Enable graceful shutdown of rq workers (#5214)

* Enable graceful shutdown of rq workers

* Use `exec` in the `worker` command of the entrypoint to propagate
  the `TERM` signal
* Allow rq processes managed by supervisor to exit without restart on
  expected status codes
* Allow supervisorctl to contact the running supervisor
* Add a `shutdown_worker` command that will send `TERM` to all running
  worker processes and then sleep. This allows orchestration systems
  to initiate a graceful shutdown before sending `SIGTERM` to
  supervisord

* Use Heroku worker as the BaseWorker

This implements a graceful shutdown on SIGTERM, which simplifies
external shutdown procedures.

* Fix imports based upon review

* Remove supervisorctl config
This commit is contained in:
Josh Bohde
2020-11-05 03:49:45 -06:00
committed by GitHub
parent c6bf8a1c55
commit e2e8714155
2 changed files with 6 additions and 5 deletions

View File

@@ -19,7 +19,7 @@ worker() {
export WORKERS_COUNT=${WORKERS_COUNT:-2}
export QUEUES=${QUEUES:-}
supervisord -c worker.conf
exec supervisord -c worker.conf
}
dev_worker() {

View File

@@ -3,7 +3,8 @@ import os
import signal
import time
from redash import statsd_client
from rq import Worker as BaseWorker, Queue as BaseQueue, get_current_job
from rq import Queue as BaseQueue, get_current_job
from rq.worker import HerokuWorker # HerokuWorker implements graceful shutdown on SIGTERM
from rq.utils import utcnow
from rq.timeouts import UnixSignalDeathPenalty, HorseMonitorTimeoutException
from rq.job import Job as BaseJob, JobStatus
@@ -40,7 +41,7 @@ class RedashQueue(StatsdRecordingQueue, CancellableQueue):
pass
class StatsdRecordingWorker(BaseWorker):
class StatsdRecordingWorker(HerokuWorker):
"""
RQ Worker Mixin that overrides `execute_job` to increment/modify metrics via Statsd
"""
@@ -58,7 +59,7 @@ class StatsdRecordingWorker(BaseWorker):
statsd_client.incr("rq.jobs.failed.{}".format(queue.name))
class HardLimitingWorker(BaseWorker):
class HardLimitingWorker(HerokuWorker):
"""
RQ's work horses enforce time limits by setting a timed alarm and stopping jobs
when they reach their time limits. However, the work horse may be entirely blocked