性能优化之Java多线程批量拆分List导入数据库

| 前言

前两天做了一个导入的功能，导入开始的时候非常慢，导入2w条数据要1分多钟，后来一点一点的优化，从直接把list怼进Mysql中，到分配把list导入Mysql中，到多线程把list导入Mysql中。时间是一点一点的变少了。非常的爽，最后变成了10s以内。下面就展示一下过程。

| 直接把list怼进Mysql

使用mybatis的批量导入操作：

@Transactional(rollbackFor = Exception.class)public int addFreshStudentsNew2(List<FreshStudentAndStudentModel> list, String schoolNo) {if (list == null || list.isEmpty()) {return 0;}List<StudentEntity> studentEntityList = new LinkedList<>();List<EnrollStudentEntity> enrollStudentEntityList = new LinkedList<>();List<AllusersEntity> allusersEntityList = new LinkedList<>();for (FreshStudentAndStudentModel freshStudentAndStudentModel : list) {EnrollStudentEntity enrollStudentEntity = new EnrollStudentEntity();StudentEntity studentEntity = new StudentEntity();BeanUtils.copyProperties(freshStudentAndStudentModel, studentEntity);BeanUtils.copyProperties(freshStudentAndStudentModel, enrollStudentEntity);String operator = TenancyContext.UserID.get();String studentId = BaseUuidUtils.base58Uuid();enrollStudentEntity.setId(BaseUuidUtils.base58Uuid());enrollStudentEntity.setStudentId(studentId);enrollStudentEntity.setIdentityCardId(freshStudentAndStudentModel.getIdCard());enrollStudentEntity.setOperator(operator);studentEntity.setId(studentId);studentEntity.setIdentityCardId(freshStudentAndStudentModel.getIdCard());studentEntity.setOperator(operator);studentEntityList.add(studentEntity);enrollStudentEntityList.add(enrollStudentEntity);AllusersEntity allusersEntity = new AllusersEntity();allusersEntity.setId(enrollStudentEntity.getId());allusersEntity.setUserCode(enrollStudentEntity.getNemtCode());allusersEntity.setUserName(enrollStudentEntity.getName());allusersEntity.setSchoolNo(schoolNo);allusersEntity.setTelNum(enrollStudentEntity.getTelNum());allusersEntity.setPassword(enrollStudentEntity.getNemtCode());  //密码设置为考生号allusersEntityList.add(allusersEntity);}enResult = enrollStudentDao.insertAll(enrollStudentEntityList);stuResult = studentDao.insertAll(studentEntityList);allResult = allusersFacade.insertUserList(allusersEntityList);if (enResult > 0 && stuResult > 0 && allResult) {return 10;}return -10;}

Mapper.xml

<insert id="insertAll" parameterType="com.dmsdbj.itoo.basicInfo.entity.EnrollStudentEntity">insert into tb_enroll_student<trim prefix="(" suffix=")" suffixOverrides=",">id,  remark,  nEMT_aspiration,  nEMT_code,  nEMT_score,  student_id,  identity_card_id,  level,  major,  name,  nation,  secondary_college,  operator,  sex,  is_delete,  account_address,  native_place,  original_place,  used_name,  pictrue,  join_party_date,  political_status,  tel_num,  is_registry,  graduate_school,  create_time,  update_time        </trim>        values<foreach collection="list" item="item" index="index" separator=",">(#{item.id,jdbcType=VARCHAR},#{item.remark,jdbcType=VARCHAR},#{item.nemtAspiration,jdbcType=VARCHAR},#{item.nemtCode,jdbcType=VARCHAR},#{item.nemtScore,jdbcType=VARCHAR},#{item.studentId,jdbcType=VARCHAR},#{item.identityCardId,jdbcType=VARCHAR},#{item.level,jdbcType=VARCHAR},#{item.major,jdbcType=VARCHAR},#{item.name,jdbcType=VARCHAR},#{item.nation,jdbcType=VARCHAR},#{item.secondaryCollege,jdbcType=VARCHAR},#{item.operator,jdbcType=VARCHAR},#{item.sex,jdbcType=VARCHAR},0,#{item.accountAddress,jdbcType=VARCHAR},#{item.nativePlace,jdbcType=VARCHAR},#{item.originalPlace,jdbcType=VARCHAR},#{item.usedName,jdbcType=VARCHAR},#{item.pictrue,jdbcType=VARCHAR},#{item.joinPartyDate,jdbcType=VARCHAR},#{item.politicalStatus,jdbcType=VARCHAR},#{item.telNum,jdbcType=VARCHAR},#{item.isRegistry,jdbcType=TINYINT},#{item.graduateSchool,jdbcType=VARCHAR},now(),now()        )   </foreach>                </insert>

代码说明：

底层的mapper是通过逆向工程来生成的，批量插入如下，是拼接成类似：insert into tb_enroll_student()values ()，()…….();

这样的缺点是，数据库一般有一个默认的设置，就是每次sql操作的数据不能超过4M。这样插入，数据多的时候，数据库会报错Packet for query is too large (6071393 > 4194304). You can change this value on the server by setting the max_allowed_packet' variable.，虽然我们可以通过

类似修改 my.ini 加上 max_allowed_packet =67108864，67108864=64M，默认大小4194304 也就是4M

修改完成之后要重启mysql服务，如果通过命令行修改就不用重启mysql服务。

完成本次操作，但是我们不能保证项目单次最大的大小是多少，这样是有弊端的。所以可以考虑进行分组导入。

| 分组把list导入Mysql中

同样适用mybatis批量插入，区别是对每次的导入进行分组计算，然后分多次进行导入：

@Transactional(rollbackFor = Exception.class)public int addFreshStudentsNew2(List<FreshStudentAndStudentModel> list, String schoolNo) {if (list == null || list.isEmpty()) {return 0;}List<StudentEntity> studentEntityList = new LinkedList<>();List<EnrollStudentEntity> enrollStudentEntityList = new LinkedList<>();List<AllusersEntity> allusersEntityList = new LinkedList<>();for (FreshStudentAndStudentModel freshStudentAndStudentModel : list) {EnrollStudentEntity enrollStudentEntity = new EnrollStudentEntity();StudentEntity studentEntity = new StudentEntity();BeanUtils.copyProperties(freshStudentAndStudentModel, studentEntity);BeanUtils.copyProperties(freshStudentAndStudentModel, enrollStudentEntity);String operator = TenancyContext.UserID.get();String studentId = BaseUuidUtils.base58Uuid();enrollStudentEntity.setId(BaseUuidUtils.base58Uuid());enrollStudentEntity.setStudentId(studentId);enrollStudentEntity.setIdentityCardId(freshStudentAndStudentModel.getIdCard());enrollStudentEntity.setOperator(operator);studentEntity.setId(studentId);studentEntity.setIdentityCardId(freshStudentAndStudentModel.getIdCard());studentEntity.setOperator(operator);studentEntityList.add(studentEntity);enrollStudentEntityList.add(enrollStudentEntity);AllusersEntity allusersEntity = new AllusersEntity();allusersEntity.setId(enrollStudentEntity.getId());allusersEntity.setUserCode(enrollStudentEntity.getNemtCode());allusersEntity.setUserName(enrollStudentEntity.getName());allusersEntity.setSchoolNo(schoolNo);allusersEntity.setTelNum(enrollStudentEntity.getTelNum());allusersEntity.setPassword(enrollStudentEntity.getNemtCode());  //密码设置为考生号allusersEntityList.add(allusersEntity);}int c = 100;int b = enrollStudentEntityList.size() / c;int d = enrollStudentEntityList.size() % c;int enResult = 0;int stuResult = 0;boolean allResult = false;for (int e = c; e <= c * b; e = e + c) {enResult = enrollStudentDao.insertAll(enrollStudentEntityList.subList(e - c, e));stuResult = studentDao.insertAll(studentEntityList.subList(e - c, e));allResult = allusersFacade.insertUserList(allusersEntityList.subList(e - c, e));}if (d != 0) {enResult = enrollStudentDao.insertAll(enrollStudentEntityList.subList(c * b, enrollStudentEntityList.size()));stuResult = studentDao.insertAll(studentEntityList.subList(c * b, studentEntityList.size()));allResult = allusersFacade.insertUserList(allusersEntityList.subList(c * b, allusersEntityList.size()));}if (enResult > 0 && stuResult > 0 && allResult) {return 10;}return -10;}

代码说明：

这样操作，可以避免上面的错误，但是分多次插入，无形中就增加了操作实践，很容易超时。所以这种方法还是不值得提倡的。

再次改进，使用多线程分批导入。

| 多线程分批导入Mysql

依然使用mybatis的批量导入，不同的是，根据线程数目进行分组，然后再建立多线程池，进行导入。

@Transactional(rollbackFor = Exception.class)public int addFreshStudentsNew(List<FreshStudentAndStudentModel> list, String schoolNo) {if (list == null || list.isEmpty()) {return 0;}List<StudentEntity> studentEntityList = new LinkedList<>();List<EnrollStudentEntity> enrollStudentEntityList = new LinkedList<>();List<AllusersEntity> allusersEntityList = new LinkedList<>();list.forEach(freshStudentAndStudentModel -> {EnrollStudentEntity enrollStudentEntity = new EnrollStudentEntity();StudentEntity studentEntity = new StudentEntity();BeanUtils.copyProperties(freshStudentAndStudentModel, studentEntity);BeanUtils.copyProperties(freshStudentAndStudentModel, enrollStudentEntity);String operator = TenancyContext.UserID.get();String studentId = BaseUuidUtils.base58Uuid();enrollStudentEntity.setId(BaseUuidUtils.base58Uuid());enrollStudentEntity.setStudentId(studentId);enrollStudentEntity.setIdentityCardId(freshStudentAndStudentModel.getIdCard());enrollStudentEntity.setOperator(operator);studentEntity.setId(studentId);studentEntity.setIdentityCardId(freshStudentAndStudentModel.getIdCard());studentEntity.setOperator(operator);studentEntityList.add(studentEntity);enrollStudentEntityList.add(enrollStudentEntity);AllusersEntity allusersEntity = new AllusersEntity();allusersEntity.setId(enrollStudentEntity.getId());allusersEntity.setUserCode(enrollStudentEntity.getNemtCode());allusersEntity.setUserName(enrollStudentEntity.getName());allusersEntity.setSchoolNo(schoolNo);allusersEntity.setTelNum(enrollStudentEntity.getTelNum());allusersEntity.setPassword(enrollStudentEntity.getNemtCode());  //密码设置为考生号allusersEntityList.add(allusersEntity);});int nThreads = 50;int size = enrollStudentEntityList.size();ExecutorService executorService = Executors.newFixedThreadPool(nThreads);List<Future<Integer>> futures = new ArrayList<Future<Integer>>(nThreads);for (int i = 0; i < nThreads; i++) {final List<EnrollStudentEntity> EnrollStudentEntityImputList = enrollStudentEntityList.subList(size / nThreads * i, size / nThreads * (i + 1));final List<StudentEntity> studentEntityImportList = studentEntityList.subList(size / nThreads * i, size / nThreads * (i + 1));final List<AllusersEntity> allusersEntityImportList = allusersEntityList.subList(size / nThreads * i, size / nThreads * (i + 1));Callable<Integer> task1 = () -> {studentSave.saveStudent(EnrollStudentEntityImputList,studentEntityImportList,allusersEntityImportList);return 1;};futures.add(executorService.submit(task1));}executorService.shutdown();if (!futures.isEmpty() && futures != null) {return 10;}return -10;}

代码说明：

上面是通过应用ExecutorService 建立了固定的线程数，然后根据线程数目进行分组，批量依次导入。一方面可以缓解数据库的压力，另一个面线程数目多了，一定程度会提高程序运行的时间。缺点就是要看服务器的配置，如果配置好的话就可以开多点线程，配置差的话就开小点。

| 小结

通过使用这个操作真是不断的提高了，项目使用技巧也是不错。加油~~ 多线程哦~~

来源：https://blog.csdn.net/kisscatforever

/article/details/79817039