How to debug sglang using pdb?

[sleepwalker2017]

I'm trying to debug the code using pdb. but when I set breakpoints in the loop, it reports error.

> /usr/local/lib/python3.10/dist-packages/sglang/srt/managers/scheduler.py(414)event_loop_overlap()
-> print("self.run_batch in event_loop_overlap")
(Pdb)
[2024-12-13 08:38:01 TP0] Scheduler hit an exception: Traceback (most recent call last):
  File "/usr/local/lib/python3.10/dist-packages/sglang/srt/managers/scheduler.py", line 1502, in run_scheduler_process
    scheduler.event_loop_overlap()
  File "/usr/local/lib/python3.10/dist-packages/torch/utils/_contextlib.py", line 116, in decorate_context
    return func(*args, **kwargs)
  File "/usr/local/lib/python3.10/dist-packages/sglang/srt/managers/scheduler.py", line 414, in event_loop_overlap
    print("self.run_batch in event_loop_overlap")
  File "/usr/local/lib/python3.10/dist-packages/sglang/srt/managers/scheduler.py", line 414, in event_loop_overlap
    print("self.run_batch in event_loop_overlap")
  File "/usr/lib/python3.10/bdb.py", line 90, in trace_dispatch
    return self.dispatch_line(frame)
  File "/usr/lib/python3.10/bdb.py", line 115, in dispatch_line
    if self.quitting: raise BdbQuit
bdb.BdbQuit

Killed

Here is the breakpoint I set.

    @torch.no_grad()
    def event_loop_overlap(self):
        """A scheduler loop that overlaps the CPU processing and GPU computation."""
        result_queue = deque()

        while True:
            recv_reqs = self.recv_requests()
            self.process_input_requests(recv_reqs)

            batch = self.get_next_batch_to_run()
            self.cur_batch = batch
            if batch:
                import pdb; pdb.set_trace()
                print("self.run_batch in event_loop_overlap")
                result = self.run_batch(batch)
                result_queue.append((batch.copy(), result))

How can I trace the code using pdb ? Thank you!

[JamesSand]

It seems the schedulers are launched by multiprocessing. Please refer here
I think pdb cannot be used directly in multiprocessing setting. Maybe you can try to change the code when you debug, to remove the multiprocessing?