ROS 2 Jazzy: every node hangs at startup and ignores Ctrl-C after a Fast DDS crash (stale /dev/shm)
Symptom: after one node (rviz2) aborted with Fatal glibc error: pthread_mutex_lock.c:94 ... assertion failed: mutex->__data.__owner == 0 and the launch was force-killed, every later ros2 launch misbehaved. Nodes started but never got far (rviz never printed its OpenGL line), and on Ctrl-C almost every node, C++ and Python alike, ignored SIGINT and SIGTERM until launch escalated to SIGKILL. Only static_transform_publishers and a trivial node exited cleanly.
Cause: Fast DDS (default RMW) shared-memory transport segments in /dev/shm (fastrtps_*) were left behind by the SIGKILLed processes, some with a robust mutex still held. New participants block on them, and the force-kills that follow leave even more stale segments, so it feeds on itself. /dev/shm had 172 fastrtps entries.
Fix: with no ROS processes running, fastdds shm clean (ships with ROS 2; source /opt/ros/jazzy/setup.bash first). It removes only segments with no live owner. 172 entries went down to 32 (normal fastrtps_portNNNN files). Rebooting also works, since /dev/shm is tmpfs.
Diagnosis tip: grep the launch.log files under ~/.ros/log for "escalating to 'SIGKILL'" to find the first bad run, then look just before it for the crash. If it keeps recurring, force UDP-only transport with a Fast DDS XML profile, or switch RMW.
Works inside distrobox, since /dev/shm is shared with the host.