system will have its agents deployed using automation which doens't use swarm services, because the production systems don't use swarm mode.

I also am not sure that the node that is running in a swarm is necessarily a manager node - it just happens to be used for more than one purpose.

[deviantony]

Sorry had trouble to figure out your use case yesterday !

From what I can tell, your feature request should actually be opened inside the Portainer repository (even if it is tied to the agent in your use case).

The reason is that the first Portainer do after connecting to an endpoint for management is to issue a request to retrieve the status of the engine associated to the endpoint (is it part of a Swarm). If the engine is part of a Swarm then Portainer will display the required management features.

As the agent is only a wrapper around the Docker API to provide a different way to connect to it as well as extra features, this feature request is not really related to the agent if I understand your use case correctly.

[SvenDowideit]

@deviantony - IDK maybe - I assumed it was the agent refusing to go through the motions of talking to portainer because it saw that docker was in swarm mode, and it didn't have the right cfg details to do swarm mode right.