Sentry and multiprocessing?

I can’t get multiprocessing spawned worker functions to work with sentry when the worker functions raises an exception.

With below code, I expect both of the spawned worker_function to separately send logs to sentry when the ZeroDivisionError is raised. However, sentry doesn’t appear to send any logs at all. Am I missing something, or this is not how it works. Cheers.

import multiprocessing

dsn = 'https://xxx:[e-mail]/zzzz'

def worker_function():
    from raven import Client
    client = Client(dsn)
    print("worker crashes")
    1 / 0

def manager_function():
    workers_count = 2
    workers = [multiprocessing.Process(target=worker_function) for i in range(workers_count)]
    for worker in workers:
        worker.start()

manager_function()

[hitmeupallthetime]

A solution can be found below. I monkeypatched multiprocessing run, then used the synchronous transport (the async doesn’t work, because it needs to send the pending request when the process ends, but the hook isn’t there).

import multiprocessing
import threading
import sys
from raven import Client
from raven.transport.requests import HTTPTransport

dsn = "xx"


def install_thread_excepthook():
    """
    Workaround for sys.excepthook thread bug
    (https://sourceforge.net/tracker/?func=detail&atid=105470&aid=1230540&group_id=5470).
    Call once from __main__ before creating any threads.
    If using psyco, call psycho.cannotcompile(threading.Thread.run)
    since this replaces a new-style class method.
    """
    start_old = multiprocessing.Process.run
    def run(*args, **kwargs):
        client = Client(dsn, transport=HTTPTransport)
        try:
            start_old(*args, **kwargs)
        except (KeyboardInterrupt, SystemExit):
            raise
        except:
            client.captureException()
    multiprocessing.Process.run = run

install_thread_excepthook()

def worker_function():

    print("worker crashes")
    1 / 0

def manager_function():
    workers_count = 2
    workers = [multiprocessing.Process(target=worker_function) for i in range(workers_count)]
    for worker in workers:
        worker.start()

manager_function()

[ashwoods]

Sentry handling exceptions by default works by using sys.excepthook, which doesn’t work with threading and multiprocessing due to python issue 1230540. Workarounds exist but a bit tricky. Wrapping your code in a try and using client.CaptureException should work fine. A long term fix is already on our radar.