Cannot step through subprocess code in debugger using debugpy

[andrewbgross]

Type: Bug

This was working as of six months ago or so, the last time I used it. So there's definitely either a change that requires me to change some configuration or setting that I'm not aware of, or it's a regression.

My environment: several Docker containers running different services, all written in Python. One of those containers is running Celery, which creates ton of subprocess (one per task, where tasks are sometimes alive for the life of the service, but are often quite short lived-- a matter of seconds.)

I connect into the service running in the container remotely using an RPC port, even though I'm actually running VS Code on the same machine that's running all the containers/services.

This used to work exactly as I expected. But now, I can connect to the processes, and I can set breakpoints, and when I'm paused on a breakpoint, I can inspect all the values on the call stack. However, I cannot step through the code. If I use F10 or F11, they act as if I use F5. I can set breakpoints on 10 lines of code in a row, and successfully go from line to line; but if I only set the first breakpoint, when I hit F10, it will not go to the next line. There is some evidence that it's actually swithching over to one of the other subproceses, but if that was the only problem, I would expect that eventually I would get switched back to the subprocess I hit F10 in, and be on that line, but that doesn't happen; when hitting F10, it just continues the subprocess, as if I hit F5.

I saw some reports from years ago that setting "Debug code I wrote only" to True fixed that, but it didn't help for me. I also saw someone mention the setting that pauses all subprocesses when you hit a breakpoint, and doesn't unfreeze them when you hit F10 or F5 in the subprocess that hit the breakpoint; but that didn't seem to help, and I didn' spend much time experimenting, because it's not a viable solution-- the subprocesses are exchanging data with each other constantly, so having the other subprocesses paused won't allow the subprocess you're debugging make much progress.

VS Code version: Code - Insiders 1.101.0-insider (ccdd214171190f69e28c8c3def68a6315f4d9ae0, 2025-05-20T07:09:03.342Z)
OS version: Linux x64 6.8.0-60-generic
Modes:

System Info

Item	Value
CPUs	Intel(R) Celeron(R) N5095 @ 2.00GHz (4 x 2800)
GPU Status	2d_canvas: unavailable_software canvas_oop_rasterization: disabled_off direct_rendering_display_compositor: disabled_off_ok gpu_compositing: disabled_software multiple_raster_threads: enabled_on opengl: disabled_off rasterization: disabled_software raw_draw: disabled_off_ok skia_graphite: disabled_off video_decode: disabled_software video_encode: disabled_software vulkan: disabled_off webgl: unavailable_software webgl2: unavailable_software webgpu: disabled_off webnn: unavailable_software
Load (avg)	2, 2, 2
Memory (System)	15.40GB (11.70GB free)
Process Argv	--crash-reporter-id e3255827-8157-4f07-9470-8120011e919a
Screen Reader	no
VM	0%
DESKTOP_SESSION	ubuntu
XDG_CURRENT_DESKTOP	Unity
XDG_SESSION_DESKTOP	undefined
XDG_SESSION_TYPE	x11

Extensions (13)

Extension	Author (truncated)	Version
docker	doc	0.6.0
gitlens	eam	17.1.0
copilot	Git	1.322.0
vscode-docker	ms-	1.29.6
debugpy	ms-	2025.8.0
python	ms-	2025.6.1
vscode-pylance	ms-	2025.5.1
jupyter	ms-	2025.4.0
jupyter-keymap	ms-	1.1.2
jupyter-renderers	ms-	1.1.0
vscode-jupyter-cell-tags	ms-	0.1.9
vscode-jupyter-slideshow	ms-	0.1.6
even-better-toml	tam	0.21.2

A/B Experiments

vsliv368cf:30146710
vspor879:30202332
vspor708:30202333
vspor363:30204092
pythonvspyt551cf:31249598
vscod805cf:30301675
c4g48928:30535728
vscrp:30624060
962ge761:30841072
dsvsc014:30777825
dsvsc015:30821418
h48ei257:31000450
pythontbext0:30879054
cppperfnew:30980852
dwnewjupyter:31046869
pythonrstrctxt:31093868
nativeloc1:31118317
e80f6927:31120813
dwcopilot:31158714
c3hdf307:31184662
6074i472:31201624
dwoutputs:31242946
customenabled:31248079
9064b325:31222308
copilot_t_ci:31222730
g012b348:31231168
pythoneinst12:31251391
bgtreat:31252123
c7cif404:31309980
pythonpulldiag:31287486
6gi0g917:31259950
editoredits36407:31282444
996jf627:31264550
pythonrdcb7:31268811
usemplatestapi:31297334
8ge78292:31309974
747dc170:31275146
pythonfullctx:31296836
6518g693:31302842
pylancecolor:31288664
generatesymbolt:31280541
convertfstringf:31280702
gendocf:31280703
pylancequickfixt:31311477
ajf51753:31312179
923gi514:31310909
b99bg931:31306656
a2ijf515:31307742
convertlamdat:31311571

[andrewbgross]

I see that this got assigned to someone, for which I am grateful. I realized after I opened this bug that I could have made this a more useful bug report by, for example, trying to set up a minimal repro-- though in this case, a minimal repro is still going to be much more complex than usual, because you need to have a linux server, you need to be running docker, you need to be running celery and have multiple celery tasks (subprocesses) running at the same time, etc. Still, if there's anything I can do to help-- recording a video of the behavior, supplying version numbers of other software, whatever-- please don't hesitate to ask. It's even possible that I might be able to set up a minimal repro environment on an Azure instance and give you access to that so you can reproduce it yourself without needing to spend considerable time creating your own environment (I'd need to get permission from my employer, but I think it would be fine.) Really, whatever you need-- I'm doing this on "my time", so it might take a couple of days to get back to you for more involved requests, but I'm willing to do whatever I can until you decide that we're no longer making progress. (I worked at Microsoft for 22+ years, and my 2nd role there was as an SDE on the Windows Sustained Engineering team, which a the time was supporting both NT4 and WindowsXP, so I have an enormous amount of both sympathy and respect for what you do.) Thanks!

[rmpIGN]

I'm having the same issue. There's no need for a Docker environment to reproduce the issue; it happened to me with a simple multiprocessing script. In the main process, the debugger works as expected, but in the subprocesses, the 'stepping over' (F10) and 'stepping into' (F11) functions behave as though the 'continue execution' (F5) function had been selected.

I tried downgrading the ms-python.debugpy extension to each version from 2025.6.0 to 2024.14.0, but it had no effect.

I'm not sure if this is relevant, but I upgraded my computer from Ubuntu 22.04 to Ubuntu 24.04 last week, and this behaviour appeared right after that. It seems pretty odd; it could just be a coincidence.

[rmpIGN]

It seems related to microsoft/debugpy#1876

Anyway, I can reproduce it with this simple script :

import multiprocessing as mp
import time


def f():
    for i in range(10):
        time.sleep(1)
        print("hello subprocess")
        print(i)


if __name__ == "__main__":
    p = mp.Process(target=f)
    print("hello main process")
    p.start()
    p.join()

[andrewbgross]

@rmpIGN I can't express how grateful I am that you took the time and went to the effort to not only read my post, but to answer me here and point me to the existing, related thread. You have saved me a lot of work, at a time that the small startup I'm working for is overwhelmed and can't afford to have me off duplicating work that is already being done. Given my lack of experience with multi-process programming and debugging, just coming up with the minimal repro you provide could very well have taken me days.

You've really made my day, and I've got a fresh injection of energy that I strongly suspect is going to make this weekend a very productive one for me.

Thanks again!

[andrewbgross]

I'm not sure if this is relevant, but I upgraded my computer from Ubuntu 22.04 to Ubuntu 24.04 last week, and this behaviour appeared right after that. It seems pretty odd; it could just be a coincidence.

Just in case this turns out to be relevant, I also upgraded last week, to Ubuntu 24.04.2 LTS. Here are the relevant lines from my /var/log/apt/history.log file:

Start-Date: 2025-05-16 06:59:16
Commandline: /usr/bin/unattended-upgrade
Install: linux-modules-6.8.0-60-generic:amd64 (6.8.0-60.63, automatic), linux-image-6.8.0-60-generic:amd64 (6.8.0-60.63, automatic), linux-headers-6.8.0-60-generic:amd64 (6.8.0-60.63, automatic), linux-tools-6.8.0-60-generic:amd64 (6.8.0-60.63, automatic), linux-tools-6.8.0-60:amd64 (6.8.0-60.63, automatic), linux-modules-extra-6.8.0-60-generic:amd64 >
Upgrade: linux-tools-common:amd64 (6.8.0-59.61, 6.8.0-60.63), linux-headers-generic:amd64 (6.8.0-59.61, 6.8.0-60.63), linux-generic:amd64 (6.8.0-59.61, 6.8.0-60.63), linux-image-generic:amd64 (6.8.0-59.61, 6.8.0-60.63), linux-libc-dev:amd64 (6.8.0-59.61, 6.8.0-60.63)
End-Date: 2025-05-16 06:59:40

[rmpIGN]

@andrewbgross you're welcome, I'm glad I could help.

I found a temporary fix: downgrading ms-code.debugpy to version 2024.10.0. This makes debugging work again.