fix(cli): add start runner server back (#4479)

src/bentoml_cli/start.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import json
 import logging
 import os
 import sys
@@ -32,6 +33,13 @@ def add_start_command(cli: click.Group) -> None:
         envvar="BENTOML_SERVE_DEPENDS",
         help="list of runners map",
     )
+    @click.option(
+        "--runner-map",
+        type=click.STRING,
+        envvar="BENTOML_SERVE_RUNNER_MAP",
+        help="[Deprecated] use --depends instead. "
+        "JSON string of runners map. For backword compatibility for yatai < 1.0.0",
+    )
     @click.option(
         "--bind",
         type=click.STRING,
@@ -128,6 +136,7 @@ def start_http_server(  # type: ignore (unused warning)
         bento: str,
         service_name: str,
         depends: list[str] | None,
+        runner_map: str | None,
         bind: str | None,
         port: int,
         host: str,
@@ -158,6 +167,8 @@ def start_http_server(  # type: ignore (unused warning)
             sys.path.insert(0, working_dir)
         if depends:
             runner_map_dict = dict([s.split("=", maxsplit=2) for s in depends or []])
+        elif runner_map:
+            runner_map_dict = json.loads(runner_map)
         else:
             runner_map_dict = {}
 
@@ -372,3 +383,100 @@ def start_grpc_server(  # type: ignore (unused warning)
             max_concurrent_streams=max_concurrent_streams,
             protocol_version=protocol_version,
         )
+
+    @cli.command(hidden=True)
+    @click.argument("bento", type=click.STRING, default=".")
+    @click.option(
+        "--runner-name",
+        type=click.STRING,
+        required=True,
+        envvar="BENTOML_SERVE_RUNNER_NAME",
+        help="specify the runner name to serve",
+    )
+    @click.option(
+        "--bind",
+        type=click.STRING,
+        help="[Deprecated] use --host and --port instead."
+        "Bind address for the server. For backword compatibility for yatai < 1.0.0",
+        required=False,
+    )
+    @click.option(
+        "--port",
+        type=click.INT,
+        default=BentoMLContainer.http.port.get(),
+        help="The port to listen on for the REST api server",
+        envvar="BENTOML_PORT",
+        show_default=True,
+    )
+    @click.option(
+        "--host",
+        type=click.STRING,
+        default=BentoMLContainer.http.host.get(),
+        help="The host to bind for the REST api server [defaults: 127.0.0.1(dev), 0.0.0.0(production)]",
+        envvar="BENTOML_HOST",
+    )
+    @click.option(
+        "--backlog",
+        type=click.INT,
+        default=BentoMLContainer.api_server_config.backlog.get(),
+        help="The maximum number of pending connections.",
+        show_default=True,
+    )
+    @click.option(
+        "--working-dir",
+        type=click.Path(),
+        help="When loading from source code, specify the directory to find the Service instance",
+        default=None,
+        show_default=True,
+    )
+    @click.option(
+        "--timeout",
+        type=click.INT,
+        help="Specify the timeout (seconds) for runners",
+        envvar="BENTOML_TIMEOUT",
+    )
+    @add_experimental_docstring
+    def start_runner_server(  # type: ignore (unused warning)
+        bento: str,
+        runner_name: str,
+        bind: str | None,
+        port: int,
+        host: str,
+        backlog: int,
+        working_dir: str | None,
+        timeout: int | None,
+    ) -> None:
+        """
+        Start Runner server standalone. Deprecate in 1.2.0
+        """
+        if working_dir is None:
+            if os.path.isdir(os.path.expanduser(bento)):
+                working_dir = os.path.expanduser(bento)
+            else:
+                working_dir = "."
+        if sys.path[0] != working_dir:
+            sys.path.insert(0, working_dir)
+
+        if bind is not None:
+            parsed = urlparse(bind)
+            assert parsed.scheme == "tcp"
+            host = parsed.hostname or host
+            port = parsed.port or port
+
+        from bentoml.start import start_runner_server as start_runner_server_impl
+
+        if bind is not None:
+            parsed = urlparse(bind)
+            assert parsed.scheme == "tcp"
+            host = parsed.hostname or host
+            port = parsed.port or port
+
+        start_runner_server_impl(
+            bento,
+            runner_name=runner_name,
+            working_dir=working_dir,
+            timeout=timeout,
+            port=port,
+            host=host,
+            backlog=backlog,
+        )