如何配置哨兵从python多处理池产生的进程发送异常？

Question

I have a script which is run by celery worker which uses Pool from billiard library and I spawned multiple process. I am trying to use sentry inside those process so that any unhandled/handled exception can be caught. Written below is my sample code :

from configurations import SENTRY_CLIENT


def process_data(data):
try:
    s = data/0
except ZeroDivisionError:
    print "Sentry must report this."
    SENTRY_CLIENT.captureException()

import multiprocessing
from billiard import Pool
POOL_SIZE=multiprocessing.cpu_count()
pool = Pool(POOL_SIZE)
data=[0, 1, 2, 3, 4, 5]
pool.map(process_data, data)
pool.close()
pool.terminate()

SENTRY_CLIENT is defined in configuration file which is defined as : configurations.py

from raven import Client
SENTRY_CLIENT = Client("dsn")

One way which I am trying is to pass SENTRY_CLIENT to each process but I am trying to avoid that as of now. Also as this script is executed by celery worker I had configured sentry for celery and any exception till pool.map() is well caught by sentry.

Also I tried to print the SENTRY_CLIENT.__dict__ and I got the valid items with correct values. My issue is here that why SENTRY_CLIENT is not sending exceptions to sentry dashboard. May be I am missing something in configurations.

Answer 1

I finally get the solution by some readings. Sentry works on an async event based model and killing a process just after triggering sentry will not ensure the exception is reached to the servers. Hence we need to add a delay(10s), before killing the process in case of any exception to ensure sentry does it's job.

def process_data(data):
    from configurations import SENTRY_CLIENT
    try:
       s = data/0
    except ZeroDivisionError:
       print "Sentry must report this."
       import time
       SENTRY_CLIENT.captureException()
       time.sleep(10)

import multiprocessing
from billiard import Pool
POOL_SIZE=multiprocessing.cpu_count()
pool = Pool(POOL_SIZE)
data=[0, 1, 2, 3, 4, 5]
pool.map(process_data, data)
pool.close()
pool.terminate()

Answer 2

As PoloSoares said , you should change a transport instead of adding any delay by sleep. Example of valid solution for 6.10.0 version of raven lib:

import multiprocessing

from billiard import Pool
from raven import Client
from raven.transport.http import HTTPTransport

SENTRY_CLIENT = Client("dsn", transport=HTTPTransport)


def process_data(data):
    try:
        s = data / 0
    except ZeroDivisionError:
        print("Sentry must report this.")
        SENTRY_CLIENT.captureException()


POOL_SIZE = multiprocessing.cpu_count()
pool = Pool(POOL_SIZE)
data = [0, 1, 2, 3, 4, 5]
pool.map(process_data, data)
pool.close()
pool.terminate()