# 传统查询逻辑的问题在代码评审的过程中，发现有些查询逻辑可读性太差，但是一时又没找到好的方法。复杂的查询往往需要查询大量的信息。例如下面的例子，获取测验的正确率和参与率。还有测验每道题目分别的正确率和参与率。老师可以在课堂 (Classroom) 上发起测验 (Exam)，一个测验对应多道题目 (Question)，学生 (Student) 可以对题目进行回答 (StudentQuestionAnswer) ![image.png](https://p3-xtjj-sign.byteimg.com/tos-cn-i-73owjymdk6/bd77730592744b779bf8ec9126e2733c~tplv-73owjymdk6-jj-mark-v1:0:0:0:0:5o6Y6YeR5oqA5pyv56S-5Yy6IEAgX-eugOeugOWNleWNlV8=:q75.awebp?rk3s=f64ab15b&x-expires=1773350179&x-signature=iHwHTrum%2BIB8B%2BIfKoVhs8czn00%3D) ```java public class ExamInfoQuestionQueryService { private ExamQuestionMapper questionMapper; private StudentQuestionAnswerMapper studentQuestionAnswerMapper; private ExamMapper examMapper; private StudentMapper studentMapper; /** * 获取测验和题目的正确率和参与率 * */ public ExamDto getQuestionByExamId(String examId) { Exam exam = examMapper.selectById(examId); // 获取所有题目 List questions = questionMapper.selectByExamId(examId); List examQuestionIds = questions.stream().map(Question::getId).collect(Collectors.toList()); // 获取所有学生的答题情况 List studentQuestionAnswers = studentQuestionAnswerMapper.selectByExamQuestionIds(examQuestionIds); Map> questionMap = studentQuestionAnswers.stream().collect(Collectors.groupingBy(StudentQuestionAnswer::getExamQuestionId)); // 获取所有学生, 为了获取学生数 List students = studentMapper.selectByClassroom(exam.getClassroomId()); ExamDto examDto = new ExamDto(); // 计算测验正确率 examDto.setCorrectRate(calculateCorrectCount(studentQuestionAnswers)); // 计算测验参与度 examDto.setSubmittedRate(calculateSubmittedRate(studentQuestionAnswers,students)); List questionDtos = questions.stream().map(question -> { QuestionDto questionDto = new QuestionDto(); // 计算题目正确率 List questionAnswers = questionMap.getOrDefault(question.getId(), Collections.emptyList()); questionDto.setCorrectRate(calculateCorrectCount(questionAnswers)); // 计算题目参与率 questionDto.setSubmittedRate(calculateSubmittedRate(questionAnswers, students)); return questionDto; }).collect(Collectors.toList()); examDto.setQuestions(questionDtos); return examDto; } private double calculateSubmittedRate(List questionAnswers, List students) { return 1.0 * questionAnswers.size() / students.size(); } private long calculateCorrectCount(List questionAnswers) { long correctCount = questionAnswers.stream() .filter(StudentQuestionAnswer::isCorrect) .count(); return correctCount / questionAnswers.size(); } } ``` 上面方法的特点是需要聚合不同的数据。为了性能考虑使用了大量的 Map, 导致可维护性下降。可维护性差体现在2点 1. 变量和引用离得太远了。因为所有数据都要在组装前里面准备好，所以无法用到的时候再去获取。 2. 依靠中间变量传递，难以复用。查询对象的依赖关系如下，因为相互依赖，为了性能考虑会提前算出值传进去。 ![image.png](https://p3-xtjj-sign.byteimg.com/tos-cn-i-73owjymdk6/f18a3c4a371d433cb4c813261898a48b~tplv-73owjymdk6-jj-mark-v1:0:0:0:0:5o6Y6YeR5oqA5pyv56S-5Yy6IEAgX-eugOeugOWNleWNlV8=:q75.awebp?rk3s=f64ab15b&x-expires=1773350179&x-signature=Ut3WYagYqzqrntrNy1NgGk6JBIc%3D) 例如测验的正确率和参与度，这两个指标表面没什么联系，因为底层查询公用了 studentQuestionAnswer，所以 studentQuestionAnswer 需要提前计算了，作为参数分别传到对应的计算方法里面。如果涉及到更多的复用变量，方法将有更多的参数。理想情况下计算测验正确率，传一个 examId 就可以了，如果是为了复用，定义成下面这种会更好一些，因为只要知道 examId 就足以可以计算出测验的正确率，这个定义复用性是最好的，虽然性能可能也是最差的，因为很多中间变量又要重新计算。 ```java private double calculateExamCorrectRate(String examId); ``` # 更优雅的编写方式我认为更优雅的实现方式就是去掉临时变量。在<重构> 里面也有提到通过方法替代变量的形式，目标就是让逻辑更加直观，暂时不考虑性能问题。理想的实现方式如下: ```java public class ExamInfoQuestionQueryService2 { private ExamQuestionMapper questionMapper; private StudentQuestionAnswerMapper studentQuestionAnswerMapper; private ExamMapper examMapper; private StudentMapper studentMapper; /** * 获取测验和题目的正确率和参与率 */ public ExamDto getQuestionByExamId(String examId) { ExamDto examDto = new ExamDto(); examDto.setCorrectRate(calculateExamCorrectRate(examId)); examDto.setSubmittedRate(calculateExamSubmittedRate(examId)); examDto.setQuestions(getExamQuestions(examId).stream().map(question -> { QuestionDto questionDto = new QuestionDto(); // 计算题目正确率 questionDto.setCorrectRate(calculateQuestionCorrectRate(examId, question)); // 计算题目参与率 questionDto.setSubmittedRate(calculateQuestionSubmittedRate(examId, question)); return questionDto; }).collect(Collectors.toList())); return examDto; } private double calculateExamSubmittedRate(String examId) { return 1.0 * getStudentQuestionAnswers(examId).size() / getStudents(examId).size(); } private long calculateExamCorrectRate(String examId) { return getCorrectStudentAnswerCount(examId) / getStudentQuestionAnswers(examId).size(); } private long getCorrectStudentAnswerCount(String examId) { return getStudentQuestionAnswers(examId).stream() .filter(StudentQuestionAnswer::isCorrect) .count(); } private List getStudents(String examId) { return studentMapper.selectByClassroom(getExam(examId).getClassroomId()); } private Exam getExam(String examId) { return examMapper.selectById(examId); } private List getStudentQuestionAnswers(String examId) { return studentQuestionAnswerMapper.selectByExamQuestionIds(getExamQuestions(examId) .stream() .map(Question::getId) .collect(Collectors.toList())); } private double calculateQuestionSubmittedRate(String examId, Question question) { return 1.0 * getStudentQuestionAnswerPerQuestion(examId).apply(question).size() / getStudents(examId).size(); } private long calculateQuestionCorrectRate(String examId, Question question) { return getCorrectCountPerQuestion(examId, question) / getAnswerCount(examId, question); } private int getAnswerCount(String examId, Question question) { return getStudentQuestionAnswerPerQuestion(examId).apply(question).size(); } private long getCorrectCountPerQuestion(String examId, Question question) { return getStudentQuestionAnswerPerQuestion(examId).apply(question).stream() .filter(StudentQuestionAnswer::isCorrect) .count(); } private Function> getStudentQuestionAnswerPerQuestion(String examId) { Map> studentQuestionAnswerMap = getStudentQuestionAnswers(examId).stream() .collect(Collectors.groupingBy(StudentQuestionAnswer::getExamQuestionId)); return question -> studentQuestionAnswerMap.getOrDefault(question.getId(), Collections.emptyList()); } private List getExamQuestions(String examId) { return questionMapper.selectByExamId(examId); } } ``` 去掉临时变量之后前面提到的2个问题都得以解决，不存在临时变量，需要的时候用get方法替代即可。通过函数组合的方式，没有了中间变量，方法的定义复用性更好。例如增加一个单独查询测验正确率的方法，可以直接调用 calculateCorrectRate(examId) 即可，不用再传一堆变量。虽然方法多了很多，但每个方法都单一职责，实际上用这种方式注释都可以不用加了，因为每个方法定义都表明了方法的作用。因为方法会比较细，因此针对复杂的查询最好还是单独一个类。 # 如何解决性能问题方法重复调用会带来性能问题，如何解决性能问题? 我们观察calculateExamSubmittedRate 和 calculateExamCorrectRate 共同调用了 getStudentQuestionAnswers 方法，函数的入参其实都是一样的。又因为查询是没有副作用的，相同的参数必定返回相同的结果，因此可以缓存下来。具体实现可以加本地缓存，通过线程变量来把函数的值存下来。解决了性能问题。最后再讨论技术实现方案。那如果参数会变怎么办，例如 getStudentQuestionAnswerPerQuestion 里面的 question。不先通过批量查询的方式就算加缓存也要 n 次查询。这时可以从函数式编程寻找灵感。 # 查询函数式用函数替代变量，看到了函数式的影子。在函数式编程里面，函数有几个特点 1. 函数组合: 多个函数可以合并成一个函数。 2. 引用透明: 函数的返回值只取决于输入值，也就是没有副作用。 3. 函数缓存: 函数是闭包，内部可以存数据。 4. 高阶函数: 函数可以作为参数或者返回值传递。 5. 柯里化: 将接收多个参数的函数转换成一系列嵌套的单个参数的函数。函数组合和透明引用是加缓存的基础，如果没有透明引用，是加不了缓存的。 getStudentQuestionAnswerPerQuestion 其实是传了2 个参数，一个是 examId，一个是 question。能否把函数柯里化，改成 getStudentQuestionAnswerPerQuestion(examId)(question) 呢? 好处就是通过 getStudentQuestionAnswerPerQuestion(examId) 提前把所有 question 都查出来，再利用这个函数去筛选出单个 question 的值。 ```java /** * 函数柯里化 */ private Function> getStudentQuestionAnswerPerQuestion(String examId) { Map> studentQuestionAnswerMap = getStudentQuestionAnswers(examId).stream() .collect(Collectors.groupingBy(StudentQuestionAnswer::getExamQuestionId)); return question -> studentQuestionAnswerMap.getOrDefault(question.getId(), Collections.emptyList()); } private long getCorrectCountPerQuestion(String examId, Question question) { return getStudentQuestionAnswerPerQuestion(examId).apply(question).stream() .filter(StudentQuestionAnswer::isCorrect) .count(); } ``` getStudentQuestionAnswerPerQuestion 计算出所有question 的值，并封装成一个函数返回。因为是闭包，因此 studentQuestionAnswerMap 会被缓存在这个函数里面，当要用到的时候，只要调用 getStudentQuestionAnswerPerQuestion (可以对这个函数的返回值进行缓存 ) 获取函数，再通过 apply 把question 传递进去即可，实际调用就是通过 Map 获取值，并不会每次都去计算 Map 的值，从而解决多次查询的问题。上层方法如法炮制，最终代码如下 ```java public class ExamInfoQuestionQueryService3 { private ExamQuestionMapper questionMapper; private StudentQuestionAnswerMapper studentQuestionAnswerMapper; private ExamMapper examMapper; private StudentMapper studentMapper; /** * 获取测验和题目的正确率和参与率 */ public ExamDto getQuestionByExamId(String examId) { ExamDto examDto = new ExamDto(); examDto.setCorrectRate(calculateExamCorrectRate(examId)); examDto.setSubmittedRate(calculateExamSubmittedRate(examId)); examDto.setQuestions(getExamQuestions(examId).stream().map(question -> { QuestionDto questionDto = new QuestionDto(); // 计算题目正确率 questionDto.setCorrectRate(calculateQuestionCorrectRate(examId).apply(question)); // 计算题目参与率 questionDto.setSubmittedRate(calculateQuestionSubmittedRate(examId).apply(question)); return questionDto; }).collect(Collectors.toList())); return examDto; } private double calculateExamSubmittedRate(String examId) { return 1.0 * getStudentQuestionAnswers(examId).size() / getStudents(examId).size(); } private long calculateExamCorrectRate(String examId) { return getCorrectStudentAnswerCount(examId) / getStudentQuestionAnswers(examId).size(); } private long getCorrectStudentAnswerCount(String examId) { return getStudentQuestionAnswers(examId).stream().filter(StudentQuestionAnswer::isCorrect).count(); } private List getStudents(String examId) { return studentMapper.selectByClassroom(getExam(examId).getClassroomId()); } private Exam getExam(String examId) { return examMapper.selectById(examId); } private List getStudentQuestionAnswers(String examId) { return studentQuestionAnswerMapper.selectByExamQuestionIds(getExamQuestionIds(getExamQuestions(examId))); } private static List getExamQuestionIds(List questions) { return questions.stream().map(Question::getId).collect(Collectors.toList()); } private Function calculateQuestionSubmittedRate(String examId) { return question -> 1.0 * getStudentQuestionAnswerPerQuestion(examId).apply(question).size() / getStudents(examId).size(); } private Function calculateQuestionCorrectRate(String examId) { return question -> 1.0 * getCorrectCountPerQuestion(examId).apply(question) / getAnswerCount(examId).apply(question); } private Function getAnswerCount(String examId) { return question -> getStudentQuestionAnswerPerQuestion(examId).apply(question).size(); } private Function getCorrectCountPerQuestion(String examId) { return question -> getStudentQuestionAnswerPerQuestion(examId).apply(question).stream().filter(StudentQuestionAnswer::isCorrect).count(); } private Function> getStudentQuestionAnswerPerQuestion(String examId) { Map> studentQuestionAnswerMap = getStudentQuestionAnswers(examId).stream().collect(Collectors.groupingBy(StudentQuestionAnswer::getExamQuestionId)); return question -> studentQuestionAnswerMap.getOrDefault(question.getId(), Collections.emptyList()); } private List getExamQuestions(String examId) { return questionMapper.selectByExamId(examId); } } ``` 使用 kotlin 的话效果更好，因为 kotlin 原生就支持函数式，不像 java 还需要有个 Function。 ```kotlin class ExamInfoQuestionQueryService4 { private lateinit var questionMapper: ExamQuestionMapper private lateinit var studentQuestionAnswerMapper: StudentQuestionAnswerMapper private lateinit var examMapper: ExamMapper private lateinit var studentMapper: StudentMapper /** * 获取测验和题目的正确率和参与率 */ fun getQuestionByExamId(examId: String): ExamDto { return ExamDto( correctRate = calculateExamCorrectRate(examId), submittedRate = calculateExamSubmittedRate(examId), questions = getExamQuestions(examId).map { question: Question -> QuestionDto( // 计算题目正确率 correctRate = calculateQuestionCorrectRate(examId)(question), // 计算题目参与率 submittedRate = calculateQuestionSubmittedRate(examId)(question) ) } ) } private fun calculateExamSubmittedRate(examId: String): Double { return 1.0 * getStudentQuestionAnswers(examId).size / getStudents(examId).size } private fun calculateExamCorrectRate(examId: String): Double { return 1.0 * getCorrectStudentAnswerCount(examId) / getStudentQuestionAnswers(examId).size } private fun getCorrectStudentAnswerCount(examId: String): Int { return getStudentQuestionAnswers(examId).count { obj: StudentQuestionAnswer -> obj.isCorrect } } private fun getStudents(examId: String): List { return studentMapper.selectByClassroom(getExam(examId).classroomId) } private fun getExam(examId: String): Exam { return examMapper.selectById(examId) } private fun getStudentQuestionAnswers(examId: String): List { return studentQuestionAnswerMapper.selectByExamQuestionIds(getExamQuestions(examId).map(Question::getId)) } private fun calculateQuestionSubmittedRate(examId: String): (Question) -> Double { return { question: Question -> 1.0 * getStudentQuestionAnswerPerQuestion(examId)(question).size / getStudents(examId).size } } private fun calculateQuestionCorrectRate(examId: String): (Question) -> Double { return { question: Question -> 1.0 * getCorrectCountPerQuestion(examId)(question) / getAnswerCount(examId)(question) } } private fun getAnswerCount(examId: String): (Question) -> Int { return { question: Question -> getStudentQuestionAnswerPerQuestion(examId)(question).size } } private fun getCorrectCountPerQuestion(examId: String): (Question) -> Int { return { question: Question -> getStudentQuestionAnswerPerQuestion(examId)(question) .count { obj: StudentQuestionAnswer -> obj.isCorrect } } } private fun getStudentQuestionAnswerPerQuestion(examId: String): (Question) -> List { val studentQuestionAnswerMap = getStudentQuestionAnswers(examId) .groupBy(StudentQuestionAnswer::getExamQuestionId) return { question: Question -> studentQuestionAnswerMap[question.id] ?: emptyList() } } private fun getExamQuestions(examId: String): List { return questionMapper.selectByExamId(examId) } } ``` 我把上面的实现方式称为函数式查询，利用函数的方式组合和组装查询值。 # 缓存的实现方式对方法实现做拦截，且类非接口，最简单的方式就是 cglib 了，cglib 可以通过继承对源方法进行拦截。 ![image.png](https://p3-xtjj-sign.byteimg.com/tos-cn-i-73owjymdk6/697a6375ad954146abe73e65c0d2cd25~tplv-73owjymdk6-jj-mark-v1:0:0:0:0:5o6Y6YeR5oqA5pyv56S-5Yy6IEAgX-eugOeugOWNleWNlV8=:q75.awebp?rk3s=f64ab15b&x-expires=1773350179&x-signature=3YSR83p9%2Bsm%2FMaR0TNa8TWpHlz8%3D) 通过 BeanPostProcessor 对 bean 进行代理 ```java @Order(Ordered.HIGHEST_PRECEDENCE) class QueryFunctionalBeanFactory : BeanPostProcessor { /** * 此方法可以在其他 bean 注入这个 bean 之前对这个 bean 进行替换 */ override fun postProcessBeforeInitialization(bean: Any?, beanName: String?): Any? { if (bean == null) { return null } val targetClass = AopUtils.getTargetClass(bean) // QueryFunctional 是标识要缓存的类 if (targetClass.isAnnotationPresent(QueryFunctional::class.java)) { val enhancer = Enhancer() enhancer.setSuperclass(targetClass) enhancer.setCallback(object : MethodInterceptor { override fun intercept(obj: Any, method: Method, args: Array, methodProxy: MethodProxy): Any? { return QueryFunctionalCache.cache(method) { return@cache methodProxy.invokeSuper(obj, args) } } }) // 创建 cglib 代理 val newBean = enhancer.create() // 把代理对象设进代理类 for (fileId in bean.javaClass.fields) { fileId.isAccessible = true fileId.set(newBean, fileId.get(bean)) } return newBean } return bean } override fun postProcessAfterInitialization(bean: Any?, beanName: String?): Any? { return bean } } ``` 缓存的逻辑如下 ```java class QueryFunctionalCache { companion object { /** * 线程缓存缓存方法的返回值 */ private var cacheDatas = object : ThreadLocal>() { override fun initialValue(): MutableMap { return mutableMapOf() } } fun cache(method: Method, any: () -> R): R { var cacheMethodReturnValueMap = cacheDatas.get() // 因为入参只有一个, 且不会变, 因此只要缓存一份就好 if (cacheMethodReturnValueMap.containsKey(method)) { return cacheMethodReturnValueMap.get(method) as R } val methodReturnValue = any() cacheMethodReturnValueMap = cacheMethodReturnValueMap ?: mutableMapOf() cacheMethodReturnValueMap.putAll(mutableMapOf(Pair(method, methodReturnValue))) return methodReturnValue } } } ``` 上面的代码有个缺陷，同时也函数式查询的一个不足，就是入参必须不变。上面的例子大部分方法都是 examId，每次查询都是不变的，因此每次查询缓存只有一份，加缓存并不需要把入参作为key。只要每次进入查询入口方法的时候缓存，退出的时候删除就可以了。但这也限制了提取的方法的入参都要一样的。当入口方法参数多的时候，可以提取成一个 dto，后面所有方法都以这一个 dto 为参数进行设计，这样方法更容易组装，因为入参都是同一个 dto，因此想调用什么方法都不用关心参数的问题。但这样函数的复用性就变得很差了，因为 dto 一般不同的查询是不同。解决方法可以针对通用的查询细化参数，不使用 dto，只要参数是固定的，只缓存一份的规则仍然生效，但里面组合的函数也要改成一样的参数。实际使用中单个查询里面的复用是非常多的，不同查询的复用较少。 # 总计函数式查询本质是通过函数替代变量，通过缓存和柯里化解决性能问题。虽然上面的代码经过精简不能用，但组件已经完成，有个项目已用上，效果还不错，把大量的 map 都去掉了，提升了可读性，定位问题的速度也更快了。