Add server option for serving only artifacts and proxied serving mode

dev/small-requirements.txt

@@ -12,3 +12,4 @@ pytest-localserver==0.5.0
 moto!=2.0.7
 azure-storage-blob>=12.0.0
 azure-identity>=1.6.1
+requests

examples/mlflow_artifacts/README.md

@@ -16,16 +16,19 @@ First, launch the tracking server with the artifacts service via `mlflow server`
 ```sh
 # Launch a tracking server with the artifacts service
 $ mlflow server \
+    --serve-artifacts
     --artifacts-destination ./mlartifacts \
     --default-artifact-root http://localhost:5000/api/2.0/mlflow-artifacts/artifacts/experiments \
     --gunicorn-opts "--log-level debug"
 ```
 
 Notes:
 
+- `--serve-artifacts` enables the MLflow Artifacts service endpoints to enable proxied serving of artifacts through the REST API
 - `--artifacts-destination` specifies the base artifact location from which to resolve artifact upload/download/list requests. In this examples, we're using a local directory `./mlartifacts`, but it can be changed to a s3 bucket or
 - `--default-artifact-root` points to the `experiments` directory of the artifacts service. Therefore, the default artifact location of a newly-created experiment is set to `./mlartifacts/experiments/<experiment_id>`.
 - `--gunicorn-opts "--log-level debug"` is specified to print out request logs but can be omitted if unnecessary.
+- `--artifacts-only` disables all other endpoints for the tracking server apart from those involved in listing, uploading, and downloading artifacts. This makes the MLflow server a single-purpose proxy for artifact handling only.
 
 Then, run `example.py` that performs upload, download, and list operations for artifacts:
 

examples/mlflow_artifacts/docker-compose.yml

@@ -54,6 +54,7 @@ services:
       --port 5500
       --artifacts-destination s3://bucket
       --gunicorn-opts "--log-level debug"
+      --serve-artifacts
 
   postgres:
     image: postgres

examples/mlflow_artifacts/example.py

@@ -10,6 +10,10 @@ def save_text(path, text):
         f.write(text)
 
 
+#  NOTE: ensure the tracking server has been started with --serve-artifacts to enable
+#        MLflow artifact serving functionality.
+
+
 def main():
     assert "MLFLOW_TRACKING_URI" in os.environ
 

mlflow/cli.py

@@ -320,6 +320,25 @@ def _validate_static_prefix(ctx, param, value):  # pylint: disable=unused-argume
     "Default: Within file store, if a file:/ URI is provided. If a sql backend is"
     " used, then this option is required.",
 )
+@click.option(
+    "--serve-artifacts",
+    is_flag=True,
+    help="If specified, enables serving of artifact uploads, downloads, and list requests "
+    "by routing these requests to the storage location that is specified by "
+    "'--artifact-destination' directly through a proxy. The default location that "
+    "these requests are served from is a local './mlartifacts' directory which can be "
+    "overridden via '--artifact-destination' arguments. "
+    "Default: False",
+)
+@click.option(
+    "--artifacts-only",
+    is_flag=True,
+    help="If specified, configures the mlflow server to be used only for proxied artifact serving. "
+    "With this mode enabled, functionality of the mlflow tracking service (e.g. run creation, "
+    "metric logging, and parameter logging are disabled. The server will only expose "
+    "endpoints for uploading, downloading, and listing artifacts. "
+    "Default: False",
+)
 @cli_args.ARTIFACTS_DESTINATION
 @cli_args.HOST
 @cli_args.PORT
@@ -348,6 +367,8 @@ def _validate_static_prefix(ctx, param, value):  # pylint: disable=unused-argume
 def server(
     backend_store_uri,
     default_artifact_root,
+    serve_artifacts,
+    artifacts_only,
     artifacts_destination,
     host,
     port,
@@ -395,6 +416,8 @@ def server(
         _run_server(
             backend_store_uri,
             default_artifact_root,
+            serve_artifacts,
+            artifacts_only,
             artifacts_destination,
             host,
             port,

mlflow/server/__init__.py

@@ -20,6 +20,8 @@
 ARTIFACT_ROOT_ENV_VAR = "_MLFLOW_SERVER_ARTIFACT_ROOT"
 ARTIFACTS_DESTINATION_ENV_VAR = "_MLFLOW_SERVER_ARTIFACT_DESTINATION"
 PROMETHEUS_EXPORTER_ENV_VAR = "prometheus_multiproc_dir"
+SERVE_ARTIFACTS_ENV_VAR = "_MLFLOW_SERVER_SERVE_ARTIFACTS"
+ARTIFACTS_ONLY_ENV_VAR = "_MLFLOW_SERVER_ARTIFACTS_ONLY"
 
 REL_STATIC_DIR = "js/build"
 
@@ -106,6 +108,8 @@ def _build_gunicorn_command(gunicorn_opts, host, port, workers):
 def _run_server(
     file_store_path,
     default_artifact_root,
+    serve_artifacts,
+    artifacts_only,
     artifacts_destination,
     host,
     port,
@@ -126,6 +130,10 @@ def _run_server(
         env_map[BACKEND_STORE_URI_ENV_VAR] = file_store_path
     if default_artifact_root:
         env_map[ARTIFACT_ROOT_ENV_VAR] = default_artifact_root
+    if serve_artifacts:
+        env_map[SERVE_ARTIFACTS_ENV_VAR] = "true"
+    if artifacts_only:
+        env_map[ARTIFACTS_ONLY_ENV_VAR] = "true"
     if artifacts_destination:
         env_map[ARTIFACTS_DESTINATION_ENV_VAR] = artifacts_destination
     if static_prefix: