Race condition between roscore and roslaunch --wait #1097
Description
Steps to reproduce
echo "<launch></launch>" > launch.xml
roscore & roslaunch --wait launch.xml
Rarely, both roscore and roslaunch will start the rosout node. If this happens, the rosout nodes will conflict forever as they are both started with respawn="true".
To make this more likely to happen, you can do the following:
edit _setup() in roslaunch/launch.py:
self._launch_master()
import random
time.sleep(random.random())
self._launch_core_nodes()
edit _wait_for_master() in roslaunch/rlutil.py:
while not is_running:
# time.sleep(0.1)
is_running = m.is_running()
run this script and look for multiple "started core service" lines:
#!/bin/bash
export PYTHONUNBUFFERED=yes
roscore | sed -n "/core service/ {s/^/roscore: /; p}" &
for i in $(seq 1 100); do
roslaunch --wait launch.xml | sed -n "/core service/ {s/^/roslaunch $i: /; p}" &
done
sleep 5
pkill -P $$
wait
Explanation
If you run roslaunch --wait, it will wait for the master before starting, but the --wait flag has no effect on the starting of core nodes. In roslaunch/launch.py, we have:
# start up the core: master + core nodes defined in core.xml
self._launch_master()
self._launch_core_nodes()
unconditionally of whether --core or --wait is used. In the --wait case, rlutil._wait_for_master has already been called, and _launch_master() notices that the master is already running so does nothing. However, _launch_core_nodes may still try to launch rosout if the roscore hasn't started it yet.
Workaround
Run this command before roslaunch --wait:
until rosnode info rosout | grep Pid; do sleep 1; done