feat: 새로운 게시글 스크래핑 후 역색인 및 BM25업데이트, 비동기 및 재시도

- BM25 업데이트 작업을 비동기 메서드로 분리하여 실행 (`updateBM25Async`) - 실패 시 최대 3회 재시도하도록 @retryable 적용, 재시도 간격 2초 설정 - CompletableFuture.runAsync() 사용하여 비동기 실행 후 thenRun()으로 후속 처리 추가
inpink · Nov 9, 2024 · 8d805f6 · 8d805f6
1 parent dedb84b
commit 8d805f6
Show file tree

Hide file tree

Showing 9 changed files with 155 additions and 43 deletions.
diff --git a/build.gradle b/build.gradle
@@ -38,6 +38,8 @@ dependencies {
 	implementation 'org.json:json:20190722'
 	implementation 'org.springframework.boot:spring-boot-starter-data-elasticsearch'
 	implementation 'net.sourceforge.tess4j:tess4j:4.5.4'
+	implementation 'org.springframework.retry:spring-retry'
+	implementation 'org.springframework:spring-aspects'
 
 	testImplementation 'junit:junit:4.13.1'
 	compileOnly 'org.projectlombok:lombok'

diff --git a/src/main/java/knusearch/clear/ClearApplication.java b/src/main/java/knusearch/clear/ClearApplication.java
@@ -3,10 +3,12 @@
 import org.springframework.boot.SpringApplication;
 import org.springframework.boot.autoconfigure.SpringBootApplication;
 import org.springframework.context.ApplicationContext;
+import org.springframework.retry.annotation.EnableRetry;
 import org.springframework.scheduling.annotation.EnableScheduling;
 
 @SpringBootApplication
 @EnableScheduling
+@EnableRetry
 public class ClearApplication {
 
     public static void main(String[] args) throws Exception {

diff --git a/src/main/java/knusearch/clear/jpa/repository/post/PostTermRepository.java b/src/main/java/knusearch/clear/jpa/repository/post/PostTermRepository.java
@@ -1,9 +1,7 @@
 package knusearch.clear.jpa.repository.post;
 
-import java.util.List;
 import knusearch.clear.jpa.domain.post.PostTerm;
 import org.springframework.data.jpa.repository.JpaRepository;
-import org.springframework.data.jpa.repository.Query;
 import org.springframework.stereotype.Repository;
 
 @Repository

diff --git a/src/main/java/knusearch/clear/jpa/service/ScrapingService.java b/src/main/java/knusearch/clear/jpa/service/ScrapingService.java
@@ -1,6 +1,7 @@
 package knusearch.clear.jpa.service;
 
 
+import jakarta.persistence.Tuple;
 import java.time.LocalDate;
 import java.time.format.DateTimeFormatter;
 import java.util.ArrayList;
@@ -14,6 +15,7 @@
 import knusearch.clear.jpa.domain.site.Board;
 import knusearch.clear.jpa.domain.site.Site;
 import knusearch.clear.jpa.repository.post.BasePostRepository;
+import knusearch.clear.jpa.service.post.CheckPostResult;
 import knusearch.clear.util.ImageDownloader;
 import knusearch.clear.util.OCRProcessor;
 import lombok.RequiredArgsConstructor;
@@ -240,35 +242,44 @@ public String cutString(String text, int byteSize) {
         return text;
     }
 
-    public void scrapeYesterdayPosts(Site site) {
+    public List<BasePost> scrapeYesterdayPosts(Site site) {
+        List<BasePost> basePosts = new ArrayList<>();
         String baseUrl = site.getBaseUrl();
         List<Board> boards = site.getBoards();
 
         final LocalDate yesterday = LocalDate.now().minusDays(1);
 
         for (Board board : boards) {
             String postUrl = board.getEncryptedName();
-            savePostsWithinPeriod(baseUrl, postUrl, yesterday);
+            List<BasePost> newPosts = savePostsWithinPeriod(baseUrl, postUrl, yesterday);
+            basePosts.addAll(newPosts);
         }
+
+        return basePosts;
     }
 
     @Transactional
-    public void savePostsWithinPeriod(String baseUrl, String postUrl, LocalDate yesterday) {
+    public List<BasePost> savePostsWithinPeriod(String baseUrl, String postUrl, LocalDate yesterday) {
+        List<BasePost> newPosts = new ArrayList<>();
         int pageIdx = 1;
         boolean isTimeToBreak = false;
         while (isTimeToBreak) {
             Elements links = getAllLinksFromOnePage(baseUrl, postUrl, pageIdx);
-            isTimeToBreak = checkWithinPeriodAndSave(baseUrl, postUrl, yesterday, links);
+            CheckPostResult checkPostResult = checkWithinPeriodAndSave(baseUrl, postUrl, yesterday, links);
+            isTimeToBreak = checkPostResult.isShouldBreak();
+            newPosts.addAll(checkPostResult.getNewPosts());
             pageIdx++;
         }
+        return newPosts;
     }
 
-    public boolean checkWithinPeriodAndSave(
+    private CheckPostResult checkWithinPeriodAndSave(
         String baseUrl,
         String postUrl,
         LocalDate yesterday,
         Elements links
     ) {
+        List<BasePost> newPosts = new ArrayList<>();
         for (Element linkElement : links) {
             BasePost basePost = setURLValues(linkElement, baseUrl, postUrl);
 
@@ -277,11 +288,12 @@ public boolean checkWithinPeriodAndSave(
 
             LocalDate dateTime = basePost.getDateTime();
             if (dateTime.isBefore(yesterday)) {
-                return true;
+                return new CheckPostResult(true, newPosts);
             }
 
             basePostRepository.save(basePost);
+            newPosts.add(basePost);
         }
-        return false;
+        return new CheckPostResult(false, newPosts);
     }
 }
diff --git a/src/main/java/knusearch/clear/jpa/service/post/BM25Service.java b/src/main/java/knusearch/clear/jpa/service/post/BM25Service.java
@@ -11,10 +11,15 @@
 import knusearch.clear.jpa.repository.post.BasePostRepository;
 import knusearch.clear.jpa.repository.post.PostTermRepository;
 import lombok.Getter;
+import lombok.extern.slf4j.Slf4j;
+import org.springframework.retry.annotation.Backoff;
+import org.springframework.retry.annotation.Retryable;
 import org.springframework.stereotype.Service;
+import org.springframework.transaction.annotation.Transactional;
 
 @Service
 @Getter
+@Slf4j
 public class BM25Service {
 
     private double k1 = 1.5;  // BM25 조정 파라미터
@@ -29,10 +34,11 @@ public class BM25Service {
 
     private List<BasePost> documents;
 
-    private BasePostRepository basePostRepository;
+    private final BasePostService basePostService;
     private final PostTermRepository postTermRepository;
 
-    public BM25Service(BasePostRepository basePostRepository, PostTermRepository postTermRepository) {
+    public BM25Service(BasePostRepository basePostRepository, PostTermRepository postTermRepository,
+        BasePostService basePostService) {
         List<BasePost> documents = basePostRepository.findAll();
 
         List<PostTerm> postTerms = postTermRepository.findAll();
@@ -42,19 +48,13 @@ public BM25Service(BasePostRepository basePostRepository, PostTermRepository pos
         }
 
         postTerms.clear();
-        postTerms = null;
 
         this.totalDocs = documents.size();
         this.avgDocLength = calculateAvgDocLength(documents);  // 평균 문서 길이 캐싱
         this.docFreqs = calculateDocFreqs(documents);  // 단어별 문서 빈도 계산
         this.postTermRepository = postTermRepository;
         this.documents = basePostRepository.findAll();
-
-//        System.out.println("totalDocs: " + totalDocs);
-//        System.out.println("avgDocLength: " + avgDocLength);
-
-//        System.out.println("docWords: " + docWords.get(1L));
-//        System.out.println("docWords: " + docWords.get(2L));
+        this.basePostService = basePostService;
     }
 
     // 문서의 평균 길이 계산 (캐싱해둠)
@@ -96,10 +96,13 @@ public double calculateBM25(BasePost doc, List<String> query) {
 
         for (String word : query) {
             int termFreq = termFrequency(word, doc);  // 단어 빈도 계산
-            if (termFreq == 0) continue;  // 문서에 단어가 없으면 건너뜀
+            if (termFreq == 0) {
+                continue;  // 문서에 단어가 없으면 건너뜀
+            }
 
             double idf = getCachedIDF(word);  // IDF 캐싱 사용
-            score += idf * ((termFreq * (k1 + 1)) / (termFreq + k1 * (1 - b + b * (docLength / avgDocLength))));
+            score += idf * ((termFreq * (k1 + 1)) / (termFreq + k1 * (1 - b + b * (docLength
+                / avgDocLength))));
         }
         // 시간 가중치를 BM25 점수에 곱해서 반영
         return score * timeWeight;
@@ -114,10 +117,13 @@ public double calculateBM25WithAi(BasePost doc, List<String> query,
 
         for (String word : query) {
             int termFreq = termFrequency(word, doc);  // 단어 빈도 계산
-            if (termFreq == 0) continue;  // 문서에 단어가 없으면 건너뜀
+            if (termFreq == 0) {
+                continue;  // 문서에 단어가 없으면 건너뜀
+            }
 
             double idf = getCachedIDF(word);  // IDF 캐싱 사용
-            score += idf * ((termFreq * (k1 + 1)) / (termFreq + k1 * (1 - b + b * (docLength / avgDocLength))));
+            score += idf * ((termFreq * (k1 + 1)) / (termFreq + k1 * (1 - b + b * (docLength
+                / avgDocLength))));
         }
         // 시간 가중치를 BM25 점수에 곱해서 반영
 
@@ -138,7 +144,8 @@ private int termFrequency(String word, BasePost doc) {
     // 캐싱된 문서 길이 반환
     private double getCachedDocLength(BasePost doc) {
 //        System.out.println(doc.getId()+ "docLengthCache: " + docLengthCache.get(doc.getId()));
-        return docLengthCache.computeIfAbsent(doc.getId(), id -> (double) doc.getContent().length());
+        return docLengthCache.computeIfAbsent(doc.getId(),
+            id -> (double) doc.getContent().length());
     }
 
     // 캐싱된 IDF 값 반환
@@ -154,4 +161,19 @@ private double calculateIDF(String term) {
         return idf;
     }
 
+    @Retryable(
+        value = {BM25UpdateException.class},
+        maxAttempts = 3,
+        backoff = @Backoff(delay = 2000)
+    )
+    @Transactional
+    public void updateIndex(List<BasePost> basePosts) {
+        try {
+            documents.addAll(basePosts);
+            basePostService.tokenizeAndSaveBasePostTerms(basePosts);
+        } catch (Exception e) {
+            log.error("BM25 update failed, retrying...", e);
+            throw new BM25UpdateException("BM25 update failed", e);
+        }
+    }
 }
diff --git a/src/main/java/knusearch/clear/jpa/service/post/BM25UpdateException.java b/src/main/java/knusearch/clear/jpa/service/post/BM25UpdateException.java
@@ -0,0 +1,7 @@
+package knusearch.clear.jpa.service.post;
+
+public class BM25UpdateException extends RuntimeException {
+    public BM25UpdateException(String message, Throwable cause) {
+        super(message, cause);
+    }
+}