Merge branch 'main' of github.com:zhaochenyang20/sglang-diffusion-routing

Author: zhaochenyang20

.github/ISSUE_TEMPLATE/1-bug-report.yml

@@ -0,0 +1,40 @@
+name: 🐞 Bug report
+description: Report a bug to help us reproduce and fix it.
+title: "[Bug] "
+labels: ['Bug']
+
+body:
+- type: checkboxes
+  attributes:
+    label: Checklist
+    options:
+      - label: I searched related issues but found no solution.
+      - label: The bug persists in the latest version.
+      - label: Issues without environment info and a minimal reproducible demo are hard to resolve and may receive no feedback.
+      - label: If this is not a bug report but a general question, please start a discussion in Issues. Otherwise, it will be closed.
+      - label: Please use English. Otherwise, it will be closed.
+- type: textarea
+  attributes:
+    label: Describe the bug
+    description: A clear, concise description of the bug.
+  validations:
+    required: true
+- type: textarea
+  attributes:
+    label: Reproduction
+    description: Command/script run and configuration used.
+    placeholder: Paste the command here.
+  validations:
+    required: true
+- type: textarea
+  attributes:
+    label: Environment
+    description: |
+      Provide your environment info, including:
+      - OS version
+      - Python version
+      - sglang-diffusion-routing version (`pip show sglang-diffusion-routing`)
+      - Related dependency versions (fastapi, httpx, uvicorn)
+    placeholder: Paste environment output here.
+  validations:
+    required: true

.github/ISSUE_TEMPLATE/2-feature-request.yml

@@ -0,0 +1,23 @@
+name: 🚀 Feature request
+description: Suggest an idea for this project
+title: "[Feature] "
+
+body:
+- type: checkboxes
+  attributes:
+    label: Checklist
+    options:
+      - label: If this is not a feature request but a general question, please start a discussion in Issues. Otherwise, it will be closed.
+      - label: Please use English. Otherwise, it will be closed.
+- type: textarea
+  attributes:
+    label: Motivation
+    description: |
+      Clearly and concisely describe the feature's motivation.
+  validations:
+    required: true
+- type: textarea
+  attributes:
+    label: Related resources
+    description: |
+      Provide official releases or third-party implementations if available.

.github/pull_request_template.md

@@ -0,0 +1,31 @@
+<!-- Thank you for your contribution! Please follow these guidelines to enhance your pull request. If anything is unclear, submit your PR and reach out to maintainers for assistance. -->
+
+## Motivation
+
+<!-- Describe the purpose and goals of this pull request. -->
+
+## Modifications
+
+<!-- Detail the changes made in this pull request. -->
+
+## Accuracy Tests
+
+<!-- If this pull request affects routing behavior or model outputs, provide accuracy test results. -->
+
+## Benchmarking and Profiling
+
+<!-- If this pull request impacts routing performance or throughput, provide benchmarking and profiling results. -->
+
+## Checklist
+
+- [ ] Format your code with `pre-commit run --all-files`.
+- [ ] Add or update unit tests if applicable.
+- [ ] Update documentation if applicable.
+- [ ] Provide accuracy and performance benchmark results if applicable.
+- [ ] Ensure all CI checks pass.
+
+## Review Process
+
+1. Get approvals from maintainers and other reviewers.
+2. Ensure all CI tests pass.
+3. After green CI and required approvals, ask maintainers to merge.

.pre-commit-config.yaml

@@ -17,6 +17,7 @@ repos:
     rev: 5.13.2
     hooks:
       - id: isort
+        args: ["--profile", "black"]
 
   - repo: https://github.com/astral-sh/ruff-pre-commit
     rev: v0.11.7

README.md

@@ -152,6 +152,30 @@ curl -X POST http://localhost:30081/update_weights_from_disk \
     -d '{"model_path": "Qwen/Qwen-Image-2512"}'
 ```
 
+### Auto-launch workers via YAML config
+
+Instead of starting workers manually, you can let the router spawn and manage
+them through a launcher backend.
+
+**Local subprocess launcher** (`examples/local_launcher.yaml`):
+
+```bash
+sglang-d-router --port 30081 --launcher-config examples/local_launcher.yaml
+```
+
+```yaml
+launcher:
+  backend: local
+  model: Qwen/Qwen-Image
+  num_workers: 2
+  num_gpus_per_worker: 1
+  worker_base_port: 10090
+  wait_timeout: 600
+```
+
+Fields not set in the YAML fall back to defaults defined in each backend's
+config dataclass (see `LocalLauncherConfig`).
+
 ## Acknowledgment
 
 This project is derived from [radixark/miles#544](https://github.com/radixark/miles/pull/544). Thanks to the original authors.

development.md

@@ -6,7 +6,7 @@
 pip install -e .
 ```
 
-Run tests:
+Run CPU only tests:
 
 ```bash
 pip install pytest

examples/local_launcher.yaml

@@ -0,0 +1,11 @@
+launcher:
+  model: Qwen/Qwen-Image
+
+  num_workers: 8
+  num_gpus_per_worker: 1
+  worker_host: "127.0.0.1"
+  worker_base_port: 10090
+
+  worker_extra_args: "--dit-cpu-offload false --text-encoder-cpu-offload false"
+
+  wait_timeout: 600

pyproject.toml

@@ -12,6 +12,7 @@ license = { text = "MIT" }
 dependencies = [
     "fastapi>=0.110",
     "httpx>=0.27",
+    "omegaconf>=2.3",
     "uvicorn>=0.30",
 ]
 classifiers = [

src/sglang_diffusion_routing/cli/main.py

@@ -6,13 +6,15 @@
 import argparse
 import asyncio
 import sys
+import threading
 
 from sglang_diffusion_routing import DiffusionRouter
+from sglang_diffusion_routing.launcher import config as _lcfg
 
 
 def _run_router_server(
     args: argparse.Namespace,
-    worker_urls: list[str] | None = None,
+    router: DiffusionRouter,
     log_prefix: str = "[router]",
 ) -> None:
     try:
@@ -22,10 +24,7 @@ def _run_router_server(
             "uvicorn is required to run router. Install with: pip install uvicorn"
         ) from exc
 
-    worker_urls = list(
-        worker_urls if worker_urls is not None else args.worker_urls or []
-    )
-    router = DiffusionRouter(args, verbose=args.verbose)
+    worker_urls = list(args.worker_urls or [])
     refresh_tasks = []
     for url in worker_urls:
         normalized_url = router.normalize_worker_url(url)
@@ -97,13 +96,52 @@ def _add_router_args(parser: argparse.ArgumentParser) -> None:
     parser.add_argument(
         "--log-level", type=str, default="info", help="Uvicorn log level."
     )
+    parser.add_argument(
+        "--launcher-config",
+        type=str,
+        default=None,
+        dest="launcher_config",
+        help="YAML config for launching router managed workers (see examples/local_launcher.yaml).",
+    )
 
 
 def _handle_router(args: argparse.Namespace) -> int:
-    _run_router_server(
-        args, worker_urls=list(args.worker_urls), log_prefix="[sglang-d-router]"
-    )
-    return 0
+    log_prefix, backend, router = "[sglang-d-router]", None, None
+
+    try:
+        router = DiffusionRouter(args, verbose=args.verbose)
+        if args.launcher_config is not None:
+            launcher_cfg = _lcfg.load_launcher_config(args.launcher_config)
+            wait_timeout = launcher_cfg.wait_timeout
+            backend = _lcfg.create_backend(launcher_cfg)
+            backend.launch()
+            threading.Thread(
+                target=backend.wait_ready_and_register,
+                kwargs=dict(
+                    register_func=router.register_worker,
+                    timeout=wait_timeout,
+                    log_prefix=log_prefix,
+                ),
+                daemon=True,
+            ).start()
+
+        _run_router_server(args, router=router, log_prefix=log_prefix)
+        return 0
+    finally:
+        # TODO (mengyang, shuwen, chenyang): refactor the exit logic of router and backend.
+        if router is not None:
+            try:
+                asyncio.run(router.client.aclose())
+            except Exception as exc:
+                print(
+                    f"{log_prefix} warning: failed to close router client: {exc}",
+                    file=sys.stderr,
+                    flush=True,
+                )
+        if backend is not None:
+            print(f"{log_prefix} shutting down managed workers...", flush=True)
+            backend.shutdown()
+            print(f"{log_prefix} all managed workers terminated.", flush=True)
 
 
 def build_parser() -> argparse.ArgumentParser:

src/sglang_diffusion_routing/launcher/__init__.py

@@ -0,0 +1,26 @@
+"""Launcher backends for spinning up SGLang diffusion workers.
+
+Right now only supports local backend, which launches workers as local subprocesses.
+We leave this module for future extensions on slurm or kubernetes.
+"""
+
+from sglang_diffusion_routing.launcher.backend import (
+    LaunchedWorker,
+    LauncherBackend,
+    WorkerLaunchResult,
+)
+from sglang_diffusion_routing.launcher.config import (
+    create_backend,
+    load_launcher_config,
+)
+from sglang_diffusion_routing.launcher.local import LocalLauncher, LocalLauncherConfig
+
+__all__ = [
+    "LaunchedWorker",
+    "LauncherBackend",
+    "LocalLauncher",
+    "LocalLauncherConfig",
+    "WorkerLaunchResult",
+    "create_backend",
+    "load_launcher_config",
+]