What is a process group?
In Unix/Linux terms, a process group is a way the OS groups related processes so they can be signaled and managed as a unit. Each group is identified by a process group ID (PGID).
1
kill -TERM -12345
You’re not killing PID 12345 — the leading minus means send the signal to the entire process group with PGID 12345.
Process groups are used mainly for job control and terminal/shell management.
Example:
1
ros2 run my_pkg my_node | tee output.log
That single pipeline creates multiple processes:
bashros2- your node process
teeteereads standard input and writes it both to standard output and to one or more files — effectively “tees” the stream so you can view output and save it at the same time.
When you press Ctrl+C, the terminal sends SIGINT to the foreground process group, so all of those processes receive the signal and stop together.
Process group vs parent/child (important distinction)
- Parent/child: who started whom.
- Process group: which processes should be controlled together.
These are orthogonal concepts. For example:
- One parent, many children, all in the same process group.
- One parent with children in different groups.
Creating a new process group
1
2
3
4
subprocess.Popen(
cmd,
start_new_session=True
)
On Linux this uses setsid() under the hood and creates a new session and a new process group; the child becomes PGID = PID.
Avoid patterns that create zombies or make process trees hard to manage:
- Don’t use
pkill -9inside_start_processes()(this can create zombies and kill unrelated processes). - Prefer launching without
bash -cwhen possible (shell wrappers complicate process trees).
Typical process tree example: subprocess.Popen() → bash → ros2 → actual node process. If you kill the node (a grandchild), the parent may become a defunct/zombie process if it doesn’t reap its children properly.
