Closed QiangWuCN closed 2 years ago
@QiangWuCN I suspect it is the same as https://github.com/eclipse-cyclonedds/cyclonedds/issues/479, so it would be worthwhile to try changing that if it is possible for you to do a custom build of cyclonedds
.
@eboasson Thanks a lot. The problem is fixed by building a custom build of cyclonedds.
As for a 128 cores machine, the max number of threads is changed from "64" to "256" (Tested "128" but still failed) in the source file: cyclonedds/src/core/ddsc/src/dds_init.c Line 113 in b84eee5
thread_states_init (64);
Just for the record, the following branches of cyclonedds and rmw_cyclonedds are used:
git clone https://github.com/eclipse-cyclonedds/cyclonedds eclipse-cyclonedds/cyclonedds -b release/0.7.x
git clone https://github.com/ros2/rmw_cyclonedds ros2/rmw_cyclonedds -b foxy
Bug report
Required Info:
Steps to reproduce issue
Expected behavior
Gazebo runs smoothly. Tested successfully w/ Fast DDS.
Actual behavior
Gzserver crashes within several minutes after starting with the following logging.
Additional information
Ubuntu 20.04 + ROS 2 Foxy + Gazebo 11 is being used in a container, while the host OS is Ubuntu 20.04 and the Docker version is version 20.10.16.
Tried to enlarge the receive buffer size by using the below command, but didn't resolve the problem:
sudo sysctl -w net.core.rmem_max=8388608 net.core.rmem_default=8388608