From 022c9e54565c91b0c88760f347019325a8f59d5b Mon Sep 17 00:00:00 2001 From: kris Date: Mon, 6 Apr 2026 12:47:31 +0800 Subject: [PATCH] feat: surface asr runtime state in workbench --- CHANGELOG.md | 6 ++ collector-service/app/core_main.py | 31 +++++++++- tests/test_production_baseline.py | 57 +++++++++++++++++++ web/storyforge-web-v4/assets/app.js | 31 +++++++++- .../tests/workbench-pages.test.mjs | 18 ++++++ 5 files changed, 141 insertions(+), 2 deletions(-) diff --git a/CHANGELOG.md b/CHANGELOG.md index 4ad8ee4..fb4e867 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -4,6 +4,12 @@ ## 2026-04-06 +### 工作台依赖健康现在会显示 ASR 真实运行模式 + +- `collector` 的 `/v2/integrations/health` 现在会带出 ASR 的 `language_mode / runtime_device_mode / runtime_compute_type_mode / active_device / active_compute_type / model_name`。 +- 工作台里的依赖健康卡不再只是显示 `ASR 在线`，而是会直接展示 `在线 · GPU` 或 `在线 · CPU`，并补充当前模型、语言模式和 compute type。 +- 这样以后排查“Windows ASR 当前到底有没有吃到 GPU”时，不需要再手查命令行或单独打 `/health`。 + ### Windows ASR GPU 失败时自动回退 CPU - Windows `ASR HTTP` 现在在 `auto` 模式下仍会优先尝试 `cuda + int8_float16`，但如果在真正推理阶段命中 `cublas/cudnn/cuda` 运行库缺失，会自动切回 `cpu + int8` 重试，不再把整次转写卡死在 GPU 路径。 diff --git a/collector-service/app/core_main.py b/collector-service/app/core_main.py index 8a02521..8cb35e4 100644 --- a/collector-service/app/core_main.py +++ b/collector-service/app/core_main.py @@ -3152,6 +3152,23 @@ def probe_http(url: str, path: str = "", timeout: float = 3.0) -> dict[str, Any] return tcp +def probe_http_json(url: str, path: str = "", timeout: float = 3.0) -> dict[str, Any]: + detail = probe_http(url, path=path, timeout=timeout) + detail["json"] = {} + if not detail.get("configured") or not detail.get("reachable"): + return detail + target_url = detail.get("url") or (urljoin(url if url.endswith("/") else f"{url}/", path.lstrip("/")) if url else "") + try: + response = httpx.get(target_url or url, timeout=timeout, follow_redirects=True) + if "application/json" in (response.headers.get("content-type") or ""): + payload = response.json() + if isinstance(payload, dict): + detail["json"] = payload + except Exception: + pass + return detail + + def live_recorder_request(method: str, path: str, payload: dict[str, Any] | None = None, timeout: float = 20.0) -> Any: if not LIVE_RECORDER_BASE_URL: raise HTTPException(status_code=503, detail="LIVE_RECORDER_BASE_URL is not configured") @@ -3274,6 +3291,8 @@ def integrations_health(account: dict[str, Any] = Depends(require_approved)) -> _ = account cutvideo_bootstrap = probe_http(CUTVIDEO_BASE_URL, "/api/bootstrap", timeout=5.0) cutvideo_uploads = probe_http(CUTVIDEO_BASE_URL, "/api/uploads", timeout=5.0) + asr_probe = probe_http_json(ASR_HTTP_BASE_URL, "/health", timeout=5.0) + asr_runtime = asr_probe.get("json") if isinstance(asr_probe.get("json"), dict) else {} cutvideo_supports_uploads = bool( cutvideo_uploads.get("configured") and cutvideo_uploads.get("reachable") @@ -3303,7 +3322,17 @@ def integrations_health(account: dict[str, Any] = Depends(require_approved)) -> }, "asr": { "base_url": ASR_HTTP_BASE_URL, - **probe_tcp(ASR_HTTP_BASE_URL), + "configured": asr_probe.get("configured", False), + "reachable": asr_probe.get("reachable", False), + "status_code": int(asr_probe.get("status_code") or 0), + "error": str(asr_probe.get("error") or ""), + "url": str(asr_probe.get("url") or ASR_HTTP_BASE_URL or ""), + "language_mode": str(asr_runtime.get("language") or ""), + "runtime_device_mode": str(asr_runtime.get("device") or ""), + "runtime_compute_type_mode": str(asr_runtime.get("compute_type") or ""), + "active_device": str(asr_runtime.get("active_device") or ""), + "active_compute_type": str(asr_runtime.get("active_compute_type") or ""), + "model_name": str(asr_runtime.get("model_name") or ""), }, "live_recorder": { "base_url": LIVE_RECORDER_BASE_URL, diff --git a/tests/test_production_baseline.py b/tests/test_production_baseline.py index 57cf0aa..364d775 100644 --- a/tests/test_production_baseline.py +++ b/tests/test_production_baseline.py @@ -269,6 +269,63 @@ class ProductionBaselineTests(unittest.TestCase): self.assertIn("cutvideoRouteMode", payload["lanRouting"]) self.assertIn("cutvideoBaseUrl", payload["lanRouting"]) + def test_integrations_health_exposes_asr_runtime_summary(self) -> None: + ctx = self._seed_context("asr_runtime", exhausted=False) + headers = {"Authorization": f"Bearer {ctx['token']}"} + original_base_url = self.core.ASR_HTTP_BASE_URL + original_probe_http_json = getattr(self.core, "probe_http_json", None) + try: + self.core.ASR_HTTP_BASE_URL = "http://asr.example:8088" + + def fake_probe_http_json(url: str, path: str = "", timeout: float = 3.0) -> dict[str, Any]: + if url == "http://asr.example:8088" and path == "/health": + return { + "configured": True, + "reachable": True, + "status_code": 200, + "error": "", + "url": "http://asr.example:8088/health", + "json": { + "service": "storyforge-windows-asr", + "model_name": "base", + "language": "auto", + "device": "auto", + "compute_type": "auto", + "active_device": "cuda", + "active_compute_type": "int8_float16", + }, + } + return { + "configured": False, + "reachable": False, + "status_code": 0, + "error": "not_configured", + "url": "", + "json": {}, + } + + self.core.probe_http_json = fake_probe_http_json + response = self.client.get("/v2/integrations/health", headers=headers) + finally: + self.core.ASR_HTTP_BASE_URL = original_base_url + if original_probe_http_json is None: + try: + delattr(self.core, "probe_http_json") + except AttributeError: + pass + else: + self.core.probe_http_json = original_probe_http_json + + self.assertEqual(response.status_code, 200, response.text) + payload = response.json() + self.assertIn("asr", payload) + self.assertEqual(payload["asr"]["active_device"], "cuda") + self.assertEqual(payload["asr"]["active_compute_type"], "int8_float16") + self.assertEqual(payload["asr"]["runtime_device_mode"], "auto") + self.assertEqual(payload["asr"]["runtime_compute_type_mode"], "auto") + self.assertEqual(payload["asr"]["language_mode"], "auto") + self.assertEqual(payload["asr"]["model_name"], "base") + def test_collector_deploy_script_exposes_health_retry_controls(self) -> None: script_path = ROOT / "scripts" / "deploy_fnos_storyforge_collector.sh" content = script_path.read_text(encoding="utf-8") diff --git a/web/storyforge-web-v4/assets/app.js b/web/storyforge-web-v4/assets/app.js index f89babf..1f4b2e4 100644 --- a/web/storyforge-web-v4/assets/app.js +++ b/web/storyforge-web-v4/assets/app.js @@ -3932,7 +3932,13 @@ function getIntegrationDetail(key) { supportsUploads: raw?.supports_uploads !== undefined ? Boolean(raw?.supports_uploads) : true, uploadStatusCode: Number(raw?.upload_status_code || 0), uploadError: String(raw?.upload_error || ""), - uploadUrl: String(raw?.upload_url || "") + uploadUrl: String(raw?.upload_url || ""), + runtimeDeviceMode: String(raw?.runtime_device_mode || ""), + runtimeComputeTypeMode: String(raw?.runtime_compute_type_mode || ""), + activeDevice: String(raw?.active_device || ""), + activeComputeType: String(raw?.active_compute_type || ""), + languageMode: String(raw?.language_mode || ""), + modelName: String(raw?.model_name || "") }; } @@ -3954,6 +3960,16 @@ function getCutvideoIntegrationUrlLabel(detail) { return isFnosTunnelCutvideo(detail) ? "fnOS NAS 隧道入口" : "Windows 直连"; } +function getAsrRuntimeBadge(detail) { + if (!detail || detail.key !== "asr") return ""; + const activeDevice = String(detail.activeDevice || "").trim().toLowerCase(); + if (activeDevice === "cuda") return "GPU"; + if (activeDevice === "cpu") return "CPU"; + const runtimeMode = String(detail.runtimeDeviceMode || "").trim().toLowerCase(); + if (runtimeMode === "auto") return "自动"; + return runtimeMode ? runtimeMode.toUpperCase() : ""; +} + function getIntegrationStatus(detail) { if (!detail.available) { return { tone: "blue", summary: "未拉取" }; @@ -3962,6 +3978,10 @@ function getIntegrationStatus(detail) { return { tone: "orange", summary: "缺上传能力" }; } if (detail.reachable) { + if (detail.key === "asr") { + const runtimeBadge = getAsrRuntimeBadge(detail); + return { tone: "green", summary: runtimeBadge ? `在线 · ${runtimeBadge}` : "在线" }; + } return { tone: "green", summary: "在线" }; } if (detail.configured) { @@ -4069,6 +4089,15 @@ function getIntegrationCards() { ? `当前通过 fnOS NAS 隧道访问 ${detail.baseUrl || detail.url || "cutvideo"}` : `当前直连 ${detail.baseUrl || detail.url || "cutvideo"}`; } + if (key === "asr") { + const runtimeBadge = getAsrRuntimeBadge(detail) || "待热身"; + const computeLabel = detail.activeComputeType || detail.runtimeComputeTypeMode || "auto"; + const languageLabel = detail.languageMode || "auto"; + extra = `当前转写：${runtimeBadge} · ${computeLabel} · 语言 ${languageLabel}`; + if (detail.modelName) { + extra += ` · 当前模型：${detail.modelName}`; + } + } if (detail.available && !detail.configured && isSuperAdmin()) { actions = [ actions, diff --git a/web/storyforge-web-v4/tests/workbench-pages.test.mjs b/web/storyforge-web-v4/tests/workbench-pages.test.mjs index fed47fe..cd5b1c8 100644 --- a/web/storyforge-web-v4/tests/workbench-pages.test.mjs +++ b/web/storyforge-web-v4/tests/workbench-pages.test.mjs @@ -1477,6 +1477,24 @@ test("live-first workbench flows no longer advertise stale missing-capability pl assert.match(APP, /暂未识别当前动作/); }); +test("integration cards surface ASR runtime mode and model details", () => { + const detailSource = extractBetween(APP, "function getIntegrationDetail(key) {", "function isFnosTunnelCutvideo"); + const statusSource = extractBetween(APP, "function getIntegrationStatus(detail) {", "function describeIntegrationFailure"); + const cardsSource = extractBetween(APP, "function getIntegrationCards()", "function renderLiveRecorderSummaryHtml()"); + + assert.match(APP, /function getAsrRuntimeBadge\(/); + assert.match(detailSource, /runtimeDeviceMode:/); + assert.match(detailSource, /runtimeComputeTypeMode:/); + assert.match(detailSource, /activeDevice:/); + assert.match(detailSource, /activeComputeType:/); + assert.match(detailSource, /languageMode:/); + assert.match(detailSource, /modelName:/); + assert.match(statusSource, /detail\.key === "asr"/); + assert.match(statusSource, /在线 ·/); + assert.match(cardsSource, /当前转写：/); + assert.match(cardsSource, /当前模型：/); +}); + test("smart discovery entrypoints prefer direct execute before falling back to forms", () => { const clicks = extractBetween(APP, "document.addEventListener(\"click\", async (event) => {", "document.addEventListener(\"submit\", async (event) => {"); assert.match(clicks, /name === "open-similar-search"[\s\S]*const account = getSelectedAccount\(\);/);