Merge f489ea2f59 into 37673c82da

(fix)(headless) add empty check for metric alias to prevent incorrect DataFormatType assignment

assembly/bin/supersonic-build.bat

@@ -43,10 +43,26 @@ if "%service%"=="webapp" (
    call mvn -f %projectDir% clean package -DskipTests -Dspotless.skip=true
    IF ERRORLEVEL 1 (
       ECHO Failed to build backend Java modules.
+      ECHO Please check Maven and Java versions are compatible.
+      ECHO Current Java: %JAVA_HOME%
+      ECHO Current Maven: %MAVEN_HOME%
       EXIT /B 1
    )
+   
+   REM extract and copy files to deployment directory
+   cd %projectDir%\launchers\%model_name%\target
+   if exist "launchers-%model_name%-%MVN_VERSION%-bin.tar.gz" (
+       echo "Extracting launchers-%model_name%-%MVN_VERSION%-bin.tar.gz..."
+       tar -xf "launchers-%model_name%-%MVN_VERSION%-bin.tar.gz"
+       if exist "launchers-%model_name%-%MVN_VERSION%" (
+           echo "Copying files to deployment directory..."
+           xcopy /E /Y "launchers-%model_name%-%MVN_VERSION%\*" "%buildDir%\supersonic-%model_name%-%MVN_VERSION%\"
+       )
+   )
+   
    copy /y %projectDir%\launchers\%model_name%\target\*.tar.gz %buildDir%\
    echo "finished building supersonic-%model_name% service"
+   cd %baseDir%
    goto :EOF
 
 
@@ -72,22 +88,55 @@ if "%service%"=="webapp" (
    cd %buildDir%
    if exist %release_dir% rmdir /s /q %release_dir%
    if exist %release_dir%.zip del %release_dir%.zip
-   mkdir %release_dir%
+   
-   rem package webapp
+   rem check if release directory already exists from buildJavaService
-   tar xvf supersonic-webapp.tar.gz
+   if exist %release_dir% (
-   move /y supersonic-webapp webapp
+       echo "Release directory already prepared by buildJavaService"
-   echo {"env": ""} > webapp\supersonic.config.json
+   ) else (
-   move /y webapp %release_dir%
+       mkdir %release_dir%
-   rem package java service
+       
-   tar xvf %service_name%-bin.tar.gz
+       rem package java service
-   for /d %%D in ("%service_name%\*") do (
+       tar xvf %service_name%-bin.tar.gz 2>nul
-       move "%%D" "%release_dir%"
+       if errorlevel 1 (
+           echo "Warning: tar command failed, trying PowerShell extraction..."
+           powershell -Command "Expand-Archive -Path '%service_name%-bin.tar.gz' -DestinationPath '.' -Force"
+       )
+       for /d %%D in ("%service_name%\*") do (
+           move "%%D" "%release_dir%"
+       )
+       rmdir /s /q %service_name% 2>nul
    )
+   
+   rem package webapp
+   if exist supersonic-webapp.tar.gz (
+       tar xvf supersonic-webapp.tar.gz 2>nul
+       if errorlevel 1 (
+           echo "Warning: tar command failed, trying PowerShell extraction..."
+           powershell -Command "Expand-Archive -Path 'supersonic-webapp.tar.gz' -DestinationPath '.' -Force"
+       )
+       move /y supersonic-webapp webapp
+       echo {"env": ""} > webapp\supersonic.config.json
+       move /y webapp %release_dir%
+       del supersonic-webapp.tar.gz 2>nul
+   )
+   
+   rem verify deployment structure
+   if exist "%release_dir%\lib\launchers-%model_name%-%MVN_VERSION%.jar" (
+       echo "Deployment structure verified successfully"
+   ) else (
+       echo "Warning: Main jar file not found in deployment structure"
+       echo "Expected: %release_dir%\lib\launchers-%model_name%-%MVN_VERSION%.jar"
+   )
+   
    rem generate zip file
-   powershell Compress-Archive -Path %release_dir% -DestinationPath %release_dir%.zip
+   powershell -Command "Compress-Archive -Path '%release_dir%' -DestinationPath '%release_dir%.zip' -Force"
-   del %service_name%-bin.tar.gz
+   if errorlevel 1 (
-   del supersonic-webapp.tar.gz
+       echo "Warning: PowerShell compression failed, release directory still available: %release_dir%"
-   rmdir /s /q %service_name%
+   ) else (
+       echo "Successfully created release package: %release_dir%.zip"
+   )
+   
+   del %service_name%-bin.tar.gz 2>nul
    echo "finished packaging supersonic release"
    goto :EOF
 

auth/authentication/src/main/java/com/tencent/supersonic/auth/authentication/adaptor/DefaultUserAdaptor.java

@@ -19,6 +19,7 @@ import lombok.extern.slf4j.Slf4j;
 import org.springframework.beans.BeanUtils;
 
 import java.sql.Timestamp;
+import java.util.Date;
 import java.util.List;
 import java.util.Optional;
 import java.util.Set;
@@ -222,8 +223,9 @@ public class DefaultUserAdaptor implements UserAdaptor {
                 new UserWithPassword(userDO.getId(), userDO.getName(), userDO.getDisplayName(),
                         userDO.getEmail(), userDO.getPassword(), userDO.getIsAdmin());
 
-        String token =
+        // 使用令牌名称作为生成key ，这样可以区分正常请求和api 请求，api 的令牌失效时间很长，需考虑令牌泄露的情况
-                tokenService.generateToken(UserWithPassword.convert(userWithPassword), expireTime);
+        String token = tokenService.generateToken(UserWithPassword.convert(userWithPassword),
+                "SysDbToken:" + name, (new Date().getTime() + expireTime));
         UserTokenDO userTokenDO = saveUserToken(name, userName, token, expireTime);
         return convertUserToken(userTokenDO);
     }

auth/authentication/src/main/java/com/tencent/supersonic/auth/authentication/persistence/repository/UserRepository.java

@@ -21,6 +21,8 @@ public interface UserRepository {
 
     UserTokenDO getUserToken(Long tokenId);
 
+    UserTokenDO getUserTokenByName(String tokenName);
+
     void deleteUserTokenByName(String userName);
 
     void deleteUserToken(Long tokenId);

auth/authentication/src/main/java/com/tencent/supersonic/auth/authentication/persistence/repository/impl/UserRepositoryImpl.java

@@ -65,6 +65,13 @@ public class UserRepositoryImpl implements UserRepository {
         return userTokenDOMapper.selectById(tokenId);
     }
 
+    @Override
+    public UserTokenDO getUserTokenByName(String tokenName) {
+        QueryWrapper<UserTokenDO> queryWrapper = new QueryWrapper<>();
+        queryWrapper.lambda().eq(UserTokenDO::getName, tokenName);
+        return userTokenDOMapper.selectOne(queryWrapper);
+    }
+
     @Override
     public void deleteUserTokenByName(String userName) {
         QueryWrapper<UserTokenDO> queryWrapper = new QueryWrapper<>();

auth/authentication/src/main/java/com/tencent/supersonic/auth/authentication/utils/TokenService.java

@@ -6,7 +6,10 @@ import javax.crypto.spec.SecretKeySpec;
 
 import com.tencent.supersonic.auth.api.authentication.config.AuthenticationConfig;
 import com.tencent.supersonic.auth.api.authentication.pojo.UserWithPassword;
+import com.tencent.supersonic.auth.authentication.persistence.dataobject.UserTokenDO;
+import com.tencent.supersonic.auth.authentication.persistence.repository.UserRepository;
 import com.tencent.supersonic.common.pojo.exception.AccessException;
+import com.tencent.supersonic.common.util.ContextUtils;
 import io.jsonwebtoken.Claims;
 import io.jsonwebtoken.Jwts;
 import io.jsonwebtoken.SignatureAlgorithm;
@@ -71,6 +74,7 @@ public class TokenService {
         return generateToken(UserWithPassword.convert(appUser), request);
     }
 
+
     public Optional<Claims> getClaims(HttpServletRequest request) {
         String token = request.getHeader(authenticationConfig.getTokenHttpHeaderKey());
         String appKey = getAppKey(request);
@@ -90,6 +94,14 @@ public class TokenService {
 
     public Optional<Claims> getClaims(String token, String appKey) {
         try {
+            if (StringUtils.isNotBlank(appKey) && appKey.startsWith("SysDbToken:")) {// 如果是配置的长期令牌，需校验数据库是否存在该配置
+                UserRepository userRepository = ContextUtils.getBean(UserRepository.class);
+                UserTokenDO dbToken =
+                        userRepository.getUserTokenByName(appKey.substring("SysDbToken:".length()));
+                if (dbToken == null || !dbToken.getToken().equals(token.replace("Bearer ", ""))) {
+                    throw new AccessException("Token does not exist :" + appKey);
+                }
+            }
             String tokenSecret = getTokenSecret(appKey);
             Claims claims =
                     Jwts.parser().setSigningKey(tokenSecret.getBytes(StandardCharsets.UTF_8))
@@ -122,6 +134,16 @@ public class TokenService {
         Map<String, String> appKeyToSecretMap = authenticationConfig.getAppKeyToSecretMap();
         String secret = appKeyToSecretMap.get(appKey);
         if (StringUtils.isBlank(secret)) {
+            if (StringUtils.isNotBlank(appKey) && appKey.startsWith("SysDbToken:")) { // 是配置的长期令牌
+                String realAppKey = appKey.substring("SysDbToken:".length());
+                String tmp =
+                        "WIaO9YRRVt+7QtpPvyWsARFngnEcbaKBk783uGFwMrbJBaochsqCH62L4Kijcb0sZCYoSsiKGV/zPml5MnZ3uQ==";
+                if (tmp.length() <= realAppKey.length()) {
+                    return realAppKey;
+                } else {
+                    return realAppKey + tmp.substring(realAppKey.length());
+                }
+            }
             throw new AccessException("get secret from appKey failed :" + appKey);
         }
         return secret;

chat/server/src/main/java/com/tencent/supersonic/chat/server/executor/SqlExecutor.java

@@ -75,8 +75,12 @@ public class SqlExecutor implements ChatQueryExecutor {
             return null;
         }
 
-        QuerySqlReq sqlReq =
+        // 使用querySQL，它已经包含了所有修正（包括物理SQL修正）
-                QuerySqlReq.builder().sql(parseInfo.getSqlInfo().getCorrectedS2SQL()).build();
+        String finalSql = StringUtils.isNotBlank(parseInfo.getSqlInfo().getQuerySQL())
+                ? parseInfo.getSqlInfo().getQuerySQL()
+                : parseInfo.getSqlInfo().getCorrectedS2SQL();
+
+        QuerySqlReq sqlReq = QuerySqlReq.builder().sql(finalSql).build();
         sqlReq.setSqlInfo(parseInfo.getSqlInfo());
         sqlReq.setDataSetId(parseInfo.getDataSetId());
 
@@ -90,7 +94,7 @@ public class SqlExecutor implements ChatQueryExecutor {
         queryResult.setQueryTimeCost(System.currentTimeMillis() - startTime);
         if (queryResp != null) {
             queryResult.setQueryAuthorization(queryResp.getQueryAuthorization());
-            queryResult.setQuerySql(queryResp.getSql());
+            queryResult.setQuerySql(finalSql);
             queryResult.setQueryResults(queryResp.getResultList());
             queryResult.setQueryColumns(queryResp.getColumns());
             queryResult.setQueryState(QueryState.SUCCESS);

chat/server/src/main/java/com/tencent/supersonic/chat/server/processor/execute/DataInterpretProcessor.java

@@ -47,7 +47,8 @@ public class DataInterpretProcessor implements ExecuteResultProcessor {
         Agent agent = executeContext.getAgent();
         ChatApp chatApp = agent.getChatAppConfig().get(APP_KEY);
         return Objects.nonNull(chatApp) && chatApp.isEnable()
-                && StringUtils.isNotBlank(executeContext.getResponse().getTextResult()); // 如果都没结果，则无法处理，直接跳过
+                && StringUtils.isNotBlank(executeContext.getResponse().getTextResult()) // 如果都没结果，则无法处理
+                && StringUtils.isBlank(executeContext.getResponse().getTextSummary()); // 如果已经有汇总的结果了，无法再次处理
     }
 
     @Override
@@ -57,7 +58,16 @@ public class DataInterpretProcessor implements ExecuteResultProcessor {
         ChatApp chatApp = agent.getChatAppConfig().get(APP_KEY);
 
         Map<String, Object> variable = new HashMap<>();
-        variable.put("question", executeContext.getRequest().getQueryText());
+        String question = executeContext.getResponse().getTextResult();// 结果解析应该用改写的问题，因为改写的内容信息量更大
+        if (executeContext.getParseInfo().getProperties() != null
+                && executeContext.getParseInfo().getProperties().containsKey("CONTEXT")) {
+            Map<String, Object> context = (Map<String, Object>) executeContext.getParseInfo()
+                    .getProperties().get("CONTEXT");
+            if (context.get("queryText") != null && "".equals(context.get("queryText"))) {
+                question = context.get("queryText").toString();
+            }
+        }
+        variable.put("question", question);
         variable.put("data", queryResult.getTextResult());
 
         Prompt prompt = PromptTemplate.from(chatApp.getPrompt()).apply(variable);

chat/server/src/main/java/com/tencent/supersonic/chat/server/service/impl/AgentServiceImpl.java

@@ -22,6 +22,7 @@ import lombok.extern.slf4j.Slf4j;
 import org.springframework.beans.BeanUtils;
 import org.springframework.beans.factory.annotation.Autowired;
 import org.springframework.beans.factory.annotation.Qualifier;
+import org.springframework.context.annotation.Lazy;
 import org.springframework.stereotype.Service;
 import org.springframework.util.CollectionUtils;
 
@@ -39,6 +40,7 @@ public class AgentServiceImpl extends ServiceImpl<AgentDOMapper, AgentDO> implem
     private MemoryService memoryService;
 
     @Autowired
+    @Lazy
     private ChatQueryService chatQueryService;
 
     @Autowired

chat/server/src/main/java/com/tencent/supersonic/chat/server/service/impl/ChatQueryServiceImpl.java

@@ -49,6 +49,7 @@ import net.sf.jsqlparser.schema.Column;
 import org.apache.commons.lang3.StringUtils;
 import org.apache.commons.lang3.tuple.Pair;
 import org.springframework.beans.factory.annotation.Autowired;
+import org.springframework.context.annotation.Lazy;
 import org.springframework.stereotype.Service;
 import org.springframework.util.CollectionUtils;
 
@@ -66,6 +67,7 @@ public class ChatQueryServiceImpl implements ChatQueryService {
     @Autowired
     private SemanticLayerService semanticLayerService;
     @Autowired
+    @Lazy
     private AgentService agentService;
 
     private final List<ChatQueryParser> chatQueryParsers = ComponentFactory.getChatParsers();

common/src/main/java/com/hankcs/hanlp/LoadRemoveService.java

@@ -21,7 +21,8 @@ public class LoadRemoveService {
         List<String> resultList = new ArrayList<>(value);
         if (!CollectionUtils.isEmpty(modelIdOrDataSetIds)) {
             resultList.removeIf(nature -> {
-                if (Objects.isNull(nature)) {
+                if (Objects.isNull(nature) || !nature.startsWith("_")) { // 系统的字典是以 _ 开头的，
+                                                                         // 过滤因引用外部字典导致的异常
                     return false;
                 }
                 Long id = getId(nature);

common/src/main/java/com/tencent/supersonic/common/pojo/Text2SQLExemplar.java

@@ -22,4 +22,6 @@ public class Text2SQLExemplar implements Serializable {
     private String dbSchema;
 
     private String sql;
+
+    protected double similarity; // 传递相似度，可以作为样本筛选的依据
 }

common/src/main/java/com/tencent/supersonic/common/service/impl/ExemplarServiceImpl.java

@@ -72,7 +72,10 @@ public class ExemplarServiceImpl implements ExemplarService, CommandLineRunner {
                 embeddingService.retrieveQuery(collection, retrieveQuery, num);
         results.forEach(ret -> {
             ret.getRetrieval().forEach(r -> {
-                exemplars.add(JsonUtil.mapToObject(r.getMetadata(), Text2SQLExemplar.class));
+                Text2SQLExemplar tmp = // 传递相似度，可以作为样本筛选的依据
+                        JsonUtil.mapToObject(r.getMetadata(), Text2SQLExemplar.class);
+                tmp.setSimilarity(r.getSimilarity());
+                exemplars.add(tmp);
             });
         });
 

headless/api/src/main/java/com/tencent/supersonic/headless/api/pojo/SemanticParseInfo.java

@@ -18,6 +18,7 @@ import java.util.List;
 import java.util.Map;
 import java.util.Objects;
 import java.util.Set;
+import java.util.stream.Collectors;
 
 import static com.tencent.supersonic.common.pojo.Constants.DEFAULT_DETAIL_LIMIT;
 import static com.tencent.supersonic.common.pojo.Constants.DEFAULT_METRIC_LIMIT;
@@ -65,12 +66,23 @@ public class SemanticParseInfo implements Serializable {
             DataSetMatchResult mr2 = getDataSetMatchResult(o2.getElementMatches());
 
             double difference = mr1.getMaxDatesetSimilarity() - mr2.getMaxDatesetSimilarity();
-            if (difference == 0) {
+            if (Math.abs(difference) < 0.0005) { // 看完全匹配的个数，实践证明，可以用户输入规范后，该逻辑具有优势
+                if (!o1.getDataSetId().equals(o2.getDataSetId())) {
+                    List<SchemaElementMatch> elementMatches1 = o1.getElementMatches().stream()
+                            .filter(e -> e.getSimilarity() == 1).collect(Collectors.toList());
+                    List<SchemaElementMatch> elementMatches2 = o2.getElementMatches().stream()
+                            .filter(e -> e.getSimilarity() == 1).collect(Collectors.toList());
+                    if (elementMatches1.size() > elementMatches2.size()) {
+                        return -1;
+                    } else if (elementMatches1.size() < elementMatches2.size()) {
+                        return 1;
+                    }
+                }
                 difference = mr1.getMaxMetricSimilarity() - mr2.getMaxMetricSimilarity();
-                if (difference == 0) {
+                if (Math.abs(difference) < 0.0005) {
                     difference = mr1.getTotalSimilarity() - mr2.getTotalSimilarity();
                 }
-                if (difference == 0) {
+                if (Math.abs(difference) < 0.0005) {
                     difference = mr1.getMaxMetricUseCnt() - mr2.getMaxMetricUseCnt();
                 }
             }

headless/api/src/main/java/com/tencent/supersonic/headless/api/pojo/SqlInfo.java

@@ -16,4 +16,7 @@ public class SqlInfo implements Serializable {
 
     // SQL to be executed finally
     private String querySQL;
+
+    // Physical SQL corrected by LLM for performance optimization
+    private String correctedQuerySQL;
 }

headless/api/src/main/java/com/tencent/supersonic/headless/api/pojo/enums/ChatWorkflowState.java

@@ -8,5 +8,6 @@ public enum ChatWorkflowState {
     VALIDATING,
     SQL_CORRECTING,
     PROCESSING,
+    PHYSICAL_SQL_CORRECTING,
     FINISHED
 }

headless/chat/src/main/java/com/tencent/supersonic/headless/chat/corrector/LLMPhysicalSqlCorrector.java

@@ -0,0 +1,98 @@
+package com.tencent.supersonic.headless.chat.corrector;
+
+import com.tencent.supersonic.common.pojo.ChatApp;
+import com.tencent.supersonic.common.pojo.enums.AppModule;
+import com.tencent.supersonic.common.util.ChatAppManager;
+import com.tencent.supersonic.headless.api.pojo.SemanticParseInfo;
+import com.tencent.supersonic.headless.chat.ChatQueryContext;
+import dev.langchain4j.model.chat.ChatLanguageModel;
+import dev.langchain4j.model.input.Prompt;
+import dev.langchain4j.model.input.PromptTemplate;
+import dev.langchain4j.model.output.structured.Description;
+import dev.langchain4j.provider.ModelProvider;
+import dev.langchain4j.service.AiServices;
+import lombok.Data;
+import lombok.ToString;
+import lombok.extern.slf4j.Slf4j;
+import org.apache.commons.lang3.StringUtils;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+import java.util.HashMap;
+import java.util.Map;
+import java.util.Objects;
+
+/**
+ * 物理SQL修正器 - 使用LLM优化物理SQL性能
+ */
+@Slf4j
+public class LLMPhysicalSqlCorrector extends BaseSemanticCorrector {
+
+    private static final Logger keyPipelineLog = LoggerFactory.getLogger("keyPipeline");
+
+    public static final String APP_KEY = "PHYSICAL_SQL_CORRECTOR";
+    private static final String INSTRUCTION = ""
+            + "#Role: You are a senior database performance optimization expert experienced in SQL tuning."
+            + "\n\n#Task: You will be provided with a user question and the corresponding physical SQL query,"
+            + " please analyze and optimize this SQL to improve query performance." + "\n\n#Rules:"
+            + "\n1. DO NOT add or introduce any new fields, columns, or aliases that are not in the original SQL."
+            + "\n2. Push WHERE conditions into JOIN ON clauses when possible to reduce intermediate result sets."
+            + "\n3. Optimize JOIN order by placing smaller tables or tables with selective conditions first."
+            + "\n4. For date range conditions, ensure they are applied as early as possible in the query execution."
+            + "\n5. Remove or comment out database-specific index hints (like USE INDEX) that may cause syntax errors."
+            + "\n6. ONLY modify the structure and order of existing elements, do not change field names or add new ones."
+            + "\n7. Ensure the optimized SQL is syntactically correct and logically equivalent to the original."
+            + "\n\n#Question: {{question}}" + "\n\n#OriginalSQL: {{sql}}";
+
+    public LLMPhysicalSqlCorrector() {
+        ChatAppManager.register(APP_KEY, ChatApp.builder().prompt(INSTRUCTION).name("物理SQL修正")
+                .appModule(AppModule.CHAT).description("通过大模型对物理SQL做性能优化").enable(false).build());
+    }
+
+    @Data
+    @ToString
+    static class PhysicalSql {
+        @Description("either positive or negative")
+        private String opinion;
+
+        @Description("optimized sql if negative")
+        private String sql;
+    }
+
+    interface PhysicalSqlExtractor {
+        PhysicalSql generatePhysicalSql(String text);
+    }
+
+    @Override
+    public void doCorrect(ChatQueryContext chatQueryContext, SemanticParseInfo semanticParseInfo) {
+        ChatApp chatApp = chatQueryContext.getRequest().getChatAppConfig().get(APP_KEY);
+        if (!chatQueryContext.getRequest().getText2SQLType().enableLLM() || Objects.isNull(chatApp)
+                || !chatApp.isEnable()) {
+            return;
+        }
+
+        ChatLanguageModel chatLanguageModel =
+                ModelProvider.getChatModel(chatApp.getChatModelConfig());
+        PhysicalSqlExtractor extractor =
+                AiServices.create(PhysicalSqlExtractor.class, chatLanguageModel);
+        Prompt prompt = generatePrompt(chatQueryContext.getRequest().getQueryText(),
+                semanticParseInfo, chatApp.getPrompt());
+        PhysicalSql physicalSql =
+                extractor.generatePhysicalSql(prompt.toUserMessage().singleText());
+        keyPipelineLog.info("LLMPhysicalSqlCorrector modelReq:\n{} \nmodelResp:\n{}", prompt.text(),
+                physicalSql);
+        if ("NEGATIVE".equalsIgnoreCase(physicalSql.getOpinion())
+                && StringUtils.isNotBlank(physicalSql.getSql())) {
+            semanticParseInfo.getSqlInfo().setCorrectedQuerySQL(physicalSql.getSql());
+        }
+    }
+
+    private Prompt generatePrompt(String queryText, SemanticParseInfo semanticParseInfo,
+            String promptTemplate) {
+        Map<String, Object> variable = new HashMap<>();
+        variable.put("question", queryText);
+        variable.put("sql", semanticParseInfo.getSqlInfo().getQuerySQL());
+
+        return PromptTemplate.from(promptTemplate).apply(variable);
+    }
+}

headless/chat/src/main/java/com/tencent/supersonic/headless/chat/parser/llm/PromptHelper.java

@@ -14,10 +14,8 @@ import org.springframework.beans.factory.annotation.Autowired;
 import org.springframework.stereotype.Component;
 import org.springframework.util.CollectionUtils;
 
-import java.util.ArrayList;
+import java.util.*;
-import java.util.Collections;
+import java.util.stream.Collectors;
-import java.util.List;
-import java.util.Objects;
 
 import static com.tencent.supersonic.headless.chat.parser.ParserConfig.*;
 
@@ -51,13 +49,33 @@ public class PromptHelper {
         // use random collection of exemplars for each self-consistency inference
         for (int i = 0; i < selfConsistencyNumber; i++) {
             List<Text2SQLExemplar> shuffledList = new ArrayList<>(exemplars);
-            // only shuffle the exemplars from config
+            List<Text2SQLExemplar> same = shuffledList.stream() // 相似度极高的话，先找出来
-            List<Text2SQLExemplar> subList =
+                    .filter(e -> e.getSimilarity() > 0.989).collect(Collectors.toList());
-                    shuffledList.subList(llmReq.getDynamicExemplars().size(), shuffledList.size());
+            List<Text2SQLExemplar> noSame = shuffledList.stream()
-            Collections.shuffle(subList);
+                    .filter(e -> e.getSimilarity() <= 0.989).collect(Collectors.toList());
-            results.add(shuffledList.subList(0, Math.min(shuffledList.size(), fewShotNumber)));
+            if ((noSame.size() - same.size()) > fewShotNumber) {// 去除部分最低分
+                noSame.sort(Comparator.comparingDouble(Text2SQLExemplar::getSimilarity));
+                noSame = noSame.subList((noSame.size() - fewShotNumber) / 2, noSame.size());
+            }
+            Text2SQLExemplar mostSimilar = noSame.get(noSame.size() - 1);
+            Collections.shuffle(noSame);
+            List<Text2SQLExemplar> ts;
+            if (same.size() > 0) {// 一样的话，必须作为提示语
+                ts = new ArrayList<>();
+                int needSize = Math.min(noSame.size() + same.size(), fewShotNumber);
+                if (needSize > same.size()) {
+                    ts.addAll(noSame.subList(0, needSize - same.size()));
+                }
+                ts.addAll(same);
+            } else { // 至少要一个最像的
+                ts = noSame.subList(0, Math.min(noSame.size(), fewShotNumber));
+                if (!ts.contains(mostSimilar)) {
+                    ts.remove(ts.size() - 1);
+                    ts.add(mostSimilar);
+                }
+            }
+            results.add(ts);
         }
-
         return results;
     }
 

headless/server/src/main/java/com/tencent/supersonic/headless/server/utils/ChatWorkflowEngine.java

@@ -8,6 +8,7 @@ import com.tencent.supersonic.headless.api.pojo.request.SemanticQueryReq;
 import com.tencent.supersonic.headless.api.pojo.response.ParseResp;
 import com.tencent.supersonic.headless.api.pojo.response.SemanticTranslateResp;
 import com.tencent.supersonic.headless.chat.ChatQueryContext;
+import com.tencent.supersonic.headless.chat.corrector.LLMPhysicalSqlCorrector;
 import com.tencent.supersonic.headless.chat.corrector.SemanticCorrector;
 import com.tencent.supersonic.headless.chat.mapper.SchemaMapper;
 import com.tencent.supersonic.headless.chat.parser.SemanticParser;
@@ -76,6 +77,10 @@ public class ChatWorkflowEngine {
                     long start = System.currentTimeMillis();
                     performTranslating(queryCtx, parseResult);
                     parseResult.getParseTimeCost().setSqlTime(System.currentTimeMillis() - start);
+                    queryCtx.setChatWorkflowState(ChatWorkflowState.PHYSICAL_SQL_CORRECTING);
+                    break;
+                case PHYSICAL_SQL_CORRECTING:
+                    performPhysicalSqlCorrecting(queryCtx);
                     queryCtx.setChatWorkflowState(ChatWorkflowState.FINISHED);
                     break;
                 default:
@@ -162,4 +167,25 @@ public class ChatWorkflowEngine {
             parseResult.setErrorMsg(String.join("\n", errorMsg));
         }
     }
+
+    private void performPhysicalSqlCorrecting(ChatQueryContext queryCtx) {
+        List<SemanticQuery> candidateQueries = queryCtx.getCandidateQueries();
+        if (CollectionUtils.isNotEmpty(candidateQueries)) {
+            for (SemanticQuery semanticQuery : candidateQueries) {
+                for (SemanticCorrector corrector : semanticCorrectors) {
+                    if (corrector instanceof LLMPhysicalSqlCorrector) {
+                        corrector.correct(queryCtx, semanticQuery.getParseInfo());
+                        // 如果物理SQL被修正了，更新querySQL为修正后的版本
+                        SemanticParseInfo parseInfo = semanticQuery.getParseInfo();
+                        if (StringUtils.isNotBlank(parseInfo.getSqlInfo().getCorrectedQuerySQL())) {
+                            parseInfo.getSqlInfo().setQuerySQL(parseInfo.getSqlInfo().getCorrectedQuerySQL());
+                            log.info("Physical SQL corrected and updated querySQL: {}", 
+                                    parseInfo.getSqlInfo().getQuerySQL());
+                        }
+                        break;
+                    }
+                }
+            }
+        }
+    }
 }

headless/server/src/main/java/com/tencent/supersonic/headless/server/utils/QueryUtils.java

@@ -109,7 +109,7 @@ public class QueryUtils {
                     column.setModelId(metric.getModelId());
                 }
                 // if column nameEn contains metric alias, use metric dataFormatType
-                if (column.getDataFormatType() == null && metric.getAlias() != null) {
+                if (column.getDataFormatType() == null && StringUtils.isNotEmpty(metric.getAlias())) {
                     for (String alias : metric.getAlias().split(",")) {
                         if (nameEn.contains(alias)) {
                             column.setDataFormatType(metric.getDataFormatType());

launchers/headless/src/main/resources/META-INF/spring.factories

@@ -14,7 +14,8 @@ com.tencent.supersonic.headless.chat.parser.SemanticParser=\
 
 com.tencent.supersonic.headless.chat.corrector.SemanticCorrector=\
     com.tencent.supersonic.headless.chat.corrector.RuleSqlCorrector,\
-    com.tencent.supersonic.headless.chat.corrector.LLMSqlCorrector
+    com.tencent.supersonic.headless.chat.corrector.LLMSqlCorrector,\
+    com.tencent.supersonic.headless.chat.corrector.LLMPhysicalSqlCorrector
 
 com.tencent.supersonic.headless.chat.knowledge.file.FileHandler=\
     com.tencent.supersonic.headless.chat.knowledge.file.FileHandlerImpl

launchers/standalone/src/main/resources/META-INF/spring.factories

@@ -15,7 +15,8 @@ com.tencent.supersonic.headless.chat.parser.SemanticParser=\
 
 com.tencent.supersonic.headless.chat.corrector.SemanticCorrector=\
     com.tencent.supersonic.headless.chat.corrector.RuleSqlCorrector,\
-    com.tencent.supersonic.headless.chat.corrector.LLMSqlCorrector
+    com.tencent.supersonic.headless.chat.corrector.LLMSqlCorrector,\
+    com.tencent.supersonic.headless.chat.corrector.LLMPhysicalSqlCorrector
 
 com.tencent.supersonic.headless.chat.knowledge.file.FileHandler=\
     com.tencent.supersonic.headless.chat.knowledge.file.FileHandlerImpl

pom.xml

@@ -10,9 +10,9 @@
     <version>${revision}</version>
 
     <modules>
+        <module>common</module>
         <module>auth</module>
         <module>chat</module>
-        <module>common</module>
         <module>launchers</module>
         <module>headless</module>
     </modules>
@@ -31,6 +31,7 @@
         <java.target.version>21</java.target.version>
         <maven.compiler.source>21</maven.compiler.source>
         <maven.compiler.target>21</maven.compiler.target>
+        <maven.compiler.release>21</maven.compiler.release>
         <file.encoding>UTF-8</file.encoding>
         <jsqlparser.version>4.9</jsqlparser.version>
         <pagehelper.version>6.1.0</pagehelper.version>
@@ -254,6 +255,13 @@
                     <source>${java.source.version}</source>
                     <target>${java.target.version}</target>
                     <encoding>${file.encoding}</encoding>
+                    <annotationProcessorPaths>
+                        <path>
+                            <groupId>org.projectlombok</groupId>
+                            <artifactId>lombok</artifactId>
+                            <version>${lombok.version}</version>
+                        </path>
+                    </annotationProcessorPaths>
                 </configuration>
             </plugin>
             <plugin>

webapp/package.json

@@ -0,0 +1,5 @@
+{
+	"dependencies": {
+		"supersonic-chat-sdk": "link:packages/chat-sdk"
+	}
+}

webapp/packages/chat-sdk/src/common/type.ts

@@ -79,6 +79,7 @@ export type SqlInfoType = {
   parsedS2SQL: string;
   correctedS2SQL: string;
   querySQL: string;
+  correctedQuerySQL?: string;
 };
 
 export type ChatContextType = {

webapp/packages/chat-sdk/src/components/ChatItem/SqlItem.tsx

@@ -58,32 +58,28 @@ const SqlItem: React.FC<Props> = ({
   const getSchemaMapText = () => {
     return `
 Schema映射
-${schema?.fieldNameList?.length > 0 ? `名称：${schema.fieldNameList.join('、')}` : ''}${
+${schema?.fieldNameList?.length > 0 ? `名称：${schema.fieldNameList.join('、')}` : ''}${schema?.values?.length > 0
-      schema?.values?.length > 0
         ? `
 取值：${schema.values
-            .map((item: any) => {
+          .map((item: any) => {
-              return `${item.fieldName}: ${item.fieldValue}`;
+            return `${item.fieldName}: ${item.fieldValue}`;
-            })
+          })
-            .join('、')}`
+          .join('、')}`
         : ''
-    }${
+      }${priorExts
-      priorExts
         ? `
 附加：${priorExts}`
         : ''
-    }${
+      }${terms?.length > 0
-      terms?.length > 0
         ? `
 术语：${terms
-            .map((item: any) => {
+          .map((item: any) => {
-              return `${item.name}${item.alias?.length > 0 ? `(${item.alias.join(',')})` : ''}: ${
+            return `${item.name}${item.alias?.length > 0 ? `(${item.alias.join(',')})` : ''}: ${item.description
-                item.description
               }`;
-            })
+          })
-            .join('、')}`
+          .join('、')}`
         : ''
-    }
+      }
 
 `;
   };
@@ -91,16 +87,16 @@ ${schema?.fieldNameList?.length > 0 ? `名称：${schema.fieldNameList.join('、
   const getFewShotText = () => {
     return `
 Few-shot示例${fewShots
-      .map((item: any, index: number) => {
+        .map((item: any, index: number) => {
-        return `
+          return `
 
 示例${index + 1}：
 问题：${item.question}
 SQL：
 ${format(item.sql)}
 `;
-      })
+        })
-      .join('')}
+        .join('')}
 `;
   };
 
@@ -120,6 +116,14 @@ ${format(sqlInfo.correctedS2SQL)}
 `;
   };
 
+  const getCorrectedQuerySQLText = () => {
+    return `
+物理SQL修正
+
+${format(sqlInfo.correctedQuerySQL || '')}
+`;
+  };
+
   const getQuerySQLText = () => {
     return `
 最终执行SQL
@@ -155,6 +159,9 @@ ${executeErrorMsg}
     if (sqlInfo.correctedS2SQL) {
       text += getCorrectedS2SQLText();
     }
+    if (sqlInfo.correctedQuerySQL) {
+      text += getCorrectedQuerySQLText();
+    }
     if (sqlInfo.querySQL) {
       text += getQuerySQLText();
     }
@@ -183,9 +190,8 @@ ${executeErrorMsg}
         <div className={`${tipPrefixCls}-content-options`}>
           {llmReq && (
             <div
-              className={`${tipPrefixCls}-content-option ${
+              className={`${tipPrefixCls}-content-option ${sqlType === 'schemaMap' ? `${tipPrefixCls}-content-option-active` : ''
-                sqlType === 'schemaMap' ? `${tipPrefixCls}-content-option-active` : ''
+                }`}
-              }`}
               onClick={() => {
                 setSqlType(sqlType === 'schemaMap' ? '' : 'schemaMap');
               }}
@@ -195,9 +201,8 @@ ${executeErrorMsg}
           )}
           {fewShots.length > 0 && (
             <div
-              className={`${tipPrefixCls}-content-option ${
+              className={`${tipPrefixCls}-content-option ${sqlType === 'fewShots' ? `${tipPrefixCls}-content-option-active` : ''
-                sqlType === 'fewShots' ? `${tipPrefixCls}-content-option-active` : ''
+                }`}
-              }`}
               onClick={() => {
                 setSqlType(sqlType === 'fewShots' ? '' : 'fewShots');
               }}
@@ -207,9 +212,8 @@ ${executeErrorMsg}
           )}
           {sqlInfo.parsedS2SQL && (
             <div
-              className={`${tipPrefixCls}-content-option ${
+              className={`${tipPrefixCls}-content-option ${sqlType === 'parsedS2SQL' ? `${tipPrefixCls}-content-option-active` : ''
-                sqlType === 'parsedS2SQL' ? `${tipPrefixCls}-content-option-active` : ''
+                }`}
-              }`}
               onClick={() => {
                 setSqlType(sqlType === 'parsedS2SQL' ? '' : 'parsedS2SQL');
               }}
@@ -219,9 +223,8 @@ ${executeErrorMsg}
           )}
           {sqlInfo.correctedS2SQL && (
             <div
-              className={`${tipPrefixCls}-content-option ${
+              className={`${tipPrefixCls}-content-option ${sqlType === 'correctedS2SQL' ? `${tipPrefixCls}-content-option-active` : ''
-                sqlType === 'correctedS2SQL' ? `${tipPrefixCls}-content-option-active` : ''
+                }`}
-              }`}
               onClick={() => {
                 setSqlType(sqlType === 'correctedS2SQL' ? '' : 'correctedS2SQL');
               }}
@@ -229,16 +232,26 @@ ${executeErrorMsg}
               修正S2SQL
             </div>
           )}
+          {sqlInfo.correctedQuerySQL && (
+            <div
+              className={`${tipPrefixCls}-content-option ${sqlType === 'correctedQuerySQL' ? `${tipPrefixCls}-content-option-active` : ''
+                }`}
+              onClick={() => {
+                setSqlType(sqlType === 'correctedQuerySQL' ? '' : 'correctedQuerySQL');
+              }}
+            >
+              物理SQL修正
+            </div>
+          )}
           {sqlInfo.querySQL && (
             <div
-              className={`${tipPrefixCls}-content-option ${
+              className={`${tipPrefixCls}-content-option ${sqlType === 'querySQL' ? `${tipPrefixCls}-content-option-active` : ''
-                sqlType === 'querySQL' ? `${tipPrefixCls}-content-option-active` : ''
+                }`}
-              }`}
               onClick={() => {
                 setSqlType(sqlType === 'querySQL' ? '' : 'querySQL');
               }}
             >
-              最终执行SQL
+              {sqlInfo.correctedQuerySQL ? '最终执行SQL' : '最终执行SQL'}
             </div>
           )}
           <Button className={`${prefixCls}-export-log`} size="small" onClick={onExportLog}>
@@ -248,13 +261,12 @@ ${executeErrorMsg}
         </div>
       </div>
       <div
-        className={`${prefixCls} ${
+        className={`${prefixCls} ${!window.location.pathname.includes('/chat') &&
-          !window.location.pathname.includes('/chat') &&
           integrateSystem &&
           integrateSystem !== 'wiki'
-            ? `${prefixCls}-copilot`
+          ? `${prefixCls}-copilot`
-            : ''
+          : ''
-        }`}
+          }`}
       >
         {sqlType === 'schemaMap' && (
           <div className={`${prefixCls}-code`}>
@@ -290,9 +302,8 @@ ${executeErrorMsg}
                 <div className={`${prefixCls}-schema-content`}>
                   {terms
                     .map((item: any) => {
-                      return `${item.name}${
+                      return `${item.name}${item.alias?.length > 0 ? `(${item.alias.join(',')})` : ''
-                        item.alias?.length > 0 ? `(${item.alias.join(',')})` : ''
+                        }: ${item.description}`;
-                      }: ${item.description}`;
                     })
                     .join('、')}
                 </div>

webapp/packages/chat-sdk/src/components/ChatItem/index.tsx

@@ -416,11 +416,10 @@ const ChatItem: React.FC<Props> = ({
 
   const onExportData = () => {
     const { queryColumns, queryResults } = data || {};
-    if (!!queryResults) {
+    if (!!queryResults && !!queryColumns) {
       const exportData = queryResults.map(item => {
-        return Object.keys(item).reduce((result, key) => {
+        return queryColumns.reduce((result, column) => {
-          const columnName = queryColumns?.find(column => column.nameEn === key)?.name || key;
+          result[column.name || column.nameEn] = item[column.nameEn];
-          result[columnName] = item[key];
           return result;
         }, {});
       });