rospy.ServiceProxy master port mismatch
On my dev machine I run ros master on default port 11311 via ROSLaunchParent as follows:
os.environ["ROS_IP"] = "<dev-machine-ip>"
os.environ["ROS_MASTER_URI"] = "http://<dev-machine-ip>:11311"
roslaunch = ROSLaunchParent('', [], is_core=True, port=11311)
roslaunch.start()
stdout states that the roslaunch server is running on port 43121 (changes randomly between starts of roslaunch):
started roslaunch server http://<dev-machine-ip>:43121/
The master seems to be started on port 11311:
auto-starting new master
process[master]: started with pid [15304]
ROS_MASTER_URI=http://<dev-machine-ip>:11311/
I use docker-py (dc is a client, di is an images instance, dn is a network) to run a ROS node <some-node> in a docker container which connects to the dev machines master.
dn = dc.networks.create(name='some_test', driver="bridge")
some_node = dc.containers.run(image=di[0].id, environment=["ROS_IP=<some-ip-in-the-dev-machine-subnet>", "ROS_MASTER_URI=http://<dev-machine-ip>:11311"], name='<some-node>', network='test', command=rosrun_command, detach=True, tty=True)
import rosnode; nodes = rosnode.get_node_names(); print(nodes) lists /<some-node> (and /rosout). some-node is (at least) discoverable in the network. So far so good.
If I send multiple service requests with
rc_itempick_start = rospy.ServiceProxy('/<some-node-namespace>/<some-service>', <service-type>)
rc_itempick_start()
a random port is used instead of 11311:
core.py WARNING Unknown error initiating TCP/IP socket to <some-ip-in-the-dev-machine-subnet>:35516 (rosrpc://<some-ip-in-the-dev-machine-subnet>:35516)
(...)
error: [Errno 113] No route to host
core.py WARNING Unknown error initiating TCP/IP socket to <some-ip-in-the-dev-machine-subnet>:60739 (rosrpc://<some-ip-in-the-dev-machine-subnet>:60739)
(...)
error: [Errno 113] No route to host
(etc.)
Can someone help out in cross-checking what I missed?
The port in the
rosrpcURIs is not11311(or the port of the roslaunch server) because those URIs are not pointing to the master (or the roslaunch server), but to the port the service server uses to accept an incoming service request.after your edit:
This points to your service server not having been configured with the correct value for either
ROS_IPorROS_HOSTNAME. Or if it is the correct IP, then the docker container can't reach it.Right. I am not sure what to look at else w.r.t. service server config. I am not sure what the default docker network behaviour is. probably good to continue debugging here...
The Python stuff you show seems to setup a
bridged network, so IPs in thedev-machine-subnetshould probably work. Issome-ip-in-the-dev-machine-subnetin therosrpcURI an expected value? Can youpingthat IP from the container? If not, that would be the first thing to fix.Btw: I don't see you configuring the IP for the container to use. If the container uses a 'random' IP instead of the one you set in
ROS_IP, then things probably won't work.Edit: in fact, if the code sample you show is what you actually run I'm pretty sure that is the problem. The ..
.. container must have the same IP as you pass in for
ROS_IP. Even iftestis abridged network, handing out IPs from another subnet will most likely not work. You'll have to use IPs from thetestnetwork. Afaik,bridgedfor a Docker network is not like VMware'sbridged(for example).Right, totally missed that. When using docker-py setting the ip of the container explicitly requires to configuring the network (subnets , etc.).