FT.CREATE custom-index ON JSON PREFIX 1 "doc:" SCHEMA $.user_id AS user_id TAG $.content AS content TEXT $.embedding AS embedding VECTOR HNSW 6 TYPE FLOAT32 DIM 1024 DISTANCE_METRIC COSINE
publicinterfaceChatMemory{StringDEFAULT_CONVERSATION_ID="default";StringCONVERSATION_ID="chat_memory_conversation_id";defaultvoidadd(String conversationId,Message message){Assert.hasText(conversationId,"conversationId cannot be null or empty");Assert.notNull(message,"message cannot be null");this.add(conversationId,List.of(message));}voidadd(String conversationId,List<Message> messages);List<Message>get(String conversationId);voidclear(String conversationId);}
Spring AI为我们默认实现了一个实现类InMemoryChatMemoryRepository，可将会话保存到本地内存中用于测试，如果我们没有自定义ChatMemory实现类注入，默认的InMemoryChatMemoryRepository将会注入
这里采用Java实现一个多线程的DNS代理服务器，对于内网域名直接通过内网DNS的UDP:53进行解析，对于外网域名则以加密的DOH（DNS Over Https）方式通过阿里云DNS进行解析，并解析DNS服务器返回的报文并打印日志。需要依赖dnsjava这个类库的支持，程序启动后，只需要将网卡DNS服务器地址和备用地址修改为127.0.0.1和127.0.0.2即可实现DNS的分流。
<dependencies><!-- DNS 处理库 --><dependency><groupId>dnsjava</groupId><artifactId>dnsjava</artifactId><version>3.6.0</version></dependency><!-- HTTP 客户端（用于DoH请求） --><dependency><groupId>org.apache.httpcomponents.client5</groupId><artifactId>httpclient5</artifactId><version>5.3</version></dependency></dependencies>
大模型中最常见的大语言模型(Large Language Models，LLM)，就是采用了深度学习中的自然语言处理这一分支，在自然语言处理(Natural Language Processing，NLP)中，有一项关键技术叫Transformer，这是一种先进的神经网络模型，是现如今AI高速发展的最主要原因，我们所熟知的大语言模型，例如GPT、Deepseek底层都是采用Transformer神经网络模型
Agent + Function calling（智能体） AI拆解任务，通过将AI能力和业务端的能力相结合，通过调用业务端提供的接口实现复杂业务，大模型可以适时调用业务端提供的函数来获取信息来进一步做判断，可以应用于数据提取和聚合分析等，例如要用大模型来进行行程规划同时提供一个天气的function给大模型，来为大模型做行程规划提供天气信息。
SELECT id, employee_id, fun_first_name((select n from users u where u.id = e.uid))as first_name ,(select last_name from users u where u.id = e.uid)as last_name,EXISTS(SELECT1FROM projects WHERE manager_id = e.employee_id)AS is_managerFROM employees e;
解析并处理后得到SQL：
SELECT id, employee_id, fun_first_name((SELECT n FROM users u WHERE u.id = e.uid AND users.scope =12))AS first_name,(SELECT last_name FROM users u WHERE u.id = e.uid AND users.scope =12)AS last_name,EXISTS(SELECT1FROM projects WHERE manager_id = e.employee_id AND projects.scope =12)AS is_manager FROM employees eWHERE employees.scope =12
EXISTS (...) as ..不能写成( EXISTS (...) ) as ..，否则不会被解析为Select而是会被解析为Parenthesis，而该方法没有提供Parenthesis的解析，会导致被忽略
SELECT name FROMuser u WHERE u.math_score <(SELECTavg(score)FROM math )OR u.english_score >(SELECTavg(score)FROM english )AND(SELECT order_num FROM student )= u.order_num AND u.role_id IN(SELECT id FROM role )ANDEXISTS(SELECT*FROM customer WHERE id =6)ANDNOTEXISTS(SELECT*FROM customer WHERE id =7)
在这段SQL中，通过plainSelect.getWhere()得到的where的部分是：u.math_score < (SELECT avg(score) FROM math) OR u.english_score > (SELECT avg(score) FROM english) AND (SELECT order_num FROM student) = u.order_num AND u.role_id IN (SELECT id FROM role) AND EXISTS (SELECT * FROM customer WHERE id = 6) AND NOT EXISTS (SELECT * FROM customer WHERE id = 7)，该部分会作为参数传入Expression where中，这段复杂的where表达式中的子查询是采用拆分的方法解析到的，具体解析和追加的步骤如下：
expression.getLeftExpression() => u.math_score < (SELECT avg(score) FROM math)
expression.getRightExpression() => u.english_score > (SELECT avg(score) FROM english) AND (SELECT order_num FROM student) = u.order_num AND u.role_id IN (SELECT id FROM role) AND EXISTS (SELECT * FROM customer WHERE id = 6) AND NOT EXISTS (SELECT * FROM customer WHERE id = 7)
第二次拆分：执行到processWhereSubSelect(expression.getLeftExpression(), whereSegment)处，将u.math_score < (SELECT avg(score) FROM math)传入processWhereSubSelect递归解析，这次执行仍然满足where instanceof FromItem == false，where instanceof BinaryExpression == true，u.math_score < (SELECT avg(score) FROM math)将被拆分为：
expression.getLeftExpression() => u.math_score
expression.getRightExpression() => (SELECT avg(score) FROM math)
接下来还会递归执行到processWhereSubSelect(expression.getLeftExpression(), whereSegment)处，将u.math_score传入processWhereSubSelect递归解析，没有满足条件的分支直接跳过，紧接着执行processWhereSubSelect(expression.getRightExpression(), whereSegment)，将(SELECT avg(score) FROM math)传入processWhereSubSelect递归解析，这次执行满足where instanceof FromItem的条件，不需要拆分，执行processOtherFromItem对(SELECT avg(score) FROM math)进行过滤条件追加，至此，第一步拆分拆出来的bexpression.getLeftExpression()部分解析处理完成，第一段递归随即跳出。
expression.getLeftExpression() => u.english_score > (SELECT avg(score) FROM english) AND (SELECT order_num FROM student) = u.order_num AND u.role_id IN (SELECT id FROM role) AND EXISTS (SELECT * FROM customer WHERE id = 6)
expression.getRightExpression() => NOT EXISTS (SELECT * FROM customer WHERE id = 7)
同理，取出expression.getLeftExpression()进行第四次拆分：
where instanceof BinaryExpression:
expression.getLeftExpression() => u.english_score > (SELECT avg(score) FROM english) AND (SELECT order_num FROM student) = u.order_num AND u.role_id IN (SELECT id FROM role)
expression.getRightExpression() => EXISTS (SELECT * FROM customer WHERE id = 6)
第五次拆分：
where instanceof BinaryExpression:
expression.getLeftExpression() => u.english_score > (SELECT avg(score) FROM english) AND (SELECT order_num FROM student) = u.order_num
expression.getRightExpression() => u.role_id IN (SELECT id FROM role)
第六次拆分：
where instanceof BinaryExpression:
expression.getLeftExpression() => u.english_score > (SELECT avg(score) FROM english)
expression.getRightExpression() => (SELECT order_num FROM student) = u.order_num
第七次拆分：
where instanceof BinaryExpression:
expression.getLeftExpression() => u.english_score
expression.getRightExpression() => (SELECT avg(score) FROM english)
expression.getRightExpression() => (SELECT avg(score) FROM english)
u.english_score不满足任何分支，直接跳过，(SELECT avg(score) FROM english)是子查询，调用processOtherFromItem()处理。
处理第六次拆分的RightExpression：
where instanceof BinaryExpression:
expression.getLeftExpression() => (SELECT order_num FROM student)
expression.getRightExpression() => u.order_num
(SELECT order_num FROM student)是子查询，调用processOtherFromItem()处理，u.order_num不满足任何分支，直接跳过
处理第五次拆分的RightExpression：
where instanceof InExpression:
expression.getLeftExpression() => u.role_id
expression.getRightExpression() => (SELECT id FROM role)
u.role_id不满足任何分支，直接跳过，(SELECT id FROM role)，通过IN解析子查询，然后调用processOtherFromItem()处理
处理第四次拆分的RightExpression：
where instanceof ExistsExpression:
expression.getRightExpression() => (SELECT * FROM customer WHERE id = 6)
EXISTS (SELECT * FROM customer WHERE id = 6)满足where instanceof ExistsExpression的情况，提取出(SELECT * FROM customer WHERE id = 6)子查询，调用processOtherFromItem()处理
处理第三次拆分的RightExpression：
where instanceof NotExpression:
expression.getExpression() => EXISTS (SELECT * FROM customer WHERE id = 7)
先调用processWhereSubSelect()从NOT EXISTS (SELECT * FROM customer WHERE id = 7)中提取出EXISTS (SELECT * FROM customer WHERE id = 7)，再走到where instanceof ExistsExpression分支提取出子查询(SELECT * FROM customer WHERE id = 7)调用processOtherFromItem()处理
至此，WHERE语句中所有需要追加条件的表都解析追加完成了，最终得到SQL如下：
SELECT name FROMuser u WHERE(u.math_score <(SELECTavg(score)FROM math WHERE math.scope =12)OR u.english_score >(SELECTavg(score)FROM english WHERE english.scope =12)AND(SELECT order_num FROM student WHERE student.scope =12)= u.order_num AND u.role_id IN(SELECT id FROM role WHERE role.scope =12)ANDEXISTS(SELECT*FROM customer WHERE id =6AND customer.scope =12)ANDNOTEXISTS(SELECT*FROM customer WHERE id =7AND customer.scope =12))ANDuser.scope =12
/** * 处理子查询等 */protectedvoidprocessOtherFromItem(FromItem fromItem,finalString whereSegment){// 去除括号// while (fromItem instanceof ParenthesisFromItem) {// fromItem = ((ParenthesisFromItem) fromItem).getFromItem();// }if(fromItem instanceofParenthesedSelect){Select subSelect =(Select) fromItem;processSelectBody(subSelect, whereSegment);}elseif(fromItem instanceofParenthesedFromItem){ logger.debug("Perform a subQuery, if you do not give us feedback");}}
SELECT u.id, u.name FROM userinfo u, dept d, role r WHERE u.p =1AND u.dept_id = d.id AND u.rid = r.id AND userinfo.scope =12AND dept.scope =12AND role.scope =12
3.7.2 INNER JOIN
SELECT u.id, u.name FROM userinfo u INNERJOIN dept d ON u.dept_id = d.id INNERJOIN role r ON u.rid = r.id WHERE u.p =1
SELECT u.id, u.name FROM userinfo u INNERJOIN dept d ON u.dept_id = d.id AND userinfo.scope =12AND dept.scope =12INNERJOIN role r ON u.rid = r.id AND role.scope =12WHERE u.p =1
3.7.3 LEFT JOIN
SELECT u.id, u.name FROM userinfo u LEFTJOIN dept d ON u.dept_id = d.id LEFTJOIN role r ON u.rid = r.id WHERE u.p =1
LEFT JOIN取的是FROM表的全部数据，是最简单的一种情况，方法开始执行时，参数mainTables中传入userinfo，joins中存放的则是dept，role两张表，局部变量mainTable和leftTable均为userinfo，因为LEFT JOIN取的是userinfo表的全部数据，因此mainTables中的userinfo就是驱动表，过滤条件加在WHERE上。LEFT JOIN的dept和role两张表都是被驱动表，过滤条件加在ON上。
SELECT u.id, u.name FROM userinfo u LEFTJOIN dept d ON u.dept_id = d.id AND dept.scope =12LEFTJOIN role r ON u.rid = r.id AND role.scope =12WHERE u.p =1AND userinfo.scope =12
3.7.4 RIGHT JOIN
SELECT u.id, u.name FROM userinfo u RIGHTJOIN dept d ON u.dept_id = d.id RIGHTJOIN role r ON u.rid = r.id WHERE u.p =1
RIGHT JOIN取的是JOIN后的表的全部数据，和LEFT JOIN正好相反，方法开始执行时，参数mainTables中传入userinfo，joins中存放的则是dept，role两张表，局部变量mainTable和leftTable均为userinfo
SELECT u.id, u.name FROM userinfo u RIGHTJOIN dept d ON u.dept_id = d.id AND userinfo.scope =12RIGHTJOIN role r ON u.rid = r.id AND dept.scope =12WHERE u.p =1AND role.scope =12
3.7.5 先INNER再RIGHT
SELECT u.id, u.name FROM userinfo u INNERJOIN dept d ON u.dept_id = d.id RIGHTJOIN role r ON u.rid = r.id WHERE u.p =1
SELECT u.id, u.nameFROM userinfo uINNERJOIN dept d ON u.dept_id = d.id AND userinfo.scope =12AND dept.scope =12RIGHTJOIN role r ON u.rid = r.id AND dept.scope =12WHERE u.p =1AND role.scope =12
3.7.6 先RIGHT再INNER
SELECT u.id, u.name FROM userinfo u RIGHTJOIN dept d ON u.dept_id = d.id INNERJOIN role r ON u.rid = r.id WHERE u.p =1
SELECT u.id, u.name FROM userinfo u RIGHTJOIN dept d ON u.dept_id = d.id AND userinfo.scope =12INNERJOIN role r ON u.rid = r.id AND dept.scope =12AND role.scope =12WHERE u.p =1
3.7.7 先INNER再LEFT
SELECT u.id, u.name FROM userinfo u INNERJOIN dept d ON u.dept_id = d.id LEFTJOIN role r ON u.rid = r.id WHERE u.p =1
SELECT u.id, u.name FROM userinfo u INNERJOIN dept d ON u.dept_id = d.id AND userinfo.scope =12AND dept.scope =12LEFTJOIN role r ON u.rid = r.id AND role.scope =12WHERE u.p =1
3.7.8 先LEFT再INNER
SELECT u.id, u.name FROM userinfo u LEFTJOIN dept d ON u.dept_id = d.id INNERJOIN role r ON u.rid = r.id WHERE u.p =1
SELECT u.id, u.name FROM userinfo u LEFTJOIN dept d ON u.dept_id = d.id AND dept.scope =12INNERJOIN role r ON u.rid = r.id AND userinfo.scope =12AND role.scope =12WHERE u.p =1
3.7.9 先RIGHT再LEFT
SELECT u.id, u.name FROM userinfo u RIGHTJOIN dept d ON u.dept_id = d.id LEFTJOIN role r ON u.rid = r.id WHERE u.p =1
SELECT u.id, u.name FROM userinfo u RIGHTJOIN dept d ON u.dept_id = d.id AND userinfo.scope =12LEFTJOIN role r ON u.rid = r.id AND role.scope =12WHERE u.p =1AND dept.scope =12
3.7.10 先LEFT再RIGHT
SELECT u.id, u.name FROM userinfo u LEFTJOIN dept d ON u.dept_id = d.id RIGHTJOIN role r ON u.rid = r.id WHERE u.p =1
SELECT u.id, u.name FROM userinfo u LEFTJOIN dept d ON u.dept_id = d.id AND dept.scope =12RIGHTJOIN role r ON u.rid = r.id AND userinfo.scope =12WHERE u.p =1AND role.scope =12
3.7.11 FROM子查询JOIN表
LEFT JOIN：
SELECT u.id, u.name FROM(SELECT*FROM userinfo ) u LEFTJOIN dept d ON u.dept_id = d.id LEFTJOIN role r ON u.rid = r.id WHERE u.p =1
SELECT u.id, u.name FROM(SELECT*FROM userinfo WHERE userinfo.scope =12) u LEFTJOIN dept d ON u.dept_id = d.id AND dept.scope =12LEFTJOIN role r ON u.rid = r.id AND role.scope =12WHERE u.p =1
RIGHT JOIN：
SELECT u.id, u.name FROM(SELECT*FROM userinfo ) u RIGHTJOIN dept d ON u.dept_id = d.id RIGHTJOIN role r ON u.rid = r.id WHERE u.p =1
SELECT u.id, u.name FROM(SELECT*FROM userinfo WHERE userinfo.scope =12) u RIGHTJOIN dept d ON u.dept_id = d.id RIGHTJOIN role r ON u.rid = r.id AND dept.scope =12WHERE u.p =1AND role.scope =12
3.7.12 FROM表JOIN子查询
RIGHT JOIN：
SELECT u.id, u.name FROM userinfo u RIGHTJOIN(SELECT*FROM dept ) d ON u.dept_id = d.id RIGHTJOIN(SELECT*FROM role ) r ON u.rid = r.id WHERE u.p =1
SELECT u.id, u.name FROM userinfo u RIGHTJOIN(SELECT*FROM dept WHERE dept.scope =12) d ON u.dept_id = d.id RIGHTJOIN(SELECT*FROM role WHERE role.scope =12) r ON u.rid = r.id WHERE u.p =1AND userinfo.scope =12
LEFT JOIN：
SELECT u.id, u.name FROM userinfo u LEFTJOIN(SELECT*FROM dept ) d ON u.dept_id = d.id LEFTJOIN(SELECT*FROM role ) r ON u.rid = r.id WHERE u.p =1
处理LEFT的情况和RIGHT是一样的，得到的SQL形式也相同：
SELECT u.id, u.name FROM userinfo u LEFTJOIN(SELECT*FROM dept WHERE dept.scope =12) d ON u.dept_id = d.id LEFTJOIN(SELECT*FROM role WHERE role.scope =12) r ON u.rid = r.id WHERE u.p =1AND userinfo.scope =12
3.7.13 FROM子查询JOIN子查询
SELECT u.id, u.name FROM(SELECT*FROM userinfo ) u RIGHTJOIN(SELECT*FROM dept ) d ON u.dept_id = d.id RIGHTJOIN(SELECT*FROM role ) r ON u.rid = r.id WHERE u.p =1
这种情况本质上和FROM表JOIN子查询是一样的
SELECT u.id, u.name FROM(SELECT*FROM userinfo WHERE userinfo.scope =12) u RIGHTJOIN(SELECT*FROM dept WHERE dept.scope =12) d ON u.dept_id = d.id RIGHTJOIN(SELECT*FROM role WHERE role.scope =12) r ON u.rid = r.id WHERE u.p =1
3.7.14 不支持的情况
processJoins()方法似乎并不是万能的，有几种我遇到的不能支持的极端情况：
1.JOIN表和JOIN子查询混用时，使用了RIGHT会导致丢掉某个表的过滤条件
以下两个是重写过的SQL，都会导致userinfo表的scope条件丢失
SELECT u.id, u.name FROM userinfo u LEFTJOIN(SELECT*FROM dept WHERE dept.scope =12) d ON u.dept_id = d.id RIGHTJOIN role r ON u.rid = r.id LEFTJOIN(SELECT*FROM job WHERE job.scope =12) j ON u.jid = j.id WHERE u.p =1AND role.scope =12
SELECT u.id, u.nameFROM userinfo uRIGHTJOIN(SELECT*FROM dept WHERE dept.scope =12) d ON u.dept_id = d.idRIGHTJOIN role r ON u.rid = r.idWHERE u.p =1AND role.scope =12
例：这是一个重写过的SQL，因为from后的表不存在(因为是子查询)，在执行leftTable = mainTable == null ? joinTable时，将left join的dept表错误的作为了驱动表，导致下次right join时以dept表为基准，将dept又追加一次dept.scope = 12，实际应当以(SELECT * FROM userinfo WHERE userinfo.scope = 12)为基准，这样就导致(SELECT * FROM userinfo WHERE userinfo.scope = 12)的记录不全
SELECT u.id, u.name FROM(SELECT*FROM userinfo WHERE userinfo.scope =12) u LEFTJOIN dept d ON u.dept_id = d.id AND dept.scope =12RIGHTJOIN role r ON u.rid = r.id AND dept.scope =12WHERE u.p =1AND role.scope =12
SELECTCASEWHEN id >=90THEN(SELECT id FROM system_users WHERE parent_dept_id =9)WHEN id >=80THEN(SELECT id FROM system_users WHERE parent_dept_id =6)WHEN(SELECT id FROM system_users WHERE parent_dept_id =5)>=70THEN(SELECT id FROM system_users WHERE parent_dept_id =5)ELSE100ENDAS grade FROM system_users WHERE system_users.scope =12
BinaryExpression expression =(BinaryExpression) obj;Expression left = expression.getLeftExpression();Expression right = expression.getRightExpression();
/* * Copyright 2002-2021 the original author or authors. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at * * https://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, software * distributed under the License is distributed on an "AS IS" BASIS, * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. * See the License for the specific language governing permissions and * limitations under the License. */packageorg.springframework.messaging.handler.invocation;importcn.iocoder.yudao.framework.tenant.core.context.TenantContextHolder;importcn.iocoder.yudao.framework.tenant.core.util.TenantUtils;importorg.springframework.core.DefaultParameterNameDiscoverer;importorg.springframework.core.MethodParameter;importorg.springframework.core.ParameterNameDiscoverer;importorg.springframework.core.ResolvableType;importorg.springframework.lang.Nullable;importorg.springframework.messaging.Message;importorg.springframework.messaging.handler.HandlerMethod;importorg.springframework.util.ObjectUtils;importjava.lang.reflect.InvocationTargetException;importjava.lang.reflect.Method;importjava.lang.reflect.Type;importjava.util.Arrays;importstaticcn.iocoder.yudao.framework.web.core.util.WebFrameworkUtils.HEADER_TENANT_ID;/** * Extension of {@link HandlerMethod} that invokes the underlying method with * argument values resolved from the current HTTP request through a list of * {@link HandlerMethodArgumentResolver}. * * 针对 rabbitmq-spring 和 kafka-spring，不存在合适的拓展点，可以实现 Consumer 消费前，读取 Header 中的 tenant-id 设置到 {@link TenantContextHolder} 中 * TODO 芋艿：持续跟进，看看有没新的拓展点 * * @author Rossen Stoyanchev * @author Juergen Hoeller * @since 4.0 */publicclassInvocableHandlerMethodextendsHandlerMethod{.........@NullablepublicObjectinvoke(Message<?> message,Object... providedArgs)throwsException{Object[] args =getMethodArgumentValues(message, providedArgs);if(logger.isTraceEnabled()){ logger.trace("Arguments: "+Arrays.toString(args));}// 注意：如下是本类的改动点！！！// 情况一：无租户编号的情况Long tenantId=parseTenantId(message);if(tenantId ==null){returndoInvoke(args);}// 情况二：有租户的情况下returnTenantUtils.execute(tenantId,()->doInvoke(args));}privateLongparseTenantId(Message<?> message){Object tenantId = message.getHeaders().get(HEADER_TENANT_ID);if(tenantId ==null){returnnull;}if(tenantId instanceofLong){return(Long) tenantId;}if(tenantId instanceofNumber){return((Number) tenantId).longValue();}if(tenantId instanceofString){returnLong.parseLong((String) tenantId);}if(tenantId instanceofbyte[]){returnLong.parseLong(newString((byte[]) tenantId));}thrownewIllegalArgumentException("未知的数据类型："+ tenantId);}.........}
登录
