diff --git a/bin/docker-entrypoint b/bin/docker-entrypoint
index f54dd20643..b5d7b0ac17 100755
--- a/bin/docker-entrypoint
+++ b/bin/docker-entrypoint
@@ -18,8 +18,8 @@ worker() {
 
   export WORKERS_COUNT=${WORKERS_COUNT:-2}
   export QUEUES=${QUEUES:-}
-  
-  supervisord -c worker.conf
+
+  exec supervisord -c worker.conf
 }
 
 dev_worker() {
diff --git a/redash/tasks/worker.py b/redash/tasks/worker.py
index 2fc67415a5..d4ca454a31 100644
--- a/redash/tasks/worker.py
+++ b/redash/tasks/worker.py
@@ -3,7 +3,8 @@
 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 @@ def execute_job(self, job, queue):
                 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