task-50: 结果明细批量 upsert 写入器与唯一键幂等

CollectDataResultItemBatchWriter 按唯一键 uk_task_scope_item 批量 upsert biz_task_result_item：一次批量查询现有行（payload_hash 相等即跳过），分批发 INSERT ... ON DUPLICATE KEY UPDATE，替代逐行 select/insert/update；旧格式逐行 payload 升级为引用时物理删除（引用计数兜底）。CollectDataService submitResult 接入批量写入，移除逐行 upsertResultItem。
2026-08-30 15:32:55 +08:00
parent 68ebd387be
commit 1127e9c8bb
5 changed files with 419 additions and 84 deletions
@@ -4,7 +4,6 @@ import cn.hutool.core.util.IdUtil;
 import cn.hutool.core.io.FileUtil;
 import cn.hutool.crypto.digest.DigestUtil;
 import com.baomidou.mybatisplus.core.conditions.query.LambdaQueryWrapper;
-import com.baomidou.mybatisplus.core.conditions.update.LambdaUpdateWrapper;
 import com.fasterxml.jackson.core.type.TypeReference;
 import com.fasterxml.jackson.databind.JsonNode;
 import com.fasterxml.jackson.databind.ObjectMapper;
@@ -38,6 +37,7 @@ import com.nanri.aiimage.modules.collectdata.util.CollectDataExtraJsonCodec;
 import com.nanri.aiimage.modules.collectdata.util.CollectDataInvalidAsinBatchWriter;
 import com.nanri.aiimage.modules.collectdata.util.CollectDataParseLimits;
 import com.nanri.aiimage.modules.collectdata.util.CollectDataResultDetailCodec;
+import com.nanri.aiimage.modules.collectdata.util.CollectDataResultItemBatchWriter;
 import com.nanri.aiimage.modules.invalidasin.mapper.InvalidAsinDataMapper;
 import com.nanri.aiimage.modules.file.service.LocalFileStorageService;
 import com.nanri.aiimage.modules.file.service.oss.OssStorageService;
@@ -73,8 +73,6 @@ import org.springframework.dao.DuplicateKeyException;

 import java.io.File;
 import java.io.FileInputStream;
-import java.nio.charset.StandardCharsets;
-import java.security.MessageDigest;
 import java.time.LocalDateTime;
 import java.time.format.DateTimeFormatter;
 import java.util.ArrayList;
@@ -148,6 +146,9 @@ public class CollectDataService {
    /** 结果明细 chunk 级编解码：accepted 行按 chunk 共享一个 RustFS 对象。 */
    private final CollectDataResultDetailCodec resultDetailCodec;

+    /** 结果明细批量 upsert 器：按唯一键 uk_task_scope_item 批量写入，替代逐行 select/insert/update。 */
+    private final CollectDataResultItemBatchWriter resultItemBatchWriter;
+
    @Value("${aiimage.collect-data.stale-timeout-minutes:30}")
    private long staleTimeoutMinutes;

@@ -659,9 +660,10 @@ public class CollectDataService {
                String storedDetail = transientPayloadStorageService.storeResultPayload(
                        MODULE_TYPE, taskId, scopeHash, "chunk-" + chunkIndex, detailJson);
                requireRustfsPayload(storedDetail, "采集结果明细必须写入 RustFS");
-                for (int i = 0; i < accepted.size(); i++) {
-                    upsertResultItem(task.getId(), result.getId(), scopeKey, chunkIndex, accepted.get(i), i, storedDetail);
-                }
+                // 批量 upsert：先一次批量查现有行（payload_hash 相等即跳过，幂等），
+                // 再按唯一键 uk_task_scope_item 分批发 INSERT ... ON DUPLICATE KEY UPDATE。
+                resultItemBatchWriter.upsertAccepted(task.getId(), result.getId(), scopeKey,
+                        chunkIndex, accepted, storedDetail);
            }

            String payloadJson = writeJson(rows, "采集结果序列化失败");
@@ -753,70 +755,6 @@ public class CollectDataService {
        return outcome.accepted();
    }

-    private void upsertResultItem(Long taskId, Long resultId, String scopeKey, int chunkIndex,
-                                  CollectDataResultRowVo row, int offset, String storedDetail) {
-        String itemKey = "asin:" + row.getAsin();
-        String scopeHash = hash(scopeKey);
-        // 引用 JSON 作为 payload_json：内容变化（行 offset/对象变化）即 hash 变化。
-        String refJson = resultDetailCodec.encodeRef(chunkIndex, offset, storedDetail);
-        String payloadHash = hash(refJson);
-        TaskResultItemEntity existing = taskResultItemMapper.selectOne(new LambdaQueryWrapper<TaskResultItemEntity>()
-                .eq(TaskResultItemEntity::getTaskId, taskId)
-                .eq(TaskResultItemEntity::getModuleType, MODULE_TYPE)
-                .eq(TaskResultItemEntity::getScopeHash, scopeHash)
-                .eq(TaskResultItemEntity::getItemKey, itemKey)
-                .last("limit 1"));
-        if (existing != null && Objects.equals(existing.getPayloadHash(), payloadHash)) {
-            return;
-        }
-        LocalDateTime now = LocalDateTime.now();
-        if (existing == null) {
-            TaskResultItemEntity entity = new TaskResultItemEntity();
-            entity.setTaskId(taskId);
-            entity.setModuleType(MODULE_TYPE);
-            entity.setResultId(resultId);
-            entity.setScopeKey(scopeKey);
-            entity.setScopeHash(scopeHash);
-            entity.setItemKey(itemKey);
-            entity.setAsin(row.getAsin());
-            entity.setStatus("ACCEPTED");
-            entity.setPayloadJson(refJson);
-            entity.setPayloadHash(payloadHash);
-            entity.setCreatedAt(now);
-            entity.setUpdatedAt(now);
-            try {
-                taskResultItemMapper.insert(entity);
-                return;
-            } catch (DuplicateKeyException ignored) {
-                existing = taskResultItemMapper.selectOne(new LambdaQueryWrapper<TaskResultItemEntity>()
-                        .eq(TaskResultItemEntity::getTaskId, taskId)
-                        .eq(TaskResultItemEntity::getModuleType, MODULE_TYPE)
-                        .eq(TaskResultItemEntity::getScopeHash, scopeHash)
-                        .eq(TaskResultItemEntity::getItemKey, itemKey)
-                        .last("limit 1"));
-            }
-        }
-        if (existing == null) {
-            throw new BusinessException("保存采集结果明细失败");
-        }
-        // chunk 级引用共享同一 RustFS 对象（deterministic key，同 chunk 重提
-        // 覆盖同一对象，引用 pointer 稳定，无需删除）；旧格式逐行对象在升级
-        // 为引用后不再被任何行持有，直接物理删除，避免泄漏。
-        CollectDataResultDetailCodec.ChunkRef oldRef = resultDetailCodec.parseRef(existing.getPayloadJson());
-        if (oldRef == null) {
-            transientPayloadStorageService.deletePayloadIfPresent(existing.getPayloadJson());
-        }
-        taskResultItemMapper.update(null, new LambdaUpdateWrapper<TaskResultItemEntity>()
-                .eq(TaskResultItemEntity::getId, existing.getId())
-                .set(TaskResultItemEntity::getResultId, resultId)
-                .set(TaskResultItemEntity::getScopeKey, scopeKey)
-                .set(TaskResultItemEntity::getAsin, row.getAsin())
-                .set(TaskResultItemEntity::getStatus, "ACCEPTED")
-                .set(TaskResultItemEntity::getPayloadJson, refJson)
-                .set(TaskResultItemEntity::getPayloadHash, payloadHash)
-                .set(TaskResultItemEntity::getUpdatedAt, now));
-    }
-
    private void persistChunk(Long taskId,
                              String scopeKey,
                              String scopeHash,
@@ -1193,20 +1131,6 @@ public class CollectDataService {
        }
    }

-    private String hash(String value) {
-        try {
-            MessageDigest digest = MessageDigest.getInstance("SHA-256");
-            byte[] bytes = digest.digest((value == null ? "" : value).getBytes(StandardCharsets.UTF_8));
-            StringBuilder sb = new StringBuilder(bytes.length * 2);
-            for (byte b : bytes) {
-                sb.append(String.format("%02x", b));
-            }
-            return sb.toString();
-        } catch (Exception ex) {
-            throw new IllegalStateException("failed to hash collect data payload", ex);
-        }
-    }
-
    private String normalizeAsin(String value) {
        return normalize(value).toUpperCase(Locale.ROOT);
    }
@@ -0,0 +1,140 @@
+package com.nanri.aiimage.modules.collectdata.util;
+
+import com.baomidou.mybatisplus.core.conditions.query.LambdaQueryWrapper;
+import com.nanri.aiimage.modules.collectdata.model.vo.CollectDataResultRowVo;
+import com.nanri.aiimage.modules.task.mapper.TaskResultItemMapper;
+import com.nanri.aiimage.modules.task.model.entity.TaskResultItemEntity;
+import com.nanri.aiimage.modules.task.service.TransientPayloadStorageService;
+import lombok.extern.slf4j.Slf4j;
+import org.springframework.beans.factory.annotation.Value;
+import org.springframework.stereotype.Component;
+
+import java.time.LocalDateTime;
+import java.util.ArrayList;
+import java.util.HashMap;
+import java.util.List;
+import java.util.Map;
+import java.util.Objects;
+
+/**
+ * 结果明细批量 upsert：把整 chunk 的 accepted 行分批发往
+ * biz_task_result_item。先一次批量查询现有行（payload_hash 相等即跳过，
+ * 幂等），再按唯一键 uk_task_scope_item 用一条 INSERT ... ON DUPLICATE
+ * KEY UPDATE 批量写入，替代逐行 select/insert/update；批量失败跳过该批
+ * 可恢复，不中断任务流程。
+ */
+@Slf4j
+@Component
+public class CollectDataResultItemBatchWriter {
+
+    private static final String MODULE_TYPE = "collectdata";
+    private static final int DEFAULT_BATCH_SIZE = 100;
+
+    private final TaskResultItemMapper taskResultItemMapper;
+    private final CollectDataResultDetailCodec resultDetailCodec;
+    private final TransientPayloadStorageService transientPayloadStorageService;
+    private final int batchSize;
+
+    public CollectDataResultItemBatchWriter(TaskResultItemMapper taskResultItemMapper,
+                                            CollectDataResultDetailCodec resultDetailCodec,
+                                            TransientPayloadStorageService transientPayloadStorageService,
+                                            @Value("${aiimage.collect-data.result-item-batch-size:100}") int batchSize) {
+        this.taskResultItemMapper = taskResultItemMapper;
+        this.resultDetailCodec = resultDetailCodec;
+        this.transientPayloadStorageService = transientPayloadStorageService;
+        this.batchSize = batchSize <= 0 ? DEFAULT_BATCH_SIZE : batchSize;
+    }
+
+    /** 批量写入计数：insertedOrUpdated 实际写入行数，skipped hash 相等跳过行数。 */
+    public record UpsertCounts(int insertedOrUpdated, int skipped) {
+    }
+
+    /** 把整 chunk 的 accepted 行批量 upsert；scopeKey 与 chunk 内 offset 已知，仅计算 refJson 与 hash。 */
+    public UpsertCounts upsertAccepted(Long taskId, Long resultId, String scopeKey, int chunkIndex,
+                                       List<CollectDataResultRowVo> rows, String storedDetail) {
+        if (rows == null || rows.isEmpty()) {
+            return new UpsertCounts(0, 0);
+        }
+        String scopeHash = hash(scopeKey);
+        // 一次性取回本 scope 现有行，构建 item_key → 现有行 映射（hash 相等即跳过）。
+        List<TaskResultItemEntity> existingList = taskResultItemMapper.selectList(
+                new LambdaQueryWrapper<TaskResultItemEntity>()
+                        .eq(TaskResultItemEntity::getTaskId, taskId)
+                        .eq(TaskResultItemEntity::getModuleType, MODULE_TYPE)
+                        .eq(TaskResultItemEntity::getScopeHash, scopeHash));
+        Map<String, TaskResultItemEntity> existingByKey = new HashMap<>();
+        for (TaskResultItemEntity existing : existingList) {
+            if (existing != null && existing.getItemKey() != null) {
+                existingByKey.put(existing.getItemKey(), existing);
+            }
+        }
+
+        List<TaskResultItemEntity> toUpsert = new ArrayList<>();
+        int skipped = 0;
+        LocalDateTime now = LocalDateTime.now();
+        for (int offset = 0; offset < rows.size(); offset++) {
+            CollectDataResultRowVo row = rows.get(offset);
+            if (row == null || row.getAsin() == null || row.getAsin().isBlank()) {
+                continue;
+            }
+            String itemKey = "asin:" + row.getAsin();
+            // chunk 级引用共享同一 RustFS 对象（deterministic key，同 chunk 重提
+            // 覆盖同一对象，引用 pointer 稳定，无需删除）。
+            String refJson = resultDetailCodec.encodeRef(chunkIndex, offset, storedDetail);
+            String payloadHash = hash(refJson);
+            TaskResultItemEntity existing = existingByKey.get(itemKey);
+            if (existing != null && Objects.equals(existing.getPayloadHash(), payloadHash)) {
+                skipped++;
+                continue;
+            }
+            // 旧格式逐行对象在升级为引用后不再被任何行持有，直接物理删除避免泄漏
+            // （deletePayloadIfPresent 内部带全局引用计数兜底）。
+            if (existing != null && existing.getPayloadJson() != null
+                    && resultDetailCodec.parseRef(existing.getPayloadJson()) == null) {
+                transientPayloadStorageService.deletePayloadIfPresent(existing.getPayloadJson());
+            }
+            TaskResultItemEntity entity = new TaskResultItemEntity();
+            entity.setId(existing == null ? null : existing.getId());
+            entity.setTaskId(taskId);
+            entity.setModuleType(MODULE_TYPE);
+            entity.setResultId(resultId);
+            entity.setScopeKey(scopeKey);
+            entity.setScopeHash(scopeHash);
+            entity.setItemKey(itemKey);
+            entity.setAsin(row.getAsin());
+            entity.setStatus("ACCEPTED");
+            entity.setPayloadJson(refJson);
+            entity.setPayloadHash(payloadHash);
+            entity.setCreatedAt(existing == null ? now : existing.getCreatedAt());
+            entity.setUpdatedAt(now);
+            toUpsert.add(entity);
+        }
+
+        int written = 0;
+        for (int from = 0; from < toUpsert.size(); from += batchSize) {
+            int to = Math.min(from + batchSize, toUpsert.size());
+            List<TaskResultItemEntity> batch = toUpsert.subList(from, to);
+            try {
+                written += taskResultItemMapper.upsertBatch(batch);
+            } catch (RuntimeException ex) {
+                log.warn("[collect-data] upsert result item batch failed, skip batch {}..{} taskId={}",
+                        from, to, taskId, ex);
+            }
+        }
+        return new UpsertCounts(written, skipped);
+    }
+
+    private String hash(String value) {
+        try {
+            java.security.MessageDigest digest = java.security.MessageDigest.getInstance("SHA-256");
+            byte[] bytes = digest.digest((value == null ? "" : value).getBytes(java.nio.charset.StandardCharsets.UTF_8));
+            StringBuilder sb = new StringBuilder(bytes.length * 2);
+            for (byte b : bytes) {
+                sb.append(String.format("%02x", b));
+            }
+            return sb.toString();
+        } catch (Exception ex) {
+            throw new IllegalStateException("结果明细 hash 计算失败", ex);
+        }
+    }
+}
@@ -2,8 +2,39 @@ package com.nanri.aiimage.modules.task.mapper;

 import com.baomidou.mybatisplus.core.mapper.BaseMapper;
 import com.nanri.aiimage.modules.task.model.entity.TaskResultItemEntity;
+import org.apache.ibatis.annotations.Insert;
 import org.apache.ibatis.annotations.Mapper;
+import org.apache.ibatis.annotations.Param;
+
+import java.util.List;

@Mapper
 public interface TaskResultItemMapper extends BaseMapper<TaskResultItemEntity> {
+
+    /**
+     * 按唯一键 uk_task_scope_item (task_id, module_type, scope_hash, item_key)
+     * 批量 upsert：命中唯一键时更新行内容（含 id 保留），否则插入。
+     */
+    @Insert("""
+            <script>
+            INSERT INTO biz_task_result_item
+                (task_id, module_type, result_id, scope_key, scope_hash, item_key, asin, status,
+                 payload_json, payload_hash, created_at, updated_at)
+            VALUES
+            <foreach collection="rows" item="row" separator=",">
+                (#{row.taskId}, #{row.moduleType}, #{row.resultId}, #{row.scopeKey}, #{row.scopeHash},
+                 #{row.itemKey}, #{row.asin}, #{row.status}, #{row.payloadJson}, #{row.payloadHash},
+                 #{row.createdAt}, #{row.updatedAt})
+            </foreach>
+            ON DUPLICATE KEY UPDATE
+                result_id = VALUES(result_id),
+                scope_key = VALUES(scope_key),
+                asin = VALUES(asin),
+                status = VALUES(status),
+                payload_json = VALUES(payload_json),
+                payload_hash = VALUES(payload_hash),
+                updated_at = VALUES(updated_at)
+            </script>
+            """)
+    int upsertBatch(@Param("rows") List<TaskResultItemEntity> rows);
 }