From cba15a62829a0fe57a03c8031f981e8f790e2efc Mon Sep 17 00:00:00 2001 From: tanaka takashi Date: Sat, 12 Sep 2026 18:04:56 +0900 Subject: feat: Stable Diffusion 1.x MCPサーバを実装 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit 仕様書・設計書に基づき、ローカルCUDA GPU上のStable Diffusion 1.x互換モデルをMCP経由で利用するサーバを新規実装。 実装内容: - generate_image、generate_guided_image、inpaint_imageの3ツールをMCP SDK v2で公開 - TOMLからcheckpoint、VAE、複数LoRAと重み、ControlNet、生成設定を読み込み - モデル種別ごとの固定ディレクトリ解決と起動時の設定・ファイル存在検証を追加 - Depth/MiDaS、OpenPose body/full、ControlNet Inpaintの前処理とパイプライン切替を実装 - FP16/CUDA、1枚固定生成、ControlNetオンデマンドロード、単一GPU直列実行を実装 - Base64 PNG/JPEG入力の形式・サイズ検証と任意のdata URI入力に対応 - seed、サイズ、モード、時刻、処理時間をPNG metadataへ記録 - 任意のローカルPNG保存、生成エラー分類、CUDA OOM後のキャッシュ回復を追加 - --configと--versionを備えたstdio CLIエントリポイントを追加 開発・運用: - Python 3.11およびuvによる依存・lock管理を構成 - config.example.tomlとセットアップ・運用手順をREADMEへ追加 - Ruff、Pyright、pytestの設定を追加 - 設定、パス解決、schema、画像検証、サービスルーティング、排他制御、MCP画像返却をテスト 検証結果: - pytest: 30 passed - Ruff lint/format: passed - Pyright: 0 errors, 0 warnings - CLI versionおよびMCPクライアント経由のPNG返却を確認 - 実GPU smoke testはCUDA GPUとモデルファイルがないため対象外 --- tests/test_service.py | 118 ++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 118 insertions(+) create mode 100644 tests/test_service.py (limited to 'tests/test_service.py') diff --git a/tests/test_service.py b/tests/test_service.py new file mode 100644 index 0000000..da82204 --- /dev/null +++ b/tests/test_service.py @@ -0,0 +1,118 @@ +from __future__ import annotations + +import asyncio +import io +import threading +import time +from dataclasses import replace + +import pytest +from PIL import Image + +from mcp_sd15.config import ControlModelConfig, ControlNetConfig +from mcp_sd15.errors import GenerationError, InvalidRequestError +from mcp_sd15.schema import GuidedImageRequest, InpaintImageRequest, TextToImageRequest +from mcp_sd15.service import ImageGenerationService + + +class FakeBackend: + def __init__(self, delay: float = 0) -> None: + self.calls: list[str] = [] + self.delay = delay + self.active = 0 + self.max_active = 0 + self.guard = threading.Lock() + + def _result(self, name: str) -> Image.Image: + with self.guard: + self.active += 1 + self.max_active = max(self.max_active, self.active) + time.sleep(self.delay) + with self.guard: + self.active -= 1 + self.calls.append(name) + return Image.new("RGB", (4, 4), "red") + + def text_to_image(self, request: object, seed: int) -> Image.Image: + return self._result("text") + + def guided_image(self, request: object, image: Image.Image, seed: int) -> Image.Image: + assert image.mode == "RGB" + return self._result("guided") + + def inpaint_image( + self, request: object, image: Image.Image, mask: Image.Image, seed: int + ) -> Image.Image: + assert mask.mode == "L" + return self._result("inpaint") + + +async def test_routes_all_generation_types(app_config, image_base64: str) -> None: + backend = FakeBackend() + service = ImageGenerationService(app_config, backend) + + text = await service.generate_image(TextToImageRequest.create("cat")) + guided = await service.generate_guided_image( + GuidedImageRequest.create("cat", "depth", image_base64) + ) + inpaint = await service.inpaint_image( + InpaintImageRequest.create("cat", image_base64, image_base64) + ) + + assert backend.calls == ["text", "guided", "inpaint"] + for result in (text, guided, inpaint): + assert result.png_data.startswith(b"\x89PNG") + assert isinstance(result.metadata["seed"], int) + + +async def test_generation_is_strictly_serialized(app_config) -> None: + backend = FakeBackend(delay=0.03) + service = ImageGenerationService(app_config, backend) + await asyncio.gather( + *(service.generate_image(TextToImageRequest.create(f"image {i}")) for i in range(4)) + ) + assert backend.max_active == 1 + + +async def test_disabled_controlnet_fails_before_backend(app_config, image_base64: str) -> None: + disabled = replace( + app_config, + controlnet=ControlNetConfig( + depth=ControlModelConfig(False, None, "midas"), + openpose=app_config.controlnet.openpose, + inpaint=app_config.controlnet.inpaint, + ), + ) + service = ImageGenerationService(disabled, FakeBackend()) + with pytest.raises(InvalidRequestError, match="disabled"): + await service.generate_guided_image(GuidedImageRequest.create("cat", "depth", image_base64)) + + +async def test_backend_error_is_sanitized(app_config) -> None: + class BrokenBackend(FakeBackend): + def text_to_image(self, request: object, seed: int) -> Image.Image: + raise RuntimeError("boom") + + service = ImageGenerationService(app_config, BrokenBackend()) + with pytest.raises(GenerationError, match="boom"): + await service.generate_image(TextToImageRequest.create("cat")) + + +async def test_cuda_oom_is_classified(app_config) -> None: + class OomBackend(FakeBackend): + def text_to_image(self, request: object, seed: int) -> Image.Image: + raise RuntimeError("CUDA out of memory while allocating") + + service = ImageGenerationService(app_config, OomBackend()) + with pytest.raises(GenerationError, match="CUDA out of memory"): + await service.generate_image(TextToImageRequest.create("cat")) + + +async def test_optional_output_is_saved(app_config) -> None: + enabled = replace(app_config, output=replace(app_config.output, enabled=True)) + service = ImageGenerationService(enabled, FakeBackend()) + await service.generate_image(TextToImageRequest.create("cat")) + files = list(enabled.output.directory.glob("*.png")) + assert len(files) == 1 + with Image.open(io.BytesIO(files[0].read_bytes())) as image: + assert image.format == "PNG" -- cgit v1.2.3