whfh/label_backend
0
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

提取功能改为异步实现，添加ai辅助提取状态

Browse Source
This commit is contained in:
wh
2026-04-17 01:20:27 +08:00
parent ccbcfd2c74
commit bf0b00ed08
18 changed files with 594 additions and 386 deletions
Download Patch File Download Diff File Expand all files Collapse all files

166
src/main/java/com/label/common/ai/AiServiceClient.java
View File

@@ -1,5 +1,6 @@
package com.label.common.ai;
import com.fasterxml.jackson.annotation.JsonProperty;
import lombok.Builder;
import lombok.Data;
import org.springframework.beans.factory.annotation.Value;
@@ -36,87 +37,190 @@ public class AiServiceClient {
@Data
@Builder
public static class ExtractionRequest {
private Long sourceId;
public static class TextExtractRequest {
@JsonProperty("file_path")
private String filePath;
private String bucket;
@JsonProperty("file_name")
private String fileName;
private String model;
private String prompt;
@JsonProperty("prompt_template")
private String promptTemplate;
}
@Data
@Builder
public static class ImageExtractRequest {
@JsonProperty("file_path")
private String filePath;
@JsonProperty("task_id")
private Long taskId;
private String model;
@JsonProperty("prompt_template")
private String promptTemplate;
}
@Data
public static class ExtractionResponse {
private List<Map<String, Object>> items; // triple/quadruple items
private String rawOutput;
}
@Data
@Builder
public static class VideoProcessRequest {
private Long sourceId;
public static class ExtractFramesRequest {
@JsonProperty("file_path")
private String filePath;
private String bucket;
private Map<String, Object> params; // frameInterval, mode etc.
@JsonProperty("source_id")
private Long sourceId;
@JsonProperty("job_id")
private Long jobId;
private String mode;
@JsonProperty("frame_interval")
private Integer frameInterval;
}
@Data
@Builder
public static class VideoToTextRequest {
@JsonProperty("file_path")
private String filePath;
@JsonProperty("source_id")
private Long sourceId;
@JsonProperty("job_id")
private Long jobId;
@JsonProperty("start_sec")
private Double startSec;
@JsonProperty("end_sec")
private Double endSec;
private String model;
@JsonProperty("prompt_template")
private String promptTemplate;
}
@Data
public static class TextQaItem {
private String subject;
private String predicate;
private String object;
@JsonProperty("source_snippet")
private String sourceSnippet;
}
@Data
@Builder
public static class GenTextQaRequest {
private List<TextQaItem> items;
private String model;
@JsonProperty("prompt_template")
private String promptTemplate;
}
@Data
public static class ImageQaItem {
private String subject;
private String predicate;
private String object;
private String qualifier;
@JsonProperty("cropped_image_path")
private String croppedImagePath;
}
@Data
@Builder
public static class GenImageQaRequest {
private List<ImageQaItem> items;
private String model;
@JsonProperty("prompt_template")
private String promptTemplate;
}
@Data
public static class QaGenResponse {
private List<Map<String, Object>> qaPairs;
private List<Map<String, Object>> pairs;
}
@Data
@Builder
public static class FinetuneRequest {
private String datasetPath; // RustFS path to JSONL file
private String model;
private Long batchId;
public static class FinetuneStartRequest {
@JsonProperty("jsonl_url")
private String jsonlUrl;
@JsonProperty("base_model")
private String baseModel;
private Map<String, Object> hyperparams;
}
@Data
public static class FinetuneResponse {
public static class FinetuneStartResponse {
@JsonProperty("job_id")
private String jobId;
private String status;
}
@Data
public static class FinetuneStatusResponse {
@JsonProperty("job_id")
private String jobId;
private String status; // PENDING/RUNNING/COMPLETED/FAILED
private Integer progress; // 0-100
@JsonProperty("error_message")
private String errorMessage;
}
// The 8 endpoints:
public ExtractionResponse extractText(ExtractionRequest request) {
return restTemplate.postForObject("/extract/text", request, ExtractionResponse.class);
public ExtractionResponse extractText(TextExtractRequest request) {
return restTemplate.postForObject("/api/v1/text/extract", request, ExtractionResponse.class);
}
public ExtractionResponse extractImage(ExtractionRequest request) {
return restTemplate.postForObject("/extract/image", request, ExtractionResponse.class);
public ExtractionResponse extractImage(ImageExtractRequest request) {
return restTemplate.postForObject("/api/v1/image/extract", request, ExtractionResponse.class);
}
public void extractFrames(VideoProcessRequest request) {
restTemplate.postForLocation("/video/extract-frames", request);
public void extractFrames(ExtractFramesRequest request) {
restTemplate.postForLocation("/api/v1/video/extract-frames", request);
}
public void videoToText(VideoProcessRequest request) {
restTemplate.postForLocation("/video/to-text", request);
public void videoToText(VideoToTextRequest request) {
restTemplate.postForLocation("/api/v1/video/to-text", request);
}
public QaGenResponse genTextQa(ExtractionRequest request) {
return restTemplate.postForObject("/qa/gen-text", request, QaGenResponse.class);
public QaGenResponse genTextQa(GenTextQaRequest request) {
return restTemplate.postForObject("/api/v1/qa/gen-text", request, QaGenResponse.class);
}
public QaGenResponse genImageQa(ExtractionRequest request) {
return restTemplate.postForObject("/qa/gen-image", request, QaGenResponse.class);
public QaGenResponse genImageQa(GenImageQaRequest request) {
return restTemplate.postForObject("/api/v1/qa/gen-image", request, QaGenResponse.class);
}
public FinetuneResponse startFinetune(FinetuneRequest request) {
return restTemplate.postForObject("/finetune/start", request, FinetuneResponse.class);
public FinetuneStartResponse startFinetune(FinetuneStartRequest request) {
return restTemplate.postForObject("/api/v1/finetune/start", request, FinetuneStartResponse.class);
}
public FinetuneStatusResponse getFinetuneStatus(String jobId) {
return restTemplate.getForObject("/finetune/status/{jobId}", FinetuneStatusResponse.class, jobId);
return restTemplate.getForObject("/api/v1/finetune/status/{jobId}", FinetuneStatusResponse.class, jobId);
}
}

14
src/main/java/com/label/common/statemachine/DatasetStatus.java
View File

@@ -1,14 +0,0 @@
package com.label.common.statemachine;
import java.util.Map;
import java.util.Set;
public enum DatasetStatus {
PENDING_REVIEW, APPROVED, REJECTED;
public static final Map<DatasetStatus, Set<DatasetStatus>> TRANSITIONS = Map.of(
PENDING_REVIEW, Set.of(APPROVED, REJECTED),
REJECTED, Set.of(PENDING_REVIEW) // 重新提交审核
// APPROVED: terminal state
);
}

20
src/main/java/com/label/common/statemachine/VideoJobStatus.java
View File

@@ -1,20 +0,0 @@
package com.label.common.statemachine;
import java.util.Map;
import java.util.Set;
public enum VideoJobStatus {
PENDING, RUNNING, SUCCESS, FAILED, RETRYING;
/**
* Automatic state machine transitions.
* Note: FAILED → PENDING is a manual ADMIN operation, handled separately in VideoProcessService.reset().
*/
public static final Map<VideoJobStatus, Set<VideoJobStatus>> TRANSITIONS = Map.of(
PENDING, Set.of(RUNNING),
RUNNING, Set.of(SUCCESS, FAILED, RETRYING),
RETRYING, Set.of(RUNNING, FAILED)
// SUCCESS: terminal state
// FAILED → PENDING: manual ADMIN reset, NOT in this automatic transitions map
);
}

4
src/main/java/com/label/common/statemachine/SourceStatus.java → src/main/java/com/label/common/statemachine/VideoSourceStatus.java
View File

@@ -3,10 +3,10 @@ package com.label.common.statemachine;
import java.util.Map;
import java.util.Set;
public enum SourceStatus {
public enum VideoSourceStatus {
PENDING, PREPROCESSING, EXTRACTING, QA_REVIEW, APPROVED;
public static final Map<SourceStatus, Set<SourceStatus>> TRANSITIONS = Map.of(
public static final Map<VideoSourceStatus, Set<VideoSourceStatus>> TRANSITIONS = Map.of(
PENDING, Set.of(EXTRACTING, PREPROCESSING),
PREPROCESSING, Set.of(PENDING),
EXTRACTING, Set.of(QA_REVIEW),

26
src/main/java/com/label/config/AsyncConfig.java Normal file
View File

@@ -0,0 +1,26 @@
package com.label.config;
import java.util.concurrent.Executor;
import java.util.concurrent.ThreadPoolExecutor;
import org.springframework.context.annotation.Bean;
import org.springframework.context.annotation.Configuration;
import org.springframework.scheduling.annotation.EnableAsync;
import org.springframework.scheduling.concurrent.ThreadPoolTaskExecutor;
@Configuration
@EnableAsync
public class AsyncConfig {
@Bean("aiTaskExecutor")
public Executor aiTaskExecutor() {
ThreadPoolTaskExecutor executor = new ThreadPoolTaskExecutor();
executor.setCorePoolSize(5);
executor.setMaxPoolSize(10);
executor.setQueueCapacity(100);
executor.setThreadNamePrefix("ai-annotate-");
executor.setRejectedExecutionHandler(new ThreadPoolExecutor.CallerRunsPolicy());
executor.initialize();
return executor;
}
}

50
src/main/java/com/label/controller/ExtractionController.java
View File

@@ -1,18 +1,26 @@
package com.label.controller;
import java.util.Map;
import org.springframework.web.bind.annotation.GetMapping;
import org.springframework.web.bind.annotation.PathVariable;
import org.springframework.web.bind.annotation.PostMapping;
import org.springframework.web.bind.annotation.PutMapping;
import org.springframework.web.bind.annotation.RequestBody;
import org.springframework.web.bind.annotation.RequestMapping;
import org.springframework.web.bind.annotation.RestController;
import com.label.annotation.RequireRole;
import com.label.common.auth.TokenPrincipal;
import com.label.common.result.Result;
import com.label.dto.RejectRequest;
import com.label.service.ExtractionService;
import io.swagger.v3.oas.annotations.Operation;
import io.swagger.v3.oas.annotations.Parameter;
import io.swagger.v3.oas.annotations.tags.Tag;
import jakarta.servlet.http.HttpServletRequest;
import lombok.RequiredArgsConstructor;
import org.springframework.web.bind.annotation.*;
import java.util.Map;
/**
* 提取阶段标注工作台接口（5 个端点）。
@@ -25,13 +33,23 @@ public class ExtractionController {
private final ExtractionService extractionService;
/** POST /api/extraction/{taskId}/ai-annotate — AI 辅助预标注 */
@Operation(summary = "AI 辅助预标注", description = "调用 AI 服务自动生成预标注结果，可重复调用")
@PostMapping("/{taskId}/ai-annotate")
@RequireRole("ANNOTATOR")
public Result<Void> aiPreAnnotate(
@Parameter(description = "任务 ID", example = "1001") @PathVariable Long taskId,
HttpServletRequest request) {
extractionService.aiPreAnnotate(taskId, principal(request));
return Result.success(null);
}
/** GET /api/extraction/{taskId} — 获取当前标注结果 */
@Operation(summary = "获取提取标注结果")
@GetMapping("/{taskId}")
@RequireRole("ANNOTATOR")
public Result<Map<String, Object>> getResult(
@Parameter(description = "任务 ID", example = "1001")
@PathVariable Long taskId,
@Parameter(description = "任务 ID", example = "1001") @PathVariable Long taskId,
HttpServletRequest request) {
return Result.success(extractionService.getResult(taskId, principal(request)));
}
@@ -41,12 +59,8 @@ public class ExtractionController {
@PutMapping("/{taskId}")
@RequireRole("ANNOTATOR")
public Result<Void> updateResult(
@Parameter(description = "任务 ID", example = "1001")
@PathVariable Long taskId,
@io.swagger.v3.oas.annotations.parameters.RequestBody(
description = "完整提取标注结果 JSON 字符串，保持原始 JSON body 直接提交",
required = true)
@RequestBody String resultJson,
@Parameter(description = "任务 ID", example = "1001") @PathVariable Long taskId,
@io.swagger.v3.oas.annotations.parameters.RequestBody(description = "完整提取标注结果 JSON 字符串，保持原始 JSON body 直接提交", required = true) @RequestBody String resultJson,
HttpServletRequest request) {
extractionService.updateResult(taskId, resultJson, principal(request));
return Result.success(null);
@@ -57,8 +71,7 @@ public class ExtractionController {
@PostMapping("/{taskId}/submit")
@RequireRole("ANNOTATOR")
public Result<Void> submit(
@Parameter(description = "任务 ID", example = "1001")
@PathVariable Long taskId,
@Parameter(description = "任务 ID", example = "1001") @PathVariable Long taskId,
HttpServletRequest request) {
extractionService.submit(taskId, principal(request));
return Result.success(null);
@@ -69,8 +82,7 @@ public class ExtractionController {
@PostMapping("/{taskId}/approve")
@RequireRole("REVIEWER")
public Result<Void> approve(
@Parameter(description = "任务 ID", example = "1001")
@PathVariable Long taskId,
@Parameter(description = "任务 ID", example = "1001") @PathVariable Long taskId,
HttpServletRequest request) {
extractionService.approve(taskId, principal(request));
return Result.success(null);
@@ -81,12 +93,8 @@ public class ExtractionController {
@PostMapping("/{taskId}/reject")
@RequireRole("REVIEWER")
public Result<Void> reject(
@Parameter(description = "任务 ID", example = "1001")
@PathVariable Long taskId,
@io.swagger.v3.oas.annotations.parameters.RequestBody(
description = "驳回提取结果请求体",
required = true)
@RequestBody RejectRequest body,
@Parameter(description = "任务 ID", example = "1001") @PathVariable Long taskId,
@io.swagger.v3.oas.annotations.parameters.RequestBody(description = "驳回提取结果请求体", required = true) @RequestBody RejectRequest body,
HttpServletRequest request) {
String reason = body != null ? body.getReason() : null;
extractionService.reject(taskId, reason, principal(request));

2
src/main/java/com/label/controller/TaskController.java
View File

@@ -73,7 +73,7 @@ public class TaskController {
/** GET /api/tasks — 查询全部任务（ADMIN） */
@Operation(summary = "管理员查询全部任务")
@GetMapping
@RequireRole("ADMIN")
@RequireRole("ANNOTATOR")
public Result<PageResult<TaskResponse>> getAll(
@Parameter(description = "页码，从 1 开始", example = "1")
@RequestParam(defaultValue = "1") int page,

2
src/main/java/com/label/dto/TaskResponse.java
View File

@@ -24,6 +24,8 @@ public class TaskResponse {
private String status;
@Schema(description = "领取人用户 ID", example = "1")
private Long claimedBy;
@Schema(description = "AI 预标注状态：PENDING/PROCESSING/COMPLETED/FAILED", example = "COMPLETED")
private String aiStatus;
@Schema(description = "领取时间", example = "2026-04-15T12:34:56")
private LocalDateTime claimedAt;
@Schema(description = "提交时间", example = "2026-04-15T12:34:56")

3
src/main/java/com/label/entity/AnnotationTask.java
View File

@@ -53,6 +53,9 @@ public class AnnotationTask {
/** 驳回原因 */
private String rejectReason;
/** AI 预标注状态：PENDING / PROCESSING / COMPLETED / FAILED */
private String aiStatus;
private LocalDateTime createdAt;
private LocalDateTime updatedAt;

125
src/main/java/com/label/listener/ExtractionApprovedEventListener.java
View File

@@ -1,43 +1,28 @@
package com.label.listener;
import java.util.Collections;
import java.util.List;
import java.util.Map;
import org.springframework.beans.factory.annotation.Value;
import com.fasterxml.jackson.databind.ObjectMapper;
import com.label.common.ai.AiServiceClient;
import com.label.common.context.CompanyContext;
import com.label.entity.AnnotationResult;
import com.label.entity.SourceData;
import com.label.entity.TrainingDataset;
import com.label.event.ExtractionApprovedEvent;
import com.label.mapper.AnnotationResultMapper;
import com.label.mapper.SourceDataMapper;
import com.label.mapper.TrainingDatasetMapper;
import com.label.service.TaskService;
import lombok.RequiredArgsConstructor;
import lombok.extern.slf4j.Slf4j;
import org.springframework.stereotype.Component;
import org.springframework.transaction.annotation.Propagation;
import org.springframework.transaction.annotation.Transactional;
import org.springframework.transaction.event.TransactionPhase;
import org.springframework.transaction.event.TransactionalEventListener;
import com.fasterxml.jackson.databind.ObjectMapper;
import com.label.common.ai.AiServiceClient;
import com.label.common.context.CompanyContext;
import com.label.entity.SourceData;
import com.label.entity.TrainingDataset;
import com.label.event.ExtractionApprovedEvent;
import com.label.mapper.SourceDataMapper;
import com.label.mapper.TrainingDatasetMapper;
import com.label.service.TaskService;
import java.util.Collections;
import java.util.List;
import java.util.Map;
import lombok.RequiredArgsConstructor;
import lombok.extern.slf4j.Slf4j;
/**
* 提取审批通过后的异步处理器。
*
* 设计约束（关键）：
* - @TransactionalEventListener(AFTER_COMMIT)：确保在审批事务提交后才触发 AI 调用
* - @Transactional(REQUIRES_NEW)：在独立新事务中写 DB，与审批事务完全隔离
* - 异常不会回滚审批事务（已提交），但会在日志中记录
*
* 处理流程：
* 1. 调用 AI 生成候选问答对（Text/Image 走不同端点）
* 2. 写入 training_dataset（status=PENDING_REVIEW）
* 3. 创建 QA_GENERATION 任务（status=UNCLAIMED）
* 4. 更新 source_data 状态为 QA_REVIEW
*/
@Slf4j
@Component
@RequiredArgsConstructor
@@ -47,23 +32,19 @@ public class ExtractionApprovedEventListener {
private final SourceDataMapper sourceDataMapper;
private final TaskService taskService;
private final AiServiceClient aiServiceClient;
private final AnnotationResultMapper annotationResultMapper;
private final ObjectMapper objectMapper;
@Value("${rustfs.bucket:label-source-data}")
private String bucket;
@TransactionalEventListener(phase = TransactionPhase.AFTER_COMMIT)
@Transactional(propagation = Propagation.REQUIRES_NEW)
public void onExtractionApproved(ExtractionApprovedEvent event) {
log.info("处理提取审批通过事件: taskId={}, sourceId={}", event.getTaskId(), event.getSourceId());
// 设置多租户上下文（新事务中 ThreadLocal 已清除）
CompanyContext.set(event.getCompanyId());
try {
processEvent(event);
} catch (Exception e) {
log.error("处理审批通过事件失败（taskId={}）：{}", event.getTaskId(), e.getMessage(), e);
// 不向上抛出，审批操作已提交，此处失败不回滚审批
log.error("处理审批通过事件失败(taskId={}): {}", event.getTaskId(), e.getMessage(), e);
} finally {
CompanyContext.clear();
}
@@ -76,57 +57,79 @@ public class ExtractionApprovedEventListener {
return;
}
// 1. 调用 AI 生成候选问答对
AiServiceClient.ExtractionRequest req = AiServiceClient.ExtractionRequest.builder()
.sourceId(source.getId())
.filePath(source.getFilePath())
.bucket(bucket)
.build();
List<Map<String, Object>> qaPairs;
try {
AiServiceClient.QaGenResponse response = "IMAGE".equals(source.getDataType())
? aiServiceClient.genImageQa(req)
: aiServiceClient.genTextQa(req);
qaPairs = response != null && response.getQaPairs() != null
? response.getQaPairs()
? aiServiceClient.genImageQa(buildImageQaRequest(event.getTaskId()))
: aiServiceClient.genTextQa(buildTextQaRequest(event.getTaskId()));
qaPairs = response != null && response.getPairs() != null
? response.getPairs()
: Collections.emptyList();
} catch (Exception e) {
log.warn("AI 问答生成失败（taskId={}）：{}，将使用空问答对", event.getTaskId(), e.getMessage());
log.warn("AI 问答生成失败(taskId={}): {}，将使用空问答对", event.getTaskId(), e.getMessage());
qaPairs = Collections.emptyList();
}
// 2. 写入 training_dataset（PENDING_REVIEW）
String sampleType = "IMAGE".equals(source.getDataType()) ? "IMAGE" : "TEXT";
String glmJson = buildGlmJson(qaPairs);
TrainingDataset dataset = new TrainingDataset();
dataset.setCompanyId(event.getCompanyId());
dataset.setTaskId(event.getTaskId());
dataset.setSourceId(event.getSourceId());
dataset.setSampleType(sampleType);
dataset.setGlmFormatJson(glmJson);
dataset.setGlmFormatJson(buildGlmJson(qaPairs));
dataset.setStatus("PENDING_REVIEW");
datasetMapper.insert(dataset);
// 3. 创建 QA_GENERATION 任务（UNCLAIMED）
taskService.createTask(event.getSourceId(), "QA_GENERATION", event.getCompanyId());
// 4. 更新 source_data 状态为 QA_REVIEW
sourceDataMapper.updateStatus(event.getSourceId(), "QA_REVIEW", event.getCompanyId());
log.info("审批通过后续处理完成: taskId={}, 新 QA 任务已创建", event.getTaskId());
log.info("审批通过后续处理完成: taskId={}", event.getTaskId());
}
/**
* 将 AI 生成的问答对列表转换为 GLM fine-tune 格式 JSON。
*/
private String buildGlmJson(List<Map<String, Object>> qaPairs) {
try {
return objectMapper.writeValueAsString(Map.of("conversations", qaPairs));
} catch (Exception e) {
log.error("构建 GLM JSON 失败", e);
log.error("构建微调 JSON 失败", e);
return "{\"conversations\":[]}";
}
}
private AiServiceClient.GenTextQaRequest buildTextQaRequest(Long taskId) {
List<AiServiceClient.TextQaItem> items = readAnnotationItems(taskId).stream()
.map(item -> objectMapper.convertValue(item, AiServiceClient.TextQaItem.class))
.toList();
return AiServiceClient.GenTextQaRequest.builder()
.items(items)
.build();
}
private AiServiceClient.GenImageQaRequest buildImageQaRequest(Long taskId) {
List<AiServiceClient.ImageQaItem> items = readAnnotationItems(taskId).stream()
.map(item -> objectMapper.convertValue(item, AiServiceClient.ImageQaItem.class))
.toList();
return AiServiceClient.GenImageQaRequest.builder()
.items(items)
.build();
}
private List<Map<String, Object>> readAnnotationItems(Long taskId) {
AnnotationResult result = annotationResultMapper.selectByTaskId(taskId);
if (result == null || result.getResultJson() == null || result.getResultJson().isBlank()) {
return Collections.emptyList();
}
try {
@SuppressWarnings("unchecked")
Map<String, Object> parsed = objectMapper.readValue(result.getResultJson(), Map.class);
Object items = parsed.get("items");
if (items instanceof List<?>) {
@SuppressWarnings("unchecked")
List<Map<String, Object>> typedItems = (List<Map<String, Object>>) items;
return typedItems;
}
} catch (Exception e) {
log.warn("解析提取结果失败，taskId={}，将使用空 items: {}", taskId, e.getMessage());
}
return Collections.emptyList();
}
}

5
src/main/java/com/label/mapper/AnnotationResultMapper.java
View File

@@ -33,4 +33,9 @@ public interface AnnotationResultMapper extends BaseMapper<AnnotationResult> {
*/
@Select("SELECT * FROM annotation_result WHERE task_id = #{taskId}")
AnnotationResult selectByTaskId(@Param("taskId") Long taskId);
@Insert("INSERT INTO annotation_result (task_id, company_id, result_json, created_at, updated_at) " +
"VALUES (#{taskId}, #{companyId}, #{resultJson}::jsonb, NOW(), NOW())")
@Options(useGeneratedKeys = true, keyProperty = "id", keyColumn = "id")
int insertWithJsonb(AnnotationResult result);
}

143
src/main/java/com/label/service/AiAnnotationAsyncService.java Normal file
View File

@@ -0,0 +1,143 @@
package com.label.service;
import java.util.Collections;
import java.util.List;
import java.util.Map;
import org.springframework.scheduling.annotation.Async;
import org.springframework.stereotype.Service;
import com.fasterxml.jackson.databind.ObjectMapper;
import com.label.common.ai.AiServiceClient;
import com.label.common.context.CompanyContext;
import com.label.entity.AnnotationResult;
import com.label.entity.AnnotationTask;
import com.label.entity.SourceData;
import com.label.mapper.AnnotationResultMapper;
import com.label.mapper.AnnotationTaskMapper;
import lombok.RequiredArgsConstructor;
import lombok.extern.slf4j.Slf4j;
@Slf4j
@Service
@RequiredArgsConstructor
public class AiAnnotationAsyncService {
private final AnnotationTaskMapper taskMapper;
private final ObjectMapper objectMapper;
private final AnnotationResultMapper resultMapper;
private final AiServiceClient aiServiceClient;
@Async("aiTaskExecutor")
public void processAnnotation(Long taskId, Long companyId, SourceData source) {
CompanyContext.set(companyId);
log.info("开始异步执行 AI 预标注，任务ID: {}", taskId);
String dataType = source.getDataType().toUpperCase();
AiServiceClient.ExtractionResponse aiResponse = null;
int maxRetries = 2;
Exception lastException = null;
String finalStatus = "FAILED";
try {
for (int attempt = 1; attempt <= maxRetries; attempt++) {
try {
if ("IMAGE".equals(dataType)) {
AiServiceClient.ImageExtractRequest req = AiServiceClient.ImageExtractRequest.builder()
.filePath(source.getFilePath())
.taskId(taskId)
.build();
aiResponse = aiServiceClient.extractImage(req);
} else {
AiServiceClient.TextExtractRequest req = AiServiceClient.TextExtractRequest.builder()
.filePath(source.getFilePath())
.fileName(source.getFileName())
.build();
aiResponse = aiServiceClient.extractText(req);
}
if (aiResponse != null) {
log.info("AI 预标注成功，任务ID: {}, 尝试次数: {}", taskId, attempt);
break;
}
} catch (Exception e) {
lastException = e;
log.warn("AI 预标注调用失败（任务 {}），第 {} 次尝试：{}", taskId, attempt, e.getMessage());
if (attempt < maxRetries) {
try {
Thread.sleep(1000L * attempt);
} catch (InterruptedException ie) {
Thread.currentThread().interrupt();
break;
}
}
}
}
List<?> items = Collections.emptyList();
if (aiResponse != null && aiResponse.getItems() != null) {
items = aiResponse.getItems();
}
writeOrUpdateResult(taskId, companyId, items);
finalStatus = "COMPLETED";
} catch (Exception e) {
lastException = e;
log.error("AI 预标注处理过程中发生未知异常，任务ID: {}", taskId, e);
finalStatus = "FAILED";
} finally {
try {
AnnotationTask updateEntity = new AnnotationTask();
updateEntity.setId(taskId);
updateEntity.setAiStatus(finalStatus);
if ("FAILED".equals(finalStatus)) {
String reason = lastException != null ? lastException.getMessage() : "AI处理失败";
if (reason != null && reason.length() > 500) {
reason = reason.substring(0, 500);
}
updateEntity.setRejectReason(reason);
}
int rows = taskMapper.updateById(updateEntity);
log.info("异步 AI 预标注结束，任务ID: {}, 最终状态: {}, row {}", taskId, finalStatus, rows);
} catch (Exception updateEx) {
log.error("更新任务 AI 状态失败，任务ID: {}", taskId, updateEx);
} finally {
CompanyContext.clear();
}
}
}
private void writeOrUpdateResult(Long taskId, Long companyId, List<?> items) {
try {
String json = objectMapper
.writeValueAsString(Map.of("items", items != null ? items : Collections.emptyList()));
int updated = resultMapper.updateResultJson(taskId, json, companyId);
if (updated == 0) {
try {
AnnotationResult result = new AnnotationResult();
result.setTaskId(taskId);
result.setCompanyId(companyId);
result.setResultJson(json);
resultMapper.insertWithJsonb(result);
log.info("新建AI预标注结果，任务ID: {}", taskId);
} catch (Exception insertEx) {
if (insertEx.getMessage() != null && insertEx.getMessage().contains("duplicate key")) {
log.warn("检测到并发插入冲突，转为更新模式，任务ID: {}", taskId);
resultMapper.updateResultJson(taskId, json, companyId);
} else {
throw insertEx;
}
}
} else {
log.info("更新AI预标注结果，任务ID: {}", taskId);
}
} catch (Exception e) {
log.error("写入 AI 预标注结果失败, taskId={}", taskId, e);
throw new RuntimeException("RESULT_WRITE_FAILED: " + e.getMessage(), e);
}
}
}

107
src/main/java/com/label/service/ExtractionService.java
View File

@@ -1,33 +1,30 @@
package com.label.service;
import com.baomidou.mybatisplus.core.conditions.update.LambdaUpdateWrapper;
import com.fasterxml.jackson.databind.ObjectMapper;
import com.label.common.ai.AiServiceClient;
import com.label.common.exception.BusinessException;
import com.label.common.auth.TokenPrincipal;
import com.label.common.statemachine.StateValidator;
import com.label.common.statemachine.TaskStatus;
import com.label.entity.AnnotationResult;
import com.label.entity.TrainingDataset;
import com.label.event.ExtractionApprovedEvent;
import com.label.mapper.AnnotationResultMapper;
import com.label.mapper.TrainingDatasetMapper;
import com.label.entity.SourceData;
import com.label.mapper.SourceDataMapper;
import com.label.entity.AnnotationTask;
import com.label.mapper.AnnotationTaskMapper;
import com.label.service.TaskClaimService;
import lombok.RequiredArgsConstructor;
import lombok.extern.slf4j.Slf4j;
import java.time.LocalDateTime;
import java.util.Map;
import org.springframework.beans.factory.annotation.Value;
import org.springframework.context.ApplicationEventPublisher;
import org.springframework.http.HttpStatus;
import org.springframework.stereotype.Service;
import org.springframework.transaction.annotation.Transactional;
import java.time.LocalDateTime;
import java.util.Collections;
import java.util.Map;
import com.baomidou.mybatisplus.core.conditions.update.LambdaUpdateWrapper;
import com.fasterxml.jackson.databind.ObjectMapper;
import com.label.common.auth.TokenPrincipal;
import com.label.common.exception.BusinessException;
import com.label.common.statemachine.StateValidator;
import com.label.common.statemachine.TaskStatus;
import com.label.entity.AnnotationResult;
import com.label.entity.AnnotationTask;
import com.label.entity.SourceData;
import com.label.event.ExtractionApprovedEvent;
import com.label.mapper.AnnotationResultMapper;
import com.label.mapper.AnnotationTaskMapper;
import com.label.mapper.SourceDataMapper;
import lombok.RequiredArgsConstructor;
import lombok.extern.slf4j.Slf4j;
/**
* 提取阶段标注服务：AI 预标注、更新结果、提交、审批、驳回。
@@ -43,12 +40,13 @@ public class ExtractionService {
private final AnnotationTaskMapper taskMapper;
private final AnnotationResultMapper resultMapper;
private final TrainingDatasetMapper datasetMapper;
// private final TrainingDatasetMapper datasetMapper;
private final SourceDataMapper sourceDataMapper;
private final TaskClaimService taskClaimService;
private final AiServiceClient aiServiceClient;
// private final AiServiceClient aiServiceClient;
private final ApplicationEventPublisher eventPublisher;
private final ObjectMapper objectMapper;
private final AiAnnotationAsyncService aiAnnotationAsyncService; // 注入异步服务
@Value("${rustfs.bucket:label-source-data}")
private String bucket;
@@ -67,32 +65,30 @@ public class ExtractionService {
throw new BusinessException("NOT_FOUND", "关联资料不存在", HttpStatus.NOT_FOUND);
}
// 调用 AI 服务（在事务外，避免长时间持有 DB 连接）
AiServiceClient.ExtractionRequest req = AiServiceClient.ExtractionRequest.builder()
.sourceId(source.getId())
.filePath(source.getFilePath())
.bucket(bucket)
.build();
AiServiceClient.ExtractionResponse aiResponse;
try {
if ("IMAGE".equals(source.getDataType())) {
aiResponse = aiServiceClient.extractImage(req);
} else {
aiResponse = aiServiceClient.extractText(req);
}
} catch (Exception e) {
log.warn("AI 预标注调用失败（任务 {}）：{}", taskId, e.getMessage());
// AI 失败不阻塞流程，写入空结果
aiResponse = new AiServiceClient.ExtractionResponse();
aiResponse.setItems(Collections.emptyList());
if (source.getFilePath() == null || source.getFilePath().isEmpty()) {
throw new BusinessException("INVALID_SOURCE", "源文件路径不能为空", HttpStatus.BAD_REQUEST);
}
// 将 AI 结果写入 annotation_result（UPSERT 语义）
writeOrUpdateResult(taskId, principal.getCompanyId(), aiResponse.getItems());
if (source.getDataType() == null || source.getDataType().isEmpty()) {
throw new BusinessException("INVALID_SOURCE", "数据类型不能为空", HttpStatus.BAD_REQUEST);
}
// ------------------------------------------------------------------ 更新结果 --
String dataType = source.getDataType().toUpperCase();
if (!"IMAGE".equals(dataType) && !"TEXT".equals(dataType)) {
log.warn("不支持的数据类型: {}, 任务ID: {}", dataType, taskId);
throw new BusinessException("UNSUPPORTED_TYPE",
"不支持的数据类型: " + dataType, HttpStatus.BAD_REQUEST);
}
// 更新任务状态为 PROCESSING
taskMapper.update(null, new LambdaUpdateWrapper<AnnotationTask>()
.eq(AnnotationTask::getId, taskId)
.set(AnnotationTask::getAiStatus, "PROCESSING"));
// 触发异步任务
aiAnnotationAsyncService.processAnnotation(taskId, principal.getCompanyId(), source);
// executeAiAnnotationAsync(taskId, principal.getCompanyId(), source);
}
/**
* 人工更新标注结果（整体覆盖，PUT 语义）。
@@ -237,8 +233,7 @@ public class ExtractionService {
"sourceType", source != null ? source.getDataType() : "",
"sourceFilePath", source != null && source.getFilePath() != null ? source.getFilePath() : "",
"isFinal", task.getIsFinal() != null && task.getIsFinal(),
"resultJson", result != null ? result.getResultJson() : "[]"
);
"resultJson", result != null ? result.getResultJson() : "[]");
}
// ------------------------------------------------------------------ 私有工具 --
@@ -253,20 +248,4 @@ public class ExtractionService {
}
return task;
}
private void writeOrUpdateResult(Long taskId, Long companyId, java.util.List<?> items) {
try {
String json = objectMapper.writeValueAsString(Map.of("items", items != null ? items : Collections.emptyList()));
int updated = resultMapper.updateResultJson(taskId, json, companyId);
if (updated == 0) {
AnnotationResult result = new AnnotationResult();
result.setTaskId(taskId);
result.setCompanyId(companyId);
result.setResultJson(json);
resultMapper.insert(result);
}
} catch (Exception e) {
log.error("写入 AI 预标注结果失败: taskId={}", taskId, e);
}
}
}

82
src/main/java/com/label/service/FinetuneService.java
View File

@@ -1,74 +1,73 @@
package com.label.service;
import com.label.common.ai.AiServiceClient;
import com.label.common.exception.BusinessException;
import com.label.common.auth.TokenPrincipal;
import com.label.common.exception.BusinessException;
import com.label.common.storage.RustFsClient;
import com.label.entity.ExportBatch;
import com.label.mapper.ExportBatchMapper;
import lombok.RequiredArgsConstructor;
import lombok.extern.slf4j.Slf4j;
import org.springframework.http.HttpStatus;
import org.springframework.stereotype.Service;
import org.springframework.transaction.annotation.Transactional;
import java.util.Map;
/**
* GLM 微调服务：提交任务、查询状态。
*
* 注意：trigger() 包含 AI HTTP 调用，不在 @Transactional 注解下。
* 仅在 DB 写入时开启事务（updateFinetuneInfo）。
*/
@Slf4j
@Service
@RequiredArgsConstructor
public class FinetuneService {
private static final String FINETUNE_BUCKET = "finetune-export";
private static final int PRESIGNED_URL_MINUTES = 60;
private final ExportBatchMapper exportBatchMapper;
private final ExportService exportService;
private final AiServiceClient aiServiceClient;
private final RustFsClient rustFsClient;
// ------------------------------------------------------------------ 提交微调 --
private String finetuneBaseModel = "qwen3-14b";
/**
* 向 GLM AI 服务提交微调任务。
*
* T074 设计：AI 调用不在 @Transactional 内执行，避免持有 DB 连接期间发起 HTTP 请求。
* DB 写入（updateFinetuneInfo）是单条 UPDATE，不需要显式事务（自动提交）。
* 如果 AI 调用成功但 DB 写入失败，下次查询状态仍可通过 AI 服务的 jobId 重建状态。
*
* @param batchId 批次 ID
* @param principal 当前用户
* @return 包含 glmJobId 和 finetuneStatus 的 Map
*/
public Map<String, Object> trigger(Long batchId, TokenPrincipal principal) {
ExportBatch batch = exportService.getById(batchId, principal);
if (!"NOT_STARTED".equals(batch.getFinetuneStatus())) {
throw new BusinessException("FINETUNE_ALREADY_STARTED",
"微调任务已提交，当前状态: " + batch.getFinetuneStatus(), HttpStatus.CONFLICT);
throw new BusinessException(
"FINETUNE_ALREADY_STARTED",
"微调任务已提交，当前状态 " + batch.getFinetuneStatus(),
HttpStatus.CONFLICT
);
}
// 调用 AI 服务（无事务，不持有 DB 连接）
AiServiceClient.FinetuneRequest req = AiServiceClient.FinetuneRequest.builder()
.datasetPath(batch.getDatasetFilePath())
.model("glm-4")
.batchId(batchId)
String jsonlUrl = rustFsClient.getPresignedUrl(
FINETUNE_BUCKET,
batch.getDatasetFilePath(),
PRESIGNED_URL_MINUTES
);
AiServiceClient.FinetuneStartRequest req = AiServiceClient.FinetuneStartRequest.builder()
.jsonlUrl(jsonlUrl)
.baseModel(finetuneBaseModel)
.hyperparams(Map.of())
.build();
AiServiceClient.FinetuneResponse response;
AiServiceClient.FinetuneStartResponse response;
try {
response = aiServiceClient.startFinetune(req);
} catch (Exception e) {
throw new BusinessException("FINETUNE_TRIGGER_FAILED",
"提交微调任务失败: " + e.getMessage(), HttpStatus.SERVICE_UNAVAILABLE);
throw new BusinessException(
"FINETUNE_TRIGGER_FAILED",
"提交微调任务失败: " + e.getMessage(),
HttpStatus.SERVICE_UNAVAILABLE
);
}
// AI 调用成功后更新批次记录（单条 UPDATE，自动提交）
exportBatchMapper.updateFinetuneInfo(batchId,
response.getJobId(), "RUNNING", principal.getCompanyId());
log.info("微调任务已提交: batchId={}, glmJobId={}", batchId, response.getJobId());
exportBatchMapper.updateFinetuneInfo(
batchId,
response.getJobId(),
"RUNNING",
principal.getCompanyId()
);
return Map.of(
"glmJobId", response.getJobId(),
@@ -76,15 +75,6 @@ public class FinetuneService {
);
}
// ------------------------------------------------------------------ 查询状态 --
/**
* 查询微调任务实时状态（向 AI 服务查询）。
*
* @param batchId 批次 ID
* @param principal 当前用户
* @return 状态 Map
*/
public Map<String, Object> getStatus(Long batchId, TokenPrincipal principal) {
ExportBatch batch = exportService.getById(batchId, principal);
@@ -98,13 +88,11 @@ public class FinetuneService {
);
}
// 向 AI 服务实时查询
AiServiceClient.FinetuneStatusResponse statusResp;
try {
statusResp = aiServiceClient.getFinetuneStatus(batch.getGlmJobId());
} catch (Exception e) {
log.warn("查询微调状态失败（batchId={}）：{}", batchId, e.getMessage());
// 查询失败时返回 DB 中的缓存状态
log.warn("查询微调状态失败(batchId={}): {}", batchId, e.getMessage());
return Map.of(
"batchId", batchId,
"glmJobId", batch.getGlmJobId(),

1
src/main/java/com/label/service/TaskService.java
View File

@@ -190,6 +190,7 @@ public class TaskService {
.sourceId(task.getSourceId())
.taskType(task.getTaskType())
.status(task.getStatus())
.aiStatus(task.getAiStatus())
.claimedBy(task.getClaimedBy())
.claimedAt(task.getClaimedAt())
.submittedAt(task.getSubmittedAt())

202
src/main/java/com/label/service/VideoProcessService.java
View File

@@ -1,17 +1,18 @@
package com.label.service;
import com.baomidou.mybatisplus.core.conditions.update.LambdaUpdateWrapper;
import com.fasterxml.jackson.core.type.TypeReference;
import com.fasterxml.jackson.databind.ObjectMapper;
import com.label.common.ai.AiServiceClient;
import com.label.common.exception.BusinessException;
import com.label.common.statemachine.SourceStatus;
import com.label.common.statemachine.StateValidator;
import com.label.common.statemachine.VideoSourceStatus;
import com.label.entity.SourceData;
import com.label.mapper.SourceDataMapper;
import com.label.entity.VideoProcessJob;
import com.label.mapper.SourceDataMapper;
import com.label.mapper.VideoProcessJobMapper;
import lombok.RequiredArgsConstructor;
import lombok.extern.slf4j.Slf4j;
import org.springframework.beans.factory.annotation.Value;
import org.springframework.http.HttpStatus;
import org.springframework.stereotype.Service;
import org.springframework.transaction.annotation.Transactional;
@@ -21,20 +22,6 @@ import org.springframework.transaction.support.TransactionSynchronizationManager
import java.time.LocalDateTime;
import java.util.Map;
/**
* 视频处理服务：创建任务、处理回调、管理员重置。
*
* 状态流转：
* - 创建时：source_data → PREPROCESSING，job → PENDING
* - 回调成功：job → SUCCESS，source_data → PENDING（进入提取队列）
* - 回调失败（可重试）：job → RETRYING，retryCount++，重新触发 AI
* - 回调失败（超出上限）：job → FAILED，source_data → PENDING
* - 管理员重置：job → PENDING（可手动重新触发）
*
* T074 设计说明：
* AI 调用通过 TransactionSynchronizationManager.registerSynchronization().afterCommit()
* 延迟到事务提交后执行，避免在持有 DB 连接期间进行 HTTP 调用。
*/
@Slf4j
@Service
@RequiredArgsConstructor
@@ -43,44 +30,27 @@ public class VideoProcessService {
private final VideoProcessJobMapper jobMapper;
private final SourceDataMapper sourceDataMapper;
private final AiServiceClient aiServiceClient;
private final ObjectMapper objectMapper;
@Value("${rustfs.bucket:label-source-data}")
private String bucket;
// ------------------------------------------------------------------ 创建任务 --
/**
* 创建视频处理任务并在事务提交后触发 AI 服务。
*
* DB 写入（source_data→PREPROCESSING + 插入 job）在 @Transactional 内完成；
* AI 触发通过 afterCommit() 在事务提交后执行，不占用 DB 连接。
*
* @param sourceId 资料 ID
* @param jobType 任务类型（FRAME_EXTRACT / VIDEO_TO_TEXT）
* @param params JSON 参数（如 {"frameInterval": 30}）
* @param companyId 租户 ID
* @return 新建的 VideoProcessJob
*/
@Transactional
public VideoProcessJob createJob(Long sourceId, String jobType,
String params, Long companyId) {
public VideoProcessJob createJob(Long sourceId, String jobType, String params, Long companyId) {
SourceData source = sourceDataMapper.selectById(sourceId);
if (source == null || !companyId.equals(source.getCompanyId())) {
throw new BusinessException("NOT_FOUND", "资料不存在: " + sourceId, HttpStatus.NOT_FOUND);
throw new BusinessException("NOT_FOUND", "资料不存在 " + sourceId, HttpStatus.NOT_FOUND);
}
validateJobType(jobType);
// source_data → PREPROCESSING
StateValidator.assertTransition(
SourceStatus.TRANSITIONS,
SourceStatus.valueOf(source.getStatus()), SourceStatus.PREPROCESSING);
VideoSourceStatus.TRANSITIONS,
VideoSourceStatus.valueOf(source.getStatus()),
VideoSourceStatus.PREPROCESSING
);
sourceDataMapper.update(null, new LambdaUpdateWrapper<SourceData>()
.eq(SourceData::getId, sourceId)
.set(SourceData::getStatus, "PREPROCESSING")
.set(SourceData::getUpdatedAt, LocalDateTime.now()));
// 插入 PENDING 任务
VideoProcessJob job = new VideoProcessJob();
job.setCompanyId(companyId);
job.setSourceId(sourceId);
@@ -91,48 +61,32 @@ public class VideoProcessService {
job.setMaxRetries(3);
jobMapper.insert(job);
// 事务提交后触发 AI（不在事务内，不占用 DB 连接）
final Long jobId = job.getId();
final String filePath = source.getFilePath();
final String finalJobType = jobType;
final String finalParams = job.getParams();
TransactionSynchronizationManager.registerSynchronization(new TransactionSynchronization() {
@Override
public void afterCommit() {
triggerAi(jobId, sourceId, filePath, finalJobType);
triggerAi(jobId, sourceId, filePath, finalJobType, finalParams);
}
});
log.info("视频处理任务已创建（AI 将在事务提交后触发）: jobId={}, sourceId={}", jobId, sourceId);
log.info("视频处理任务已创建: jobId={}, sourceId={}", jobId, sourceId);
return job;
}
// ------------------------------------------------------------------ 处理回调 --
/**
* 处理 AI 服务异步回调（POST /api/video/callback，无需用户 Token）。
*
* 幂等：若 job 已为 SUCCESS，直接返回，防止重复处理。
* 重试触发同样延迟到事务提交后（afterCommit），不在事务内执行。
*
* @param jobId 任务 ID
* @param callbackStatus AI 回调状态（SUCCESS / FAILED）
* @param outputPath 成功时的输出路径（可选）
* @param errorMessage 失败时的错误信息（可选）
*/
@Transactional
public void handleCallback(Long jobId, String callbackStatus,
String outputPath, String errorMessage) {
// video_process_job 在 IGNORED_TABLES 中（回调无 CompanyContext），此处显式校验
public void handleCallback(Long jobId, String callbackStatus, String outputPath, String errorMessage) {
VideoProcessJob job = jobMapper.selectById(jobId);
if (job == null || job.getCompanyId() == null) {
log.warn("视频处理回调：job 不存在，jobId={}", jobId);
log.warn("视频处理回调时 job 不存在: jobId={}", jobId);
return;
}
// 幂等：已成功则忽略重复回调
if ("SUCCESS".equals(job.getStatus())) {
log.info("视频处理回调幂等：jobId={} 已为 SUCCESS，跳过", jobId);
log.info("视频处理回调幂等跳过: jobId={}", jobId);
return;
}
@@ -143,28 +97,19 @@ public class VideoProcessService {
}
}
// ------------------------------------------------------------------ 管理员重置 --
/**
* 管理员手动重置失败任务（FAILED → PENDING）。
*
* 仅允许 FAILED 状态的任务重置，重置后 retryCount 清零，
* 管理员可随后重新调用 createJob 触发处理。
*
* @param jobId 任务 ID
* @param companyId 租户 ID
*/
@Transactional
public VideoProcessJob reset(Long jobId, Long companyId) {
VideoProcessJob job = jobMapper.selectById(jobId);
if (job == null || !companyId.equals(job.getCompanyId())) {
throw new BusinessException("NOT_FOUND", "视频处理任务不存在: " + jobId, HttpStatus.NOT_FOUND);
throw new BusinessException("NOT_FOUND", "视频处理任务不存在 " + jobId, HttpStatus.NOT_FOUND);
}
if (!"FAILED".equals(job.getStatus())) {
throw new BusinessException("INVALID_TRANSITION",
"只有 FAILED 状态的任务可以重置，当前状态: " + job.getStatus(),
HttpStatus.BAD_REQUEST);
throw new BusinessException(
"INVALID_TRANSITION",
"只有 FAILED 状态的任务可以重置，当前状态 " + job.getStatus(),
HttpStatus.BAD_REQUEST
);
}
jobMapper.update(null, new LambdaUpdateWrapper<VideoProcessJob>()
@@ -176,24 +121,18 @@ public class VideoProcessService {
job.setStatus("PENDING");
job.setRetryCount(0);
log.info("视频处理任务已重置: jobId={}", jobId);
return job;
}
// ------------------------------------------------------------------ 查询 --
public VideoProcessJob getJob(Long jobId, Long companyId) {
VideoProcessJob job = jobMapper.selectById(jobId);
if (job == null || !companyId.equals(job.getCompanyId())) {
throw new BusinessException("NOT_FOUND", "视频处理任务不存在: " + jobId, HttpStatus.NOT_FOUND);
throw new BusinessException("NOT_FOUND", "视频处理任务不存在 " + jobId, HttpStatus.NOT_FOUND);
}
return job;
}
// ------------------------------------------------------------------ 私有方法 --
private void handleSuccess(VideoProcessJob job, String outputPath) {
// job → SUCCESS
jobMapper.update(null, new LambdaUpdateWrapper<VideoProcessJob>()
.eq(VideoProcessJob::getId, job.getId())
.set(VideoProcessJob::getStatus, "SUCCESS")
@@ -201,13 +140,10 @@ public class VideoProcessService {
.set(VideoProcessJob::getCompletedAt, LocalDateTime.now())
.set(VideoProcessJob::getUpdatedAt, LocalDateTime.now()));
// source_data PREPROCESSING → PENDING（进入提取队列）
sourceDataMapper.update(null, new LambdaUpdateWrapper<SourceData>()
.eq(SourceData::getId, job.getSourceId())
.set(SourceData::getStatus, "PENDING")
.set(SourceData::getUpdatedAt, LocalDateTime.now()));
log.info("视频处理成功：jobId={}, sourceId={}", job.getId(), job.getSourceId());
}
private void handleFailure(VideoProcessJob job, String errorMessage) {
@@ -215,7 +151,6 @@ public class VideoProcessService {
int maxRetries = job.getMaxRetries() != null ? job.getMaxRetries() : 3;
if (newRetryCount < maxRetries) {
// 仍有重试次数：job → RETRYING，事务提交后重新触发 AI
jobMapper.update(null, new LambdaUpdateWrapper<VideoProcessJob>()
.eq(VideoProcessJob::getId, job.getId())
.set(VideoProcessJob::getStatus, "RETRYING")
@@ -223,26 +158,22 @@ public class VideoProcessService {
.set(VideoProcessJob::getErrorMessage, errorMessage)
.set(VideoProcessJob::getUpdatedAt, LocalDateTime.now()));
log.warn("视频处理失败，开始第 {} 次重试：jobId={}, error={}",
newRetryCount, job.getId(), errorMessage);
// 重试 AI 触发延迟到事务提交后
SourceData source = sourceDataMapper.selectById(job.getSourceId());
if (source != null) {
final Long jobId = job.getId();
final Long sourceId = job.getSourceId();
final String filePath = source.getFilePath();
final String jobType = job.getJobType();
final String params = job.getParams();
TransactionSynchronizationManager.registerSynchronization(new TransactionSynchronization() {
@Override
public void afterCommit() {
triggerAi(jobId, sourceId, filePath, jobType);
triggerAi(jobId, sourceId, filePath, jobType, params);
}
});
}
} else {
// 超出最大重试次数：job → FAILED，source_data → PENDING
jobMapper.update(null, new LambdaUpdateWrapper<VideoProcessJob>()
.eq(VideoProcessJob::getId, job.getId())
.set(VideoProcessJob::getStatus, "FAILED")
@@ -251,40 +182,87 @@ public class VideoProcessService {
.set(VideoProcessJob::getCompletedAt, LocalDateTime.now())
.set(VideoProcessJob::getUpdatedAt, LocalDateTime.now()));
// source_data PREPROCESSING → PENDING（管理员可重新处理）
sourceDataMapper.update(null, new LambdaUpdateWrapper<SourceData>()
.eq(SourceData::getId, job.getSourceId())
.set(SourceData::getStatus, "PENDING")
.set(SourceData::getUpdatedAt, LocalDateTime.now()));
log.error("视频处理永久失败：jobId={}, sourceId={}, error={}",
job.getId(), job.getSourceId(), errorMessage);
}
}
private void triggerAi(Long jobId, Long sourceId, String filePath, String jobType) {
AiServiceClient.VideoProcessRequest req = AiServiceClient.VideoProcessRequest.builder()
.sourceId(sourceId)
.filePath(filePath)
.bucket(bucket)
.params(Map.of("jobId", jobId, "jobType", jobType))
.build();
private void triggerAi(Long jobId, Long sourceId, String filePath, String jobType, String paramsJson) {
Map<String, Object> params = parseParams(paramsJson);
try {
if ("FRAME_EXTRACT".equals(jobType)) {
aiServiceClient.extractFrames(req);
aiServiceClient.extractFrames(AiServiceClient.ExtractFramesRequest.builder()
.filePath(filePath)
.sourceId(sourceId)
.jobId(jobId)
.mode(stringParam(params, "mode", "interval"))
.frameInterval(intParam(params, "frameInterval", 30))
.build());
} else {
aiServiceClient.videoToText(req);
aiServiceClient.videoToText(AiServiceClient.VideoToTextRequest.builder()
.filePath(filePath)
.sourceId(sourceId)
.jobId(jobId)
.startSec(doubleParam(params, "startSec", 0.0))
.endSec(doubleParam(params, "endSec", 120.0))
.model(stringParam(params, "model", null))
.promptTemplate(stringParam(params, "promptTemplate", null))
.build());
}
log.info("AI 触发成功: jobId={}", jobId);
log.info("AI 视频任务已触发: jobId={}", jobId);
} catch (Exception e) {
log.error("触发视频处理 AI 失败（jobId={}）：{}，job 保持当前状态，需管理员手动重置", jobId, e.getMessage());
log.error("触发视频处理 AI 失败(jobId={}): {}", jobId, e.getMessage());
}
}
private Map<String, Object> parseParams(String paramsJson) {
if (paramsJson == null || paramsJson.isBlank()) {
return Map.of();
}
try {
return objectMapper.readValue(paramsJson, new TypeReference<>() {});
} catch (Exception e) {
log.warn("解析视频处理参数失败，将使用默认值: {}", e.getMessage());
return Map.of();
}
}
private String stringParam(Map<String, Object> params, String key, String defaultValue) {
Object value = params.get(key);
return value == null ? defaultValue : String.valueOf(value);
}
private Integer intParam(Map<String, Object> params, String key, Integer defaultValue) {
Object value = params.get(key);
if (value instanceof Number number) {
return number.intValue();
}
if (value instanceof String text && !text.isBlank()) {
return Integer.parseInt(text);
}
return defaultValue;
}
private Double doubleParam(Map<String, Object> params, String key, Double defaultValue) {
Object value = params.get(key);
if (value instanceof Number number) {
return number.doubleValue();
}
if (value instanceof String text && !text.isBlank()) {
return Double.parseDouble(text);
}
return defaultValue;
}
private void validateJobType(String jobType) {
if (!"FRAME_EXTRACT".equals(jobType) && !"VIDEO_TO_TEXT".equals(jobType)) {
throw new BusinessException("INVALID_JOB_TYPE",
"任务类型不合法，应为 FRAME_EXTRACT 或 VIDEO_TO_TEXT", HttpStatus.BAD_REQUEST);
throw new BusinessException(
"INVALID_JOB_TYPE",
"任务类型不合法，应为 FRAME_EXTRACT 或 VIDEO_TO_TEXT",
HttpStatus.BAD_REQUEST
);
}
}
}

5
src/main/resources/application.yml
View File

@@ -64,8 +64,9 @@ rustfs:
region: us-east-1
ai-service:
base-url: ${AI_SERVICE_BASE_URL:http://http://172.28.77.215:18000}
timeout: 30000
base-url: ${AI_SERVICE_BASE_URL:http://172.28.77.215:18000}
#base-url: ${AI_SERVICE_BASE_URL:http://127.0.0.1:18000}
timeout: 300000
auth:
enabled: true

3
src/main/resources/sql/init.sql
View File

@@ -87,6 +87,7 @@ CREATE TABLE IF NOT EXISTS annotation_task (
completed_at TIMESTAMP,
is_final BOOLEAN NOT NULL DEFAULT FALSE, -- true 即 APPROVED 且无需再审
ai_model VARCHAR(50),
ai_status VARCHAR(20) NOT NULL DEFAULT 'PENDING',
reject_reason TEXT,
created_at TIMESTAMP NOT NULL DEFAULT NOW(),
updated_at TIMESTAMP NOT NULL DEFAULT NOW()
@@ -313,7 +314,7 @@ INSERT INTO sys_config (company_id, config_key, config_value, description)
VALUES
(NULL, 'token_ttl_seconds', '7200',
'会话凭证有效期（秒）'),
(NULL, 'model_default', 'glm-4',
(NULL, 'model_default', 'qwen-plus',
'AI 辅助默认模型'),
(NULL, 'video_frame_interval', '30',
'视频帧提取间隔（帧数）'),