feat: 复习 Feed 随机取样改用主键区间并固定 actuator 清单

This commit is contained in:
2026-10-09 15:51:50 +08:00
parent cb450b6f47
commit 251a46887c
8 changed files with 262 additions and 6 deletions
@@ -0,0 +1,26 @@
package com.guo.learningprogresstracker.dto;
import lombok.Data;
/**
* 主键区间的聚合结果(复习-03)。
* <p>随机取样需要知道 id 的跨度,才能用「随机起点 + 顺序取窗口」替代 {@code ORDER BY RAND()}。
* 单列聚合可用 {@code selectObjs},这里需要 MIN 与 MAX 两列,因此用显式结果类 + 自定义 SQL。</p>
*/
@Data
public class IdRange {
private Integer minId;
private Integer maxId;
/** 区间内是否存在记录 */
public boolean isEmpty() {
return minId == null || maxId == null;
}
/** 区间宽度(含两端);空区间返回 0 */
public long span() {
return isEmpty() ? 0L : (long) maxId - minId + 1L;
}
}
@@ -1,5 +1,6 @@
package com.guo.learningprogresstracker.mapper;
import com.guo.learningprogresstracker.dto.IdRange;
import com.guo.learningprogresstracker.dto.TaskContentCount;
import com.guo.learningprogresstracker.entity.StudyReportFragmentsEntity;
import com.baomidou.mybatisplus.core.mapper.BaseMapper;
@@ -25,6 +26,12 @@ public interface StudyReportFragmentsMapper extends BaseMapper<StudyReportFragme
* @param taskNums 需要统计的任务编号;为空时返回空列表
*/
List<TaskContentCount> countFragmentsByTask(@Param("taskNums") Collection<String> taskNums);
/**
* 取知识碎片的主键区间(复习-03):供随机取样计算随机起点。
* <p>语句见 {@code resources/mapper/StudyReportFragmentsMapper.xml};无数据时两列为 null。</p>
*/
IdRange selectIdRange();
}
@@ -36,6 +36,11 @@ public class AiServiceClient {
private String url;
/** 超时秒数(等待 AI 任务完成的最长时间) */
private int timeoutSeconds = 600;
/**
* 入站令牌(安全-03):lpt-ai 配置了 LPT_AI_TOKEN 时,本客户端必须携带同一令牌。
* 为空表示不携带,兼容未启用令牌的部署。
*/
private String token;
private HttpClient httpClient;
private final ObjectMapper objectMapper = new ObjectMapper();
@@ -56,6 +61,11 @@ public class AiServiceClient {
return url != null && !url.isBlank();
}
/** 头里必须带值,未配置令牌时给空串(lpt-ai 未启用令牌时会忽略它) */
private String authToken() {
return token == null ? "" : token.trim();
}
// ============ 公开 API(签名不变,内部改为 submit + poll)============
/**
@@ -144,6 +154,7 @@ public class AiServiceClient {
.uri(URI.create(url + "/ai/tasks"))
.timeout(Duration.ofSeconds(30)) // 提交本身不应超时
.header("Content-Type", "application/json")
.header("x-api-token", authToken())
.POST(HttpRequest.BodyPublishers.ofString(objectMapper.writeValueAsString(body)))
.build();
@@ -182,6 +193,7 @@ public class AiServiceClient {
HttpRequest request = HttpRequest.newBuilder()
.uri(URI.create(url + "/ai/tasks/" + taskId))
.timeout(Duration.ofSeconds(10))
.header("x-api-token", authToken())
.GET()
.build();
@@ -1,6 +1,7 @@
package com.guo.learningprogresstracker.service.impl;
import com.baomidou.mybatisplus.core.toolkit.Wrappers;
import com.guo.learningprogresstracker.dto.IdRange;
import com.guo.learningprogresstracker.dto.ReviewFeedItem;
import com.guo.learningprogresstracker.dto.ReviewTaskStats;
import com.guo.learningprogresstracker.dto.TaskContentCount;
@@ -27,6 +28,7 @@ import java.time.DayOfWeek;
import java.time.LocalDate;
import java.time.LocalDateTime;
import java.util.*;
import java.util.concurrent.ThreadLocalRandom;
import java.util.stream.Collectors;
import java.util.stream.Stream;
@@ -59,8 +61,7 @@ public class ReviewServiceImpl implements ReviewService {
boolean random = RANDOM_MODE.equalsIgnoreCase(mode);
List<StudyReportFragmentsEntity> fragments = random
? studyReportFragmentsMapper.selectList(Wrappers.<StudyReportFragmentsEntity>lambdaQuery()
.last("ORDER BY RAND() LIMIT " + safeLimit))
? randomFragments(safeLimit)
: studyReportFragmentsMapper.selectList(Wrappers.<StudyReportFragmentsEntity>lambdaQuery()
.orderByDesc(StudyReportFragmentsEntity::getCreatedTime)
.last("LIMIT " + safeLimit));
@@ -72,6 +73,47 @@ public class ReviewServiceImpl implements ReviewService {
return items.stream().limit(safeLimit).collect(Collectors.toList());
}
/**
* 随机取样知识碎片(复习-03)。
*
* <p>不使用 {@code ORDER BY RAND()}:那在 MySQL 上是全表随机排序,一次 Feed 请求会触发两次。
* 改为「随机起点 + 顺序取窗口」:先取主键区间,随机选一个起点,再按 id 顺序取够数量;
* 不足时从头回绕补齐(id 与创建时间同向,回绕等价于取更早的内容)。删除造成的 id 空洞会让
* 靠后的内容被略微偏重,对「偶遇式」的 Feed 来说可以接受。</p>
*/
private List<StudyReportFragmentsEntity> randomFragments(int limit) {
IdRange range = studyReportFragmentsMapper.selectIdRange();
if (range == null || range.isEmpty()) {
return Collections.emptyList();
}
long start = range.getMinId()
+ (long) (ThreadLocalRandom.current().nextDouble() * range.span());
// 用 LinkedHashMap 去重:回绕补齐时可能取到与首轮相同的行
Map<Integer, StudyReportFragmentsEntity> picked = new LinkedHashMap<>();
collectFrom(start, limit, picked);
if (picked.size() < limit) {
collectFrom(range.getMinId(), limit - picked.size(), picked);
}
return new ArrayList<>(picked.values());
}
/** 从 startId 起按 id 顺序取至多 count 条,按主键去重后并入 picked */
private void collectFrom(long startId, int count, Map<Integer, StudyReportFragmentsEntity> picked) {
if (count <= 0) {
return;
}
List<StudyReportFragmentsEntity> window = studyReportFragmentsMapper.selectList(
Wrappers.<StudyReportFragmentsEntity>lambdaQuery()
.ge(StudyReportFragmentsEntity::getId, startId)
.orderByAsc(StudyReportFragmentsEntity::getId)
.last("LIMIT " + count));
for (StudyReportFragmentsEntity fragment : window) {
picked.putIfAbsent(fragment.getId(), fragment);
}
}
/**
* 智能模式:加权随机采样,让更需要复习的内容有更高概率出现。
* 权重 = 时间衰减(越久未见权重越高) × 任务回忆掌握度(覆盖率越低权重越高)。
@@ -80,9 +122,7 @@ public class ReviewServiceImpl implements ReviewService {
private List<ReviewFeedItem> getSmartFeed(int safeLimit) {
int candidateLimit = Math.min(safeLimit * SMART_CANDIDATE_MULTIPLIER, 500);
List<StudyReportFragmentsEntity> fragments = studyReportFragmentsMapper.selectList(
Wrappers.<StudyReportFragmentsEntity>lambdaQuery()
.last("ORDER BY RAND() LIMIT " + candidateLimit));
List<StudyReportFragmentsEntity> fragments = randomFragments(candidateLimit);
List<ReviewFeedItem> candidates = mergeAndConvert(List.of(), fragments);
if (candidates.size() <= safeLimit) {