Update the default Gunicorn API server workers count to one (#1454)

* Update the default API server GUnicorn workers to 1.

* Update dependency injection wire functionality for yatai/deployment/sagemaker/serve.

* Update wire modules in yatai/deployment/sagemaker/serve.

* Use app_server_workers in yatai/deployment/sagemaker/serve.

* Fix typo.

bentoml/configuration/containers.py

@@ -84,6 +84,9 @@ def __init__(
                 self.config["api_server"]["port"] = config("apiserver").getint(
                     "default_port"
                 )
+                self.config["api_server"]["workers"] = config("apiserver").getint(
+                    "default_gunicorn_workers_count"
+                )
                 self.config["api_server"]["max_request_size"] = config(
                     "apiserver"
                 ).getint("default_max_request_size")

bentoml/configuration/default_bentoml.cfg

@@ -76,8 +76,9 @@ default_image_input_accept_file_extensions = .jpg,.png,.jpeg,.tiff,.webp,.bmp
 
 # Set to a positive integer to take effect, otherwise will fallback to a
 # runtime calculated value based on cpu cores
-# see `bentoml.server.utils.get_gunicorn_num_of_workers` for details
-default_gunicorn_workers_count = -1
+# see `bentoml.configuration.containers.BentoMLContainer.api_server_workers`
+# for details
+default_gunicorn_workers_count = 1
 batch_request_header = Bentoml-Is-Batch-Request
 
 

bentoml/configuration/default_bentoml.yml

@@ -18,7 +18,7 @@ api_server:
   # automatically, see `bentoml.configuration.containers.BentoMLContainer.
   # api_server_workers` for details.
   #
-  workers: Null
+  workers: 1
 
   timeout: 60
 

bentoml/server/instruments.py

@@ -1,3 +1,4 @@
+import logging
 import multiprocessing
 import os
 import shutil
@@ -6,7 +7,9 @@
 from flask import Request
 
 from bentoml import config
-from bentoml.server.utils import logger
+
+
+logger = logging.getLogger(__name__)
 
 
 class InstrumentMiddleware:

bentoml/yatai/deployment/sagemaker/serve

@@ -10,13 +10,9 @@ import subprocess
 import os
 import signal
 import sys
+from dependency_injector.wiring import inject, Provide
 
-from bentoml.server.utils import get_gunicorn_num_of_workers
-
-bento_server_timeout = os.environ.get('BENTOML_GUNICORN_TIMEOUT', 60)
-bento_server_workers = int(
-    os.environ.get('BENTOML_GUNICORN_NUM_OF_WORKERS', get_gunicorn_num_of_workers())
-)
+from bentoml.configuration.containers import BentoMLConfiguration, BentoMLContainer
 
 
 def sigterm_handler(nginx_pid, gunicorn_pid):
@@ -32,7 +28,11 @@ def sigterm_handler(nginx_pid, gunicorn_pid):
     sys.exit(0)
 
 
-def _serve():
+@inject
+def _serve(
+    bento_server_timeout: int = Provide[BentoMLContainer.config.api_server.timeout],
+    bento_server_workers: int = Provide[BentoMLContainer.api_server_workers],
+):
     # link the log streams to stdout/err so they will be logged to the container logs
     subprocess.check_call(['ln', '-sf', '/dev/stdout', '/var/log/nginx/access.log'])
     subprocess.check_call(['ln', '-sf', '/dev/stderr', '/var/log/nginx/error.log'])
@@ -66,4 +66,14 @@ def _serve():
 
 
 if __name__ == '__main__':
+    container = BentoMLContainer()
+    config = BentoMLConfiguration()
+    if "BENTOML_GUNICORN_TIMEOUT" in os.environ:
+        config.override(["api_server", "timeout"], int(os.environ.get("BENTOML_GUNICORN_TIMEOUT")))
+    if "BENTOML_GUNICORN_NUM_OF_WORKERS" in os.environ:
+        config.override(["api_server", "workers"], int(os.environ.get("BENTOML_GUNICORN_NUM_OF_WORKERS")))
+    container.config.from_dict(config.as_dict())
+
+    container.wire(modules=[sys.modules[__name__]])
+
     _serve()