[ros-users] Question about roslaunch and remote processes.

[Joan Pau Beltran]

Hi everyone,

I have to questions about roslaunch and remote processes that seem to 
point some bugs out.
We are using Diamondback as current release.

Firstly, the output=screen attribute seems to be ignored, since I don't 
get any output with lines like:
> <launch>
> <node machine="fugu-c" pkg="memsense_imu" type="imu_node" 
> name="imu_node" output="screen">
> <rosparam file="$(find fugu_configurations)/imu/params.yaml" />
> <param name="frame_id" value="imu" />
> </node>
> </launch>
I am sure (because I wrote it myself) that the node is using the 
rosconsole macros for output. According to the wiki, one of the reasons 
to use the macros instead of the standard c++ streams is the possibility 
to get output from remotely launched processes. Does anyone know what 
may be happening?


The machine file looks like this:
> <launch>
> <machine name="fugu-c" address="fugu-c.local" user="user"
>     ros-root="/opt/ros/diamondback/ros/"
>     ros-package-path="/home/user/ros:/opt/ros/diamondback/stacks" />
> </launch>


And the second one, it seems that the roslaunch mechanism does not 
honour the ROS_HOSTNAME variable for remote machines. In the .bashrc 
file of the remote machine this variable is set to 'fugu-c.local', to 
allow the remote machine name to be resolved properly using Avahi (that 
seems to become quite standard in the latest distributions, this is the 
reason for the .local suffix).
However when launching the above file, connection is stablished with the 
remote machine, even the core is found if running there, but any node is 
launched. This is the output:
> started roslaunch server http://10.42.43.1:51495/
> remote[fugu-c.local-0] starting roslaunch
> remote[fugu-c.local-0]: creating ssh connection to fugu-c.local:22, 
> user[user]
> remote[fugu-c.local-0]: ssh connection created
>
> SUMMARY
> ========
>
> PARAMETERS
>  ...
> MACHINES
>  * fugu-c
>
> NODES
>   /
>     imu_node (memsense_imu/imu_node)
>
> ROS_MASTER_URI=http://fugu-c.local:11311
>
> core service [/rosout] found
> error launching on [fugu-c.local-0, uri http://fugu-c:39451/]: Name or 
> service not known
> Launch of the following nodes most likely failed: memsense_imu/imu_node
Apparently, the uri is computed as the result of the hostname command on 
the remote machine, which returns only 'fugu-c' instead of 
'fugu-c.local'. I expected that setting the ROS_HOSTNAME would overwrite 
this behaviour (actually it works for finding the remote core). If this 
is a bug or a possible improvement, where should I submmit it? If not, 
any idea on how to handle this? By now, we added the corresponding entry 
to /etc/hosts to overcome the problem, but I would expect a better solution.


Thank you in advance!

-- 
Joan Pau Beltran
Grup de Sistemes, Robòtica i Visió - DMI
Universitat de les Illes Balears