Fix missing ROS_MASTER_URI on master deployment

[marcoshuck]

Change

This PR solves an error while deploying the ROS application on Kubernetes

The ROS_MASTER_URI was not configured on the master node, and therefore a new env var was added to the master deployment.

Issue

While launching the ROS master deployment on Kubernetes, the following error was returned:

/opt/ros/melodic/etc/catkin/profile.d/10.roslaunch.sh: line 3: ROS_MASTER_URI: unbound variable

[marcoshuck]

Care to give this a review? @diegoferigo thanks! 😃

[diegoferigo]

@marcoshuck The ROS_MASTER_URI variable is exported automatically by the entrypoint.sh:

ros-kubernetes/docker/master/entrypoint.sh

Line 8 in 316101e

source "/opt/ros/$ROS_DISTRO/setup.bash"

I'm not sure exporting this variable is the cleanest way to handle it. I suspect you encountered this problem after entering the container with docker exec [...]. In this case, the entrypoint is not executed and the script is not sourced. Maybe in this case the best solution is sourcing the script in the interpreter configuration (e.g. /etc/bash.bashrc).

My point is that if you don't have that variable defined, your setup is not sourcing the ROS initialization script. Exporting manually the variable will only provide a partially configured setup that could miss something else perhaps in other cases. The best way is always source the script.

[marcoshuck]

Hey @diegoferigo, thanks for your response.

I agree it's not the cleanest solution, but I should mention that I got this problem while initializing the set of pods on my minikube cluster.

That being said, images for those pods were generated as described in the docs using docker-compose build and added to the cluster repository with minikube cache add.

Taking everything into account, I imagine the problem, even if it's related to what you've described, is happening somewhere else.

[marcoshuck]

This problem only happens on the ROS master, and that's not the case for the workers (talker and listener deployments)

[marcoshuck]

Does it make sense to reproduce this problem to see if I'm the only one who's having it?

[diegoferigo]

At the moment I don't have access to a machine in which I can install and configure minikube. Can you make sure that the entrypoint is called in the master deployment? If you browse the logs, you should read the echoed lines:

==> Executing master image entrypoint ...
-> Setting up ROS
==> Container ready

Another attempt could be to inspect the environment that roscore finds before running. I think you can get it printed to the logs by running env && roscore (ported to kubernetes syntax) in the args key:

ros-kubernetes/k8s/master-deployment.yaml

Lines 23 to 29 in 316101e

	- name: rosclustermaster
	image: roscluster/master:v0
	args:
	- roscore
	ports:
	- containerPort: 11311
	name: roscoreport

[marcoshuck]

Sure, thanks @diegoferigo, will get back as soon as I have some logs.