continue fixing

android/sample-app/src/main/java/cn/learningpad/oraltrainer/sample/MainActivity.kt

@@ -357,6 +357,20 @@ class MainActivity : Activity() {
         }
     }
 
+    private fun applyTranslatedSentences(itemId: String, translations: Map<Int, String>) {
+        if (translations.isEmpty()) return
+        val current = controller.currentTrainingItem() ?: return
+        if (current.id != itemId) return
+        val updated = current.sentences.map { sentence ->
+            sentence.copy(zhText = translations[sentence.index] ?: sentence.zhText)
+        }
+        controller.updateSentences(current.id, updated)
+        screenState.courseBoundaries[current.id]?.let { cached ->
+            screenState.courseBoundaries[current.id] = cached.copy(sentences = updated)
+        }
+        applyChineseSubtitle(controller.currentSentence())
+    }
+
     private fun ensureChineseTranslations() {
         if (screenState.activeModule != Module.TRAIN || !screenState.chineseSubtitlesEnabled) {
             return
@@ -382,18 +396,16 @@ class MainActivity : Activity() {
             item.id,
             expectedIndexes,
             object : SubtitleTranslationCallback {
+                override fun onProgress(result: SubtitleTranslationResult) {
+                    if (!screenState.isCurrentTranslationRequest(generation, item.id)) return
+                    applyTranslatedSentences(item.id, result.translations)
+                }
+
                 override fun onSuccess(result: SubtitleTranslationResult) {
                     if (!screenState.isCurrentTranslationRequest(generation, item.id)) return
+                    applyTranslatedSentences(item.id, result.translations)
                     screenState.chineseTranslationLoading = false
-                    val current = controller.currentTrainingItem() ?: return
-                    if (current.id != item.id) return
-                    val updated = current.sentences.map { sentence ->
-                        sentence.copy(zhText = result.translations[sentence.index] ?: sentence.zhText)
-                    }
-                    controller.updateSentences(current.id, updated)
-                    screenState.courseBoundaries[current.id]?.let { cached ->
-                        screenState.courseBoundaries[current.id] = cached.copy(sentences = updated)
-                    }
+                    screenState.chineseTranslationError = ""
                     applyChineseSubtitle(controller.currentSentence())
                     if (bindings.hasPlayerChrome) {
                         bindings.statusText.text = "已加载中文字幕"

android/sample-app/src/main/java/cn/learningpad/oraltrainer/sample/SubtitleTranslationApi.kt

@@ -3,9 +3,11 @@ package cn.learningpad.oraltrainer.sample
 import android.content.Context
 import android.os.Handler
 import android.os.Looper
+import android.os.SystemClock
 import org.json.JSONObject
 import java.io.IOException
 import java.net.HttpURLConnection
+import java.net.SocketTimeoutException
 import java.net.URL
 import java.util.concurrent.ExecutorService
 import java.util.concurrent.Executors
@@ -14,9 +16,13 @@ import java.util.concurrent.Future
 data class SubtitleTranslationResult(
     val videoHash: String,
     val translations: Map<Int, String>,
+    val complete: Boolean = false,
+    val status: String = "",
+    val error: String = "",
 )
 
 interface SubtitleTranslationCallback {
+    fun onProgress(result: SubtitleTranslationResult) {}
     fun onSuccess(result: SubtitleTranslationResult)
     fun onError(error: Throwable)
 }
@@ -46,7 +52,11 @@ internal class SubtitleTranslationApi(
     ): CancellableTranslationRequest {
         val future = executor.submit {
             try {
-                val result = loadBlocking(videoHash, expectedIndexes)
+                val result = loadBlocking(videoHash, expectedIndexes) { partial ->
+                    if (!Thread.currentThread().isInterrupted) {
+                        mainHandler.post { callback.onProgress(partial) }
+                    }
+                }
                 if (!Thread.currentThread().isInterrupted) {
                     mainHandler.post { callback.onSuccess(result) }
                 }
@@ -66,15 +76,71 @@ internal class SubtitleTranslationApi(
     private fun loadBlocking(
         videoHash: String,
         expectedIndexes: Collection<Int>,
+        onProgress: (SubtitleTranslationResult) -> Unit,
     ): SubtitleTranslationResult {
+        val expected = expectedIndexes.toSet()
+        val deadline = SystemClock.elapsedRealtime() + TRANSLATION_DEADLINE_MS
+        var kickJob = true
+        var lastError: Throwable? = null
+        var lastCount = -1
+        var stalledRounds = 0
+        while (SystemClock.elapsedRealtime() < deadline) {
+            if (Thread.currentThread().isInterrupted) {
+                throw InterruptedException("Subtitle translation cancelled")
+            }
             val stored = runCatching {
                 requestBlocking(videoHash, method = "GET", readTimeoutMs = 15_000)
-        }.getOrNull()
-        val expected = expectedIndexes.toSet()
-        if (stored != null && (expected.isEmpty() || expected.all { stored.translations.containsKey(it) })) {
+            }.onFailure { lastError = it }.getOrNull()
+            if (stored == null) {
+                kickJob = true
+            } else {
+                onProgress(stored)
+                if (isComplete(stored, expected)) {
                     return stored
                 }
-        return requestBlocking(videoHash, method = "POST", readTimeoutMs = 180_000)
+                if (stored.translations.size > lastCount) {
+                    lastCount = stored.translations.size
+                    stalledRounds = 0
+                } else {
+                    stalledRounds += 1
+                }
+                if (stored.status != "running" && stalledRounds >= 2) {
+                    kickJob = true
+                }
+            }
+            if (kickJob) {
+                val posted = runCatching {
+                    requestBlocking(videoHash, method = "POST", readTimeoutMs = 30_000)
+                }.onFailure { error ->
+                    lastError = error
+                    if (error !is SocketTimeoutException) {
+                        stalledRounds += 1
+                    }
+                }.getOrNull()
+                kickJob = false
+                if (posted != null) {
+                    onProgress(posted)
+                    if (isComplete(posted, expected)) {
+                        return posted
+                    }
+                    if (posted.status == "error") {
+                        lastError = IOException(posted.error.ifBlank { "Subtitle translation failed" })
+                    }
+                }
+            }
+            Thread.sleep(POLL_INTERVAL_MS)
+        }
+        throw lastError ?: IOException("Subtitle translation timed out")
+    }
+
+    private fun isComplete(
+        result: SubtitleTranslationResult,
+        expected: Set<Int>,
+    ): Boolean {
+        if (expected.isNotEmpty()) {
+            return expected.all { result.translations.containsKey(it) }
+        }
+        return result.complete
     }
 
     private fun requestBlocking(
@@ -115,9 +181,17 @@ internal class SubtitleTranslationApi(
             return SubtitleTranslationResult(
                 videoHash = root.getString("video_hash"),
                 translations = translations,
+                complete = root.optBoolean("complete", false),
+                status = root.optString("status"),
+                error = root.optString("error"),
             )
         } finally {
             connection.disconnect()
         }
     }
+
+    private companion object {
+        const val POLL_INTERVAL_MS = 1_500L
+        const val TRANSLATION_DEADLINE_MS = 15 * 60 * 1000L
+    }
 }

ios/OralTrainer/Config.swift

@@ -15,7 +15,9 @@ enum AppConfig {
     static let connectTimeout: TimeInterval = 15
     static let readTimeout: TimeInterval = 30
     static let assessmentTimeout: TimeInterval = 180
-    static let translationTimeout: TimeInterval = 180
+    static let translationTimeout: TimeInterval = 30
+    static let translationDeadline: TimeInterval = 900
+    static let translationPollIntervalNs: UInt64 = 1_500_000_000
 
     static let sha256Pattern = #"^[0-9a-fA-F]{64}$"#
 

ios/OralTrainer/Services/SentenceBoundaryAPI.swift

@@ -59,10 +59,24 @@ struct SubtitleTranslationItem: Decodable {
 struct SubtitleTranslationResponse: Decodable {
     let videoHash: String
     let translations: [SubtitleTranslationItem]
+    let complete: Bool?
+    let status: String?
+    let error: String?
 
     private enum CodingKeys: String, CodingKey {
         case translations
         case videoHash = "video_hash"
+        case complete
+        case status
+        case error
+    }
+
+    var mapped: [Int: String] {
+        var values: [Int: String] = [:]
+        for item in translations where !item.zhText.isEmpty {
+            values[item.index] = item.zhText
+        }
+        return values
     }
 }
 
@@ -100,29 +114,90 @@ extension SentenceBoundaryAPI {
             .explanation
     }
 
-    static func fetchStoredSubtitles(videoHash: String) async throws -> [Int: String] {
+    static func fetchStoredSubtitles(videoHash: String) async throws -> SubtitleTranslationResponse {
         try await requestSubtitles(videoHash: videoHash, method: "GET", timeout: AppConfig.readTimeout)
     }
 
-    static func translateSubtitles(videoHash: String) async throws -> [Int: String] {
+    static func translateSubtitles(videoHash: String) async throws -> SubtitleTranslationResponse {
         try await requestSubtitles(videoHash: videoHash, method: "POST", timeout: AppConfig.translationTimeout)
     }
 
-    static func loadSubtitles(videoHash: String, expectedIndexes: [Int]) async throws -> [Int: String] {
-        if let stored = try? await fetchStoredSubtitles(videoHash: videoHash) {
+    static func loadSubtitles(
+        videoHash: String,
+        expectedIndexes: [Int],
+        onProgress: (([Int: String]) -> Void)? = nil
+    ) async throws -> [Int: String] {
         let expected = Set(expectedIndexes)
-            if expected.isEmpty || expected.isSubset(of: Set(stored.keys)) {
-                return stored
+        let deadline = Date().addingTimeInterval(AppConfig.translationDeadline)
+        var kickJob = true
+        var lastError: Error?
+        var lastCount = -1
+        var stalledRounds = 0
+        while Date() < deadline {
+            try Task.checkCancellation()
+            if let stored = try? await fetchStoredSubtitles(videoHash: videoHash) {
+                let mapped = stored.mapped
+                onProgress?(mapped)
+                if isComplete(mapped, expected: expected, payload: stored) {
+                    return mapped
+                }
+                if mapped.count > lastCount {
+                    lastCount = mapped.count
+                    stalledRounds = 0
+                } else {
+                    stalledRounds += 1
+                }
+                if stored.status != "running" && stalledRounds >= 2 {
+                    kickJob = true
+                }
+            } else {
+                kickJob = true
+            }
+            if kickJob {
+                do {
+                    let posted = try await translateSubtitles(videoHash: videoHash)
+                    let mapped = posted.mapped
+                    onProgress?(mapped)
+                    if isComplete(mapped, expected: expected, payload: posted) {
+                        return mapped
+                    }
+                    if posted.status == "error" {
+                        lastError = HTTPClientError.badStatus(503, posted.error ?? "Subtitle translation failed")
+                    }
+                } catch {
+                    lastError = error
+                    if !isTimeout(error) {
+                        stalledRounds += 1
                     }
                 }
-        return try await translateSubtitles(videoHash: videoHash)
+                kickJob = false
+            }
+            try await Task.sleep(nanoseconds: AppConfig.translationPollIntervalNs)
+        }
+        throw lastError ?? HTTPClientError.badStatus(408, "Subtitle translation timed out")
+    }
+
+    private static func isComplete(
+        _ mapped: [Int: String],
+        expected: Set<Int>,
+        payload: SubtitleTranslationResponse
+    ) -> Bool {
+        if !expected.isEmpty {
+            return expected.isSubset(of: Set(mapped.keys))
+        }
+        return payload.complete == true
+    }
+
+    private static func isTimeout(_ error: Error) -> Bool {
+        let nsError = error as NSError
+        return nsError.domain == NSURLErrorDomain && nsError.code == NSURLErrorTimedOut
     }
 
     private static func requestSubtitles(
         videoHash: String,
         method: String,
         timeout: TimeInterval
-    ) async throws -> [Int: String] {
+    ) async throws -> SubtitleTranslationResponse {
         let hash = videoHash.lowercased()
         guard hash.range(of: AppConfig.sha256Pattern, options: .regularExpression) != nil else {
             throw SentenceBoundaryAPIError.invalidHash
@@ -142,11 +217,6 @@ extension SentenceBoundaryAPI {
             let detail = String(data: result.body, encoding: .utf8) ?? ""
             throw HTTPClientError.badStatus(result.statusCode, detail)
         }
-        let payload = try JSONDecoder().decode(SubtitleTranslationResponse.self, from: result.body)
-        var mapped: [Int: String] = [:]
-        for item in payload.translations where !item.zhText.isEmpty {
-            mapped[item.index] = item.zhText
-        }
-        return mapped
+        return try JSONDecoder().decode(SubtitleTranslationResponse.self, from: result.body)
     }
 }

ios/OralTrainer/Views/RootView.swift

@@ -627,6 +627,15 @@ struct RootView: View {
         return "暂无中文字幕"
     }
 
+    @MainActor
+    private func applyChineseTranslations(_ map: [Int: String], mediaId: String) {
+        guard controller.currentItem?.id == mediaId else { return }
+        let updated = (controller.currentItem?.sentences ?? []).map { sentence in
+            sentence.withZhText(map[sentence.index] ?? sentence.zhText)
+        }
+        controller.updateBoundaries(updated)
+    }
+
     private func ensureChineseTranslations() {
         guard showChineseSubtitles, activeModule == .train else {
             return
@@ -656,14 +665,14 @@ struct RootView: View {
                 let map = try await SentenceBoundaryAPI.loadSubtitles(
                     videoHash: mediaId,
                     expectedIndexes: expectedIndexes
-                )
+                ) { partial in
+                    Task { @MainActor in
+                        applyChineseTranslations(partial, mediaId: mediaId)
+                    }
+                }
                 guard !Task.isCancelled else { return }
                 await MainActor.run {
-                    guard controller.currentItem?.id == mediaId else { return }
-                    let updated = (controller.currentItem?.sentences ?? []).map { sentence in
-                        sentence.withZhText(map[sentence.index] ?? sentence.zhText)
-                    }
-                    controller.updateBoundaries(updated)
+                    applyChineseTranslations(map, mediaId: mediaId)
                     chineseTranslationLoading = false
                     chineseTranslationError = ""
                     controller.showTransientMessage("已加载中文字幕")

sentence_api/README.md

@@ -102,11 +102,12 @@ to the mobile clients.
 
 Chinese captions are stored on the course in the server database and shared by
 every user. `GET /api/v1/videos/{sha256}/subtitle-translation` returns the saved
-captions without calling the AI. `POST` generates any missing captions in small
-windows with neighboring context, runs several windows in parallel, retries and
-splits a window if the model response is invalid, and writes each finished
-window immediately so later viewers can reuse it. English sentence edits clear
-the saved Chinese captions for that course.
+captions without calling the AI, including `complete` and `status` so clients can
+poll long videos. `POST` starts missing-caption generation in the background and
+returns quickly; it translates in small windows with neighboring context, runs
+several windows in parallel, retries and splits a window if the model response is
+invalid, and writes each finished window immediately so later viewers can reuse
+it. English sentence edits clear the saved Chinese captions for that course.
 
 ## Assessment
 

sentence_api/main.py

@@ -29,6 +29,7 @@ from fastapi.staticfiles import StaticFiles
 from starlette.concurrency import run_in_threadpool
 from starlette.requests import ClientDisconnect
 from .ai_tutor import AITutor, AITutorError
+from .translation_jobs import SubtitleTranslationService
 
 from .assessment import AssessmentService
 from .auth import AuthenticatedUser
@@ -45,7 +46,6 @@ from .models import (
     SentenceBoundaryDocument,
     SentenceExplanationRequest,
     SentenceExplanationResponse,
-    SubtitleTranslationItem,
     SubtitleTranslationResponse,
     PasswordResetRequest,
     SentenceTextUpdate,
@@ -129,6 +129,10 @@ def create_app(
     application.state.oss_uploader = oss
     application.state.video_processor = processor
     application.state.assessment_service = assessment_service
+    application.state.subtitle_translations = SubtitleTranslationService(
+        video_repository,
+        lambda: application.state.ai_tutor,
+    )
 
     static_dir = Path(__file__).resolve().parent / "static"
     if static_dir.is_dir():
@@ -338,14 +342,7 @@ def create_app(
     def stored_subtitle_translations(
         document: SentenceBoundaryDocument,
     ) -> SubtitleTranslationResponse:
-        return SubtitleTranslationResponse(
-            video_hash=document.video_hash,
-            translations=[
-                SubtitleTranslationItem(index=sentence.index, zh_text=sentence.zh_text)
-                for sentence in document.sentences
-                if sentence.zh_text and sentence.zh_text.strip()
-            ],
-        )
+        return application.state.subtitle_translations.snapshot(document)
 
     @application.get(
         "/api/v1/videos/{video_hash}/subtitle-translation",
@@ -356,10 +353,10 @@ def create_app(
         user: AuthenticatedUser = Depends(current_user),
     ) -> SubtitleTranslationResponse:
         del user
-        document = video_repository.get_document(video_hash)
-        if document is None:
+        snapshot = application.state.subtitle_translations.current(video_hash)
+        if snapshot is None:
             raise HTTPException(status_code=404, detail="Course is not available.")
-        return stored_subtitle_translations(document)
+        return snapshot
 
     @application.post(
         "/api/v1/videos/{video_hash}/subtitle-translation",
@@ -370,52 +367,14 @@ def create_app(
         user: AuthenticatedUser = Depends(current_user),
     ) -> SubtitleTranslationResponse:
         del user
-        document = video_repository.get_document(video_hash)
-        if document is None:
-            raise HTTPException(status_code=404, detail="Course is not available.")
-        payload = [
-            (sentence.index, sentence.text or "")
-            for sentence in document.sentences
-            if (sentence.text or "").strip()
-        ]
-        if not payload:
-            raise HTTPException(status_code=422, detail="No sentence text is available to translate.")
-        cached = stored_subtitle_translations(document).translations
-        complete = len(cached) == len(payload) and all(
-            (sentence.zh_text or "").strip()
-            for sentence in document.sentences
-            if (sentence.text or "").strip()
-        )
-        if complete:
-            return stored_subtitle_translations(document)
-        video = video_repository.get_video(video_hash)
-        existing = {
-            item.index: item.zh_text
-            for item in cached
-        }
         try:
-            translated = application.state.ai_tutor.translate_sentences(
-                payload,
-                course_title=(video["title"] if video is not None else "") or "",
-                existing=existing,
-                persist=lambda part: video_repository.save_sentence_translations(
-                    video_hash,
-                    part,
-                ),
-            )
+            return application.state.subtitle_translations.ensure(video_hash)
+        except KeyError:
+            raise HTTPException(status_code=404, detail="Course is not available.") from None
+        except ValueError as exc:
+            raise HTTPException(status_code=422, detail=str(exc)) from exc
         except AITutorError as exc:
             raise HTTPException(status_code=503, detail=str(exc)) from exc
-        saved = video_repository.save_sentence_translations(video_hash, translated)
-        if saved is None:
-            raise HTTPException(status_code=404, detail="Course is not available.")
-        return SubtitleTranslationResponse(
-            video_hash=saved.video_hash,
-            translations=[
-                SubtitleTranslationItem(index=sentence.index, zh_text=sentence.zh_text)
-                for sentence in saved.sentences
-                if sentence.zh_text and sentence.zh_text.strip()
-            ],
-        )
 
     @application.post(
         "/api/v1/audio/sentence-boundaries",

sentence_api/models.py

@@ -139,6 +139,9 @@ class SubtitleTranslationResponse(BaseModel):
 
     video_hash: str
     translations: List[SubtitleTranslationItem]
+    complete: bool = True
+    status: str = "ready"
+    error: Optional[str] = None
 
 
 class SentenceBoundaryAdjust(BaseModel):

sentence_api/tests/test_api.py

@@ -1,4 +1,6 @@
 import json
+import threading
+import time
 from dataclasses import replace
 
 from fastapi.testclient import TestClient
@@ -14,6 +16,18 @@ from sentence_api.store import BoundaryStore
 
 VIDEO_HASH = "a" * 64
 
+def wait_for_subtitle_translation(client, video_hash, timeout=5.0):
+    deadline = time.time() + timeout
+    last = None
+    while time.time() < deadline:
+        last = client.get(f"/api/v1/videos/{video_hash}/subtitle-translation")
+        if last.status_code == 200 and last.json().get("complete"):
+            return last
+        time.sleep(0.05)
+    raise AssertionError(f"subtitle translation did not complete: {getattr(last, 'text', last)}")
+
+
+
 
 def make_client(tmp_path, admin_api_key=""):
     index_path = tmp_path / "boundaries.json"
@@ -172,7 +186,7 @@ def test_sentence_explanation_uses_server_side_ai_client(tmp_path, monkeypatch):
         return "这句话表示打招呼。"
 
     class StubAITutor:
-        pass
+        available = True
 
     stub = StubAITutor()
     stub.explain = explain
@@ -244,7 +258,7 @@ def test_subtitle_translation_uses_full_document_and_is_cached(tmp_path, monkeyp
         return {index: f"中文{index}" for index, _text in sentences}
 
     class StubAITutor:
-        pass
+        available = True
 
     stub = StubAITutor()
     stub.translate_sentences = translate_sentences
@@ -252,8 +266,12 @@ def test_subtitle_translation_uses_full_document_and_is_cached(tmp_path, monkeyp
 
     response = client.post(f"/api/v1/videos/{VIDEO_HASH}/subtitle-translation")
     assert response.status_code == 200, response.text
+    if not response.json().get("complete"):
+        response = wait_for_subtitle_translation(client, VIDEO_HASH)
     payload = response.json()
     assert payload["video_hash"] == VIDEO_HASH
+    assert payload["complete"] is True
+    assert payload["status"] == "ready"
     assert payload["translations"] == [
         {"index": 0, "zh_text": "中文0"},
         {"index": 1, "zh_text": "中文1"},
@@ -263,7 +281,8 @@ def test_subtitle_translation_uses_full_document_and_is_cached(tmp_path, monkeyp
 
     second = client.post(f"/api/v1/videos/{VIDEO_HASH}/subtitle-translation")
     assert second.status_code == 200
-    assert second.json() == payload
+    assert second.json()["translations"] == payload["translations"]
+    assert second.json()["complete"] is True
     assert len(requested) == 1
     assert requested[0][0] == [(0, "Hello."), (1, "Nice to meet you.")]
 
@@ -278,7 +297,7 @@ def test_subtitle_translation_is_shared_with_other_users(tmp_path, monkeypatch):
         return {index: f"中文{index}" for index, _text in sentences}
 
     class StubAITutor:
-        pass
+        available = True
 
     stub = StubAITutor()
     stub.translate_sentences = translate_sentences
@@ -286,10 +305,13 @@ def test_subtitle_translation_is_shared_with_other_users(tmp_path, monkeypatch):
 
     first = client.post(f"/api/v1/videos/{VIDEO_HASH}/subtitle-translation")
     assert first.status_code == 200, first.text
+    if not first.json().get("complete"):
+        first = wait_for_subtitle_translation(client, VIDEO_HASH)
 
     stored = client.get(f"/api/v1/videos/{VIDEO_HASH}/subtitle-translation")
     assert stored.status_code == 200
-    assert stored.json() == first.json()
+    assert stored.json()["translations"] == first.json()["translations"]
+    assert stored.json()["complete"] is True
 
     other = TestClient(client.app)
     signup = other.post(
@@ -311,7 +333,61 @@ def test_subtitle_translation_is_shared_with_other_users(tmp_path, monkeypatch):
 
     reused = other.post(f"/api/v1/videos/{VIDEO_HASH}/subtitle-translation")
     assert reused.status_code == 200
-    assert reused.json() == first.json()
+    assert reused.json()["translations"] == first.json()["translations"]
+    assert reused.json()["complete"] is True
+    assert len(requested) == 1
+
+
+
+def test_subtitle_translation_does_not_block_http_on_long_jobs(tmp_path, monkeypatch):
+    client = make_client(tmp_path)
+    _ready_video(client.app.state.video_repository)
+    started = threading.Event()
+    release = threading.Event()
+    requested = []
+
+    def translate_sentences(sentences, **kwargs):
+        requested.append(sentences)
+        persist = kwargs.get("persist")
+        if persist is not None:
+            persist({0: "中文0"})
+        started.set()
+        assert release.wait(timeout=5)
+        if persist is not None:
+            persist({1: "中文1"})
+        return {0: "中文0", 1: "中文1"}
+
+    class StubAITutor:
+        available = True
+
+    stub = StubAITutor()
+    stub.translate_sentences = translate_sentences
+    monkeypatch.setattr(client.app.state, "ai_tutor", stub)
+    client.app.state.subtitle_translations.ai_tutor = stub
+    client.app.state.subtitle_translations.post_wait_seconds = 0.2
+
+    response = client.post(f"/api/v1/videos/{VIDEO_HASH}/subtitle-translation")
+    assert started.wait(timeout=2)
+    assert response.status_code == 200, response.text
+    payload = response.json()
+    assert payload["complete"] is False
+    assert payload["status"] == "running"
+    if {"index": 0, "zh_text": "中文0"} not in payload["translations"]:
+        payload = client.get(f"/api/v1/videos/{VIDEO_HASH}/subtitle-translation").json()
+    assert {"index": 0, "zh_text": "中文0"} in payload["translations"]
+
+    stored = client.get(f"/api/v1/videos/{VIDEO_HASH}/subtitle-translation")
+    assert stored.status_code == 200
+    assert stored.json()["status"] == "running"
+    assert stored.json()["complete"] is False
+
+    release.set()
+    finished = wait_for_subtitle_translation(client, VIDEO_HASH)
+    assert finished.json()["complete"] is True
+    assert finished.json()["translations"] == [
+        {"index": 0, "zh_text": "中文0"},
+        {"index": 1, "zh_text": "中文1"},
+    ]
     assert len(requested) == 1
 
 

sentence_api/translation_jobs.py

@@ -0,0 +1,181 @@
+"""Background subtitle translation so long videos do not block HTTP."""
+
+from __future__ import annotations
+
+import logging
+import threading
+from dataclasses import dataclass
+
+from .ai_tutor import AITutor, AITutorError
+from .models import SentenceBoundaryDocument, SubtitleTranslationItem, SubtitleTranslationResponse
+from .repository import VideoRepository
+
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class _Job:
+    video_hash: str
+    done: threading.Event
+    lock: threading.Lock
+    running: bool = True
+    error: str | None = None
+    thread: threading.Thread | None = None
+
+
+class SubtitleTranslationService:
+    def __init__(
+        self,
+        repository: VideoRepository,
+        ai_tutor: AITutor,
+        *,
+        post_wait_seconds: float = 8.0,
+    ) -> None:
+        self.repository = repository
+        self._ai_tutor = ai_tutor
+        self.post_wait_seconds = post_wait_seconds
+        self._lock = threading.Lock()
+        self._jobs: dict[str, _Job] = {}
+
+    @property
+    def ai_tutor(self) -> AITutor:
+        value = self._ai_tutor
+        return value() if callable(value) else value
+
+    @ai_tutor.setter
+    def ai_tutor(self, value) -> None:
+        self._ai_tutor = value
+
+    def snapshot(self, document: SentenceBoundaryDocument) -> SubtitleTranslationResponse:
+        translations = [
+            SubtitleTranslationItem(index=sentence.index, zh_text=sentence.zh_text)
+            for sentence in document.sentences
+            if sentence.zh_text and sentence.zh_text.strip()
+        ]
+        needed = [
+            sentence
+            for sentence in document.sentences
+            if (sentence.text or "").strip()
+        ]
+        complete = bool(needed) and all((sentence.zh_text or "").strip() for sentence in needed)
+        job = self._jobs.get(document.video_hash)
+        running = False
+        error: str | None = None
+        if job is not None:
+            with job.lock:
+                running = job.running
+                error = job.error
+        if complete:
+            status = "ready"
+            error = None
+        elif running:
+            status = "running"
+        elif error:
+            status = "error"
+        else:
+            status = "idle"
+        return SubtitleTranslationResponse(
+            video_hash=document.video_hash,
+            translations=translations,
+            complete=complete,
+            status=status,
+            error=error,
+        )
+
+    def current(self, video_hash: str) -> SubtitleTranslationResponse | None:
+        document = self.repository.get_document(video_hash)
+        if document is None:
+            return None
+        return self.snapshot(document)
+
+    def ensure(self, video_hash: str) -> SubtitleTranslationResponse:
+        if not getattr(self.ai_tutor, "available", False):
+            raise AITutorError("AI translation is not configured.")
+        document = self.repository.get_document(video_hash)
+        if document is None:
+            raise KeyError(video_hash)
+        current = self.snapshot(document)
+        if current.complete:
+            return current
+        payload = [
+            (sentence.index, sentence.text or "")
+            for sentence in document.sentences
+            if (sentence.text or "").strip()
+        ]
+        if not payload:
+            raise ValueError("No sentence text is available to translate.")
+        job = self._start_if_needed(video_hash)
+        wait_seconds = max(0.0, float(self.post_wait_seconds))
+        if wait_seconds > 0:
+            job.done.wait(timeout=wait_seconds)
+        latest = self.current(video_hash)
+        if latest is None:
+            raise KeyError(video_hash)
+        return latest
+
+    def _start_if_needed(self, video_hash: str) -> _Job:
+        with self._lock:
+            existing = self._jobs.get(video_hash)
+            if existing is not None:
+                with existing.lock:
+                    still_running = existing.running or (
+                        existing.thread is not None and existing.thread.is_alive()
+                    )
+                if still_running:
+                    return existing
+            job = _Job(
+                video_hash=video_hash,
+                done=threading.Event(),
+                lock=threading.Lock(),
+            )
+            thread = threading.Thread(
+                target=self._run_job,
+                args=(job,),
+                name=f"subtitle-translate-{video_hash[:8]}",
+                daemon=True,
+            )
+            job.thread = thread
+            self._jobs[video_hash] = job
+            thread.start()
+            return job
+
+    def _run_job(self, job: _Job) -> None:
+        try:
+            document = self.repository.get_document(job.video_hash)
+            if document is None:
+                raise AITutorError("Course is not available.")
+            payload = [
+                (sentence.index, sentence.text or "")
+                for sentence in document.sentences
+                if (sentence.text or "").strip()
+            ]
+            if not payload:
+                return
+            existing = {
+                sentence.index: sentence.zh_text
+                for sentence in document.sentences
+                if sentence.zh_text and sentence.zh_text.strip()
+            }
+            video = self.repository.get_video(job.video_hash)
+            translated = self.ai_tutor.translate_sentences(
+                payload,
+                course_title=(video["title"] if video is not None else "") or "",
+                existing=existing,
+                persist=lambda part: self.repository.save_sentence_translations(
+                    job.video_hash,
+                    part,
+                ),
+            )
+            if translated:
+                self.repository.save_sentence_translations(job.video_hash, translated)
+            with job.lock:
+                job.error = None
+        except Exception as exc:  # noqa: BLE001 - surface job failure to clients
+            logger.exception("Subtitle translation job failed: video=%s", job.video_hash)
+            with job.lock:
+                job.error = str(exc)
+        finally:
+            with job.lock:
+                job.running = False
+            job.done.set()