Add environment variable with VMARGS

Author: nikhil-sk

src/sagemaker_inference/environment.py

@@ -25,6 +25,7 @@
 DEFAULT_MODEL_SERVER_TIMEOUT = "60"
 DEFAULT_STARTUP_TIMEOUT = "600"  # 10 minutes
 DEFAULT_HTTP_PORT = "8080"
+DEFAULT_VMARGS = "-XX:-UseContainerSupport"
 
 SAGEMAKER_BASE_PATH = os.path.join("/opt", "ml")  # type: str
 
@@ -70,15 +71,12 @@ def __init__(self):
             os.environ.get(parameters.MODEL_SERVER_TIMEOUT_ENV, DEFAULT_MODEL_SERVER_TIMEOUT)
         )
         self._model_server_workers = os.environ.get(parameters.MODEL_SERVER_WORKERS_ENV)
-        self._startup_timeout = int(
-            os.environ.get(parameters.STARTUP_TIMEOUT_ENV, DEFAULT_STARTUP_TIMEOUT)
-        )
-        self._default_accept = os.environ.get(
-            parameters.DEFAULT_INVOCATIONS_ACCEPT_ENV, content_types.JSON
-        )
+        self._startup_timeout = int(os.environ.get(parameters.STARTUP_TIMEOUT_ENV, DEFAULT_STARTUP_TIMEOUT))
+        self._default_accept = os.environ.get(parameters.DEFAULT_INVOCATIONS_ACCEPT_ENV, content_types.JSON)
         self._inference_http_port = os.environ.get(parameters.BIND_TO_PORT_ENV, DEFAULT_HTTP_PORT)
         self._management_http_port = os.environ.get(parameters.BIND_TO_PORT_ENV, DEFAULT_HTTP_PORT)
         self._safe_port_range = os.environ.get(parameters.SAFE_PORT_RANGE_ENV)
+        self._vmargs = os.environ.get(parameters.VMARGS, DEFAULT_VMARGS)
 
     @staticmethod
     def _parse_module_name(program_param):
@@ -140,3 +138,8 @@ def safe_port_range(self):  # type: () -> str
         specified by SageMaker for handling pings and invocations.
         """
         return self._safe_port_range
+
+    @property
+    def vmargs(self):  # type: () -> str
+        """str: vmargs can be provided for the JVM, to be overriden"""
+        return self._vmargs

src/sagemaker_inference/model_server.py

@@ -159,7 +159,7 @@ def _generate_mms_config_properties(env, handler_service=None):
         "default_workers_per_model": env.model_server_workers,
         "inference_address": "http://0.0.0.0:{}".format(env.inference_http_port),
         "management_address": "http://0.0.0.0:{}".format(env.management_http_port),
-        "vmargs": "-XX:-UseContainerSupport",
+        "vmargs": env.vmargs,
     }
     # If provided, add handler service to user config
     if handler_service:

src/sagemaker_inference/parameters.py

@@ -24,3 +24,4 @@
 BIND_TO_PORT_ENV = "SAGEMAKER_BIND_TO_PORT"  # type: str
 SAFE_PORT_RANGE_ENV = "SAGEMAKER_SAFE_PORT_RANGE"  # type: str
 MULTI_MODEL_ENV = "SAGEMAKER_MULTI_MODEL"  # type: str
+VMARGS = "VMARGS" # type: str

test/unit/test_environment.py

@@ -28,6 +28,7 @@
         parameters.DEFAULT_INVOCATIONS_ACCEPT_ENV: "text/html",
         parameters.BIND_TO_PORT_ENV: "1738",
         parameters.SAFE_PORT_RANGE_ENV: "1111-2222",
+        parameters.VMARGS: "-XX:-UseContainerSupport",
     },
     clear=True,
 )
@@ -45,6 +46,7 @@ def test_env():
     assert env.inference_http_port == "1738"
     assert env.management_http_port == "1738"
     assert env.safe_port_range == "1111-2222"
+    assert "-XX:-UseContainerSupport" in env.vmargs
 
 
 @pytest.mark.parametrize("sagemaker_program", ["program.py", "program"])