Merge pull request #3048 from ktleung2017/add-keepalive

SeldonIO · Apr 12, 2021 · 357ba4d · 357ba4d
2 parents 2858011 + 9ccd636
commit 357ba4d
Show file tree

Hide file tree

Showing 2 changed files with 9 additions and 0 deletions.
diff --git a/doc/source/python/python_server.md b/doc/source/python/python_server.md
@@ -193,6 +193,7 @@ Python Server can be configured using environmental variables or command line fl
 | `--threads`             | `GUNICORN_THREADS`                     | `10`          | Number of threads to run per Gunicorn worker.                                                                                                                      |
 | `--max-requests`        | `GUNICORN_MAX_REQUESTS`                | `0`           | Maximum number of requests gunicorn worker will process before restarting.                                                                                         |
 | `--max-requests-jitter` | `GUNICORN_MAX_REQUESTS_JITTER`         | `0`           | Maximum random jitter to add to max-requests.                                                                                                                      |
+| `--keepalive`           | `GUNICORN_KEEPALIVE`                   | `2`           | The number of seconds to wait for requests on a Keep-Alive connection.                                                                                         |
 | `--access-log`          | `GUNICORN_ACCESS_LOG`                  | `false`       | Enable gunicorn access log.                                                                                                                                        |
 | `--pidfile`             | N/A                                    | None          | A file path to use for the Gunicorn PID file.                                                                                                                      |
 | `--single-threaded`     | `FLASK_SINGLE_THREADED`                | `0`           | Force the Flask app to run single-threaded. Also applies to Gunicorn. Can be `0` or `1`.                                                                           |

diff --git a/python/seldon_core/microservice.py b/python/seldon_core/microservice.py
@@ -266,6 +266,12 @@ def main():
         default=int(os.environ.get("GUNICORN_MAX_REQUESTS_JITTER", "0")),
         help="Maximum random jitter to add to max-requests.",
     )
+    parser.add_argument(
+        "--keepalive",
+        type=int,
+        default=int(os.environ.get("GUNICORN_KEEPALIVE", "2")),
+        help="The number of seconds to wait for requests on a Keep-Alive connection.",
+    )
 
     parser.add_argument(
         "--single-threaded",
@@ -406,6 +412,7 @@ def rest_prediction_server():
                 "max_requests_jitter": args.max_requests_jitter,
                 "post_worker_init": post_worker_init,
                 "worker_exit": partial(worker_exit, seldon_metrics=seldon_metrics),
+                "keepalive": args.keepalive,
             }
             if args.pidfile is not None:
                 options["pidfile"] = args.pidfile
@@ -468,6 +475,7 @@ def rest_metrics_server():
                 "max_requests": args.max_requests,
                 "max_requests_jitter": args.max_requests_jitter,
                 "post_worker_init": post_worker_init,
+                "keepalive": args.keepalive,
             }
             if args.pidfile is not None:
                 options["pidfile"] = args.pidfile