DBCP数据库连接打满原因分析

实验背景

近一年来发生几起的数据库连接被打满的情况，初步分析是应用使用连接数量“超过了”连接池（DBCP1.4）的上限，导致数据库连接被打满，其中一个结论是连接池的bug导致

问题分析

1、DBCP连接池的Bug导致连接数超过大小设置，其根本原因是在连接池初始化的时候，有可能创建多个连接池导致

1. protected synchronized DataSource createDataSource()

2. throws SQLException {

3. if (closed) {

4. throw new SQLException("Data source is closed");

5. }

6. // Return the pool if we have already created it

7. if (dataSource != null) {

8. return (dataSource);

9. }

10. // create factory which returns raw physical connections

11. ConnectionFactory driverConnectionFactory = createConnectionFactory();

12. // create a pool for our connections

13. createConnectionPool();

14. // Set up statement pool, if desired

15. GenericKeyedObjectPoolFactory statementPoolFactory = null;

16. if (isPoolPreparedStatements()) {

17. statementPoolFactory = new GenericKeyedObjectPoolFactory(null,

18. -1, // unlimited maxActive (per key)

19. GenericKeyedObjectPool.WHEN_EXHAUSTED_FAIL,

20. 0, // maxWait

21. 1, // maxIdle (per key)

22. maxOpenPreparedStatements);

23. }

24. // Set up the poolable connection factory

25. createPoolableConnectionFactory(driverConnectionFactory, statementPoolFactory, abandonedConfig);

26. // Create and return the pooling data source to manage the connections

27. createDataSourceInstance();

28. try {

29. for (int i = 0 ; i < initialSize ; i++) {

30. connectionPool.addObject();

31. }

32. } catch (Exception e) {

33. throw new SQLNestedException("Error preloading the connection pool", e);

34. }

35. return dataSource;

36. }

createDataSource方法里边调用createConnectionPool方法，如果后面方式执行失败，比如createPoolableConnectionFactory，当再次调用createDataSource，又会再次调用createConnectionPool，从而初始化多次连接池，下面是createConnectionPool

1. protected void createConnectionPool() {

2. // Create an object pool to contain our active connections

3. GenericObjectPool gop;

4. if ((abandonedConfig != null) && (abandonedConfig.getRemoveAbandoned())) {

5. gop = new AbandonedObjectPool(null,abandonedConfig);

6. }

7. else {

8. gop = new GenericObjectPool();

9. }

10. gop.setMaxActive(maxActive);

11. gop.setMaxIdle(maxIdle);

12. gop.setMinIdle(minIdle);

13. gop.setMaxWait(maxWait);

14. gop.setTestOnBorrow(testOnBorrow);

15. gop.setTestOnReturn(testOnReturn);

16. gop.setTimeBetweenEvictionRunsMillis(timeBetweenEvictionRunsMillis);

17. gop.setNumTestsPerEvictionRun(numTestsPerEvictionRun);

18. gop.setMinEvictableIdleTimeMillis(minEvictableIdleTimeMillis);

19. gop.setTestWhileIdle(testWhileIdle);

20. connectionPool = gop;

21. }

显然这个情况只能是模块启动初始化的时候产生，且连接都不会被使用，而线上出现的问题是连接都在执行sql，不是这种情况！在1.4x的这个bug已经修复
2、启动参数分析，从启动参数入手，看一下是不是连接池主动释放连接，导致正在使用的连接被释放，从而创建新的连接，给我们的现象是“连接数”超过了连接池的限制，下面是一个入库的连接池参数

1. partition1.driverClassName=com.mysql.jdbc.Driver

2. partition1.initialSize=2

3. partition1.maxActive=25

4. partition1.minIdle=2

5. partition1.maxIdle=5

6. partition1.maxWait=3000

7. partition1.threadPoolSize=10

8. partition1.logAbandoned=true

9. partition1.testWhileIdle=true

10. partition1.testOnReturn=false

11. partition1.testOnBorrow=true

12. partition1.validationQuery=select now()

13. //在每次空闲连接回收器线程(如果有)运行时检查的连接数量

14. partition1.numTestsPerEvictionRun=5

15. //在空闲连接回收器线程运行期间休眠的时间值,以毫秒为单位

16. partition1.timeBetweenEvictionRunsMillis=30000

17. //连接在池中保持空闲而不被空闲连接回收器线程

18. partition1.minEvictableIdleTimeMillis=180000

19. //设置了rmoveAbandoned=true 那么在getNumActive()快要到getMaxActive()的时候，系统会进行无效的Connection的回收，回收的 Connection为removeAbandonedTimeout(默认300秒)中设置的秒数后没有使用的Connection

20. partition1.removeAbandoned=true

21. //强制回收连接的时间，单位秒

22. partition1.removeAbandonedTimeout=18

注意connectionProperties参数

1. <bean id="partition[j]" class="org.apache.commons.dbcp.BasicDataSource"

2. destroy-method="close">

3. <property name="driverClassName" value="${partition[j].driverClassName}" ></property>

4. <property name="url" value="${partition[j].url}" ></property>

5. <property name="username" value="${partition[j].username}" ></property>

6. <property name="password" value="${partition[j].password}" ></property>

7. <property name="defaultAutoCommit" value="false" ></property>

8. <property name="maxActive" value="${partition[j].maxActive}" ></property>

9. <property name="maxIdle" value="${partition[j].maxIdle}" ></property>

10. <property name="maxWait" value="${partition[j].maxWait}" ></property>

11. <property name="initialSize" value="${partition[j].initialSize}" ></property>

12. <property name="minIdle" value="${partition[j].minIdle}" ></property>

13. <property name="logAbandoned" value="${partition[j].logAbandoned}" ></property>

14. <property name="testWhileIdle" value="${partition[j].testWhileIdle}" ></property>

15. <property name="testOnReturn" value="${partition[j].testOnReturn}" ></property>

16. <property name="testOnBorrow" value="${partition[j].testOnBorrow}" ></property>

17. <property name="validationQuery" value="${partition[j].validationQuery}" ></property>

18. <property name="numTestsPerEvictionRun" value="${partition[j].numTestsPerEvictionRun}" ></property>

19. <property name="timeBetweenEvictionRunsMillis" value="${partition[j].timeBetweenEvictionRunsMillis}" ></property>

20. <property name="minEvictableIdleTimeMillis" value="${partition[j].minEvictableIdleTimeMillis}" ></property>

21. <property name="removeAbandoned" value="${partition[j].removeAbandoned}" ></property>

22. <property name="removeAbandonedTimeout" value="${partition[j].removeAbandonedTimeout}" ></property>

23. <property name="connectionProperties" value="useUnicode=true;

24. characterEncoding=utf8;initialTimeout=1;connectTimeout=1000;socketTimeout=6000;

25. rewriteBatchedStatements=true;autoReconnectForPools=true;autoReconnect=true;maxReconnects=1;

26. failOverReadOnly=false;roundRobinLoadBalance=true;allowMultiQueries=true"></property>

27. </bean>

从上面参数我们重点关注removeAbandonedTimeout 这个参数的意义

1. //创建连接

2. public Object borrowObject() throws Exception {

3. if (config != null

4. && config.getRemoveAbandoned()

5. && (getNumIdle() < 2)

6. && (getNumActive() > getMaxActive() - 3) ) {

7. removeAbandoned();

8. }

9. Object obj = super.borrowObject();

10. if (obj instanceof AbandonedTrace) {

11. ((AbandonedTrace) obj).setStackTrace();

12. }

13. if (obj != null && config != null && config.getRemoveAbandoned()) {

14. synchronized (trace) {

15. trace.add(obj);

16. }

17. }

18. ...

19. private void removeAbandoned() {

20. // Generate a list of abandoned connections to remove

21. long now = System.currentTimeMillis();

22. long timeout = now - (config.getRemoveAbandonedTimeout() * 1000);

23. ArrayList remove = new ArrayList();

24. synchronized (trace) {

25. Iterator it = trace.iterator();

26. while (it.hasNext()) {

27. AbandonedTrace pc = (AbandonedTrace) it.next();

28. if (pc.getLastUsed() > timeout) {

29. continue;

30. }

31. if (pc.getLastUsed() > 0) {

32. remove.add(pc);

33. }

34. }

35. }

36. // Now remove the abandoned connections

37. Iterator it = remove.iterator();

38. while (it.hasNext()) {

39. AbandonedTrace pc = (AbandonedTrace) it.next();

40. if (config.getLogAbandoned()) {

41. pc.printStackTrace();

42. }

43. try {

44. invalidateObject(pc);

45. } catch (Exception e) {

46. e.printStackTrace();

47. }

48. }

49. }

代从代码可以看出，配置了removeAbandonedOnBorrow和removeAbandonedTimeout就是触发正在使用的连接remove，假如这个时候，数据库正在执行sql，就会出现连接池已经断开连接，而数据库的连接还在执行，造成数据库的连接超过连接池的显现，如果少量慢sql执行时间大于removeAbandonedTimeout，不会出现问题，出现大量的慢sql，就会导致数据库中的这种连接越来越多，最后把数据库打满

实验过程

创建测试库

1. mkdir /usr/local/data/mysql

2. docker run -d -e MYSQL_ROOT_PASSWORD=root --name centos/mysql-57-centos7 -v /usr/local/data/mysql:/var/lib/mysql -p 3306:3306 mysql --character-set-server=utf8mb4 --collation-server=utf8mb4_unicode_ci --lower_case_table_names=1

创建测试库

1. create database test ;

构建测试类，核心代码

1. public void setUp() throws Exception {

2. ds = createDataSource();

3. ds.setDriverClassName("com.mysql.jdbc.Driver");

4. ds.setUrl("jdbc:mysql://127.0.0.1:3306/test?allowMultiQueries=true&characterEncoding=UTF-8&autoReconnect=true&failOverReadOnly=false&useSSL=false&allowPublicKeyRetrieval=true");

5. ds.setUsername("root");

6. ds.setPassword("Htbuy@2016");

7. ds.setMaxActive(1);

8. ds.setMaxWait(1000);

9. ds.setTestWhileIdle(true);

10. ds.setTestOnBorrow(true);

11. ds.setTestOnReturn(false);

12. ds.setValidationQuery("select now()");

13. ds.setNumTestsPerEvictionRun(5);

14. ds.setMinEvictableIdleTimeMillis(2000);

15. ds.setLogAbandoned(true);

16. ds.setRemoveAbandoned(true);

17. ds.setRemoveAbandonedTimeout(1);

18. }

19. public void testAbandoned() throws Exception {

20. for (int i = 0; i < 20; i++) {

21. Thread t = new Thread(new Runnable() {

22. @Override

23. public void run() {

24. try {

25. Connection conn = ds.getConnection();

26. Statement statement = conn.createStatement();

27. ResultSet resultSet = null;

28. #模拟慢sql

29. resultSet = statement.executeQuery("select sleep(100),now()");

30. while (resultSet.next()) {

31. System.out.println("result+" + resultSet.getString(1));

32. }

33. resultSet.close();

34. statement.close();

35. conn.close();

36. } catch (Exception ex) {

37. System.out.println(ex.getMessage());

38. }

39. System.out.println(Thread.currentThread().getName() + "---------------------- end----------------------");

40. }

41. });

42. t.setName(i + "");

43. t.start();

44. Thread.sleep((i + 1) * 1000);

45. }

46. System.out.println(Thread.currentThread().getName() + "---------------------- end----------------------");

47. Thread.sleep(1000000);

48. }

通过运行上面代码，并没有像我们所期望那样，超过连接的最大限制1，通过RemoveAbandoned=true和RemoveAbandonedTimeout=1 应该已经触发了连接池的Abandoned机制，但是都阻塞到下面的一行代码，DelegatingStatement的close方法

1. /**

2. * Close this DelegatingStatement, and close

3. * any ResultSets that were not explicitly closed.

4. */

5. public void close() throws SQLException {

6. try {

7. try {

8. if (_conn != null) {

9. _conn.removeTrace(this);

10. _conn = null;

11. }

12. // The JDBC spec requires that a statment close any open

13. // ResultSet's when it is closed.

14. // FIXME The PreparedStatement we're wrapping should handle this for us.

15. // See bug 17301 for what could happen when ResultSets are closed twice.

16. List resultSets = getTrace();

17. if( resultSets != null) {

18. ResultSet[] set = (ResultSet[]) resultSets.toArray(new ResultSet[resultSets.size()]);

19. for (int i = 0; i < set.length; i++) {

20. set[i].close();

21. }

22. clearTrace();

23. }

24. //阻塞地方

25. _stmt.close();

26. }

27. catch (SQLException e) {

28. handleException(e);

29. }

30. }

31. finally {

32. _closed = true;

33. }

34. }

当我们在url设置socketTimeout=1000，这时候阻塞的地方成功执行完成，超过数据库出现大量的连接数1，问题重现！同时出现典型的日志The last packet successfully received from the server was 1,001 milliseconds ago. The last packet sent successfully to the server was 1,001 milliseconds ago.

当设置socketTimeout=1000，RemoveAbandoned=false ，并没有重现问题，且都是大量的等待连接超时

原因是下面的代码差异，开启RemoveAbandoned，使用的是AbandonedObjectPool，而不开启是默认的GenericObjectPool，AbandonedObjectPool增加了Abandoned逻辑

1. */

2. protected void createConnectionPool() {

3. // Create an object pool to contain our active connections

4. GenericObjectPool gop;

5. if ((abandonedConfig != null) && (abandonedConfig.getRemoveAbandoned())) {

6. gop = new AbandonedObjectPool(null,abandonedConfig);

7. }

8. else {

9. gop = new GenericObjectPool();

10. }

11. gop.setMaxActive(maxActive);

12. gop.setMaxIdle(maxIdle);

13. gop.setMinIdle(minIdle);

14. gop.setMaxWait(maxWait);

15. gop.setTestOnBorrow(testOnBorrow);

16. gop.setTestOnReturn(testOnReturn);

17. gop.setTimeBetweenEvictionRunsMillis(timeBetweenEvictionRunsMillis);

18. gop.setNumTestsPerEvictionRun(numTestsPerEvictionRun);

19. gop.setMinEvictableIdleTimeMillis(minEvictableIdleTimeMillis);

20. gop.setTestWhileIdle(testWhileIdle);

21. connectionPool = gop;

22. }

总结

当RemoveAbandoned=true， 且执行时间超过socketTimeout ，达到RemoveAbandonedTimeout的触发点时，就会导致数据库连接数超过连接池的限制，注意这种情况关闭模块是没用的，sql还在数据库中执行，应该直接kill或者切库操作！！

优化方案

1、关闭RemoveAbandoned设置，目前wms系统的还是AP和TP混合型，有很大的几率触发这种情况
2、经过邱玉堃复核模块测试，需要关闭socketTimeout参数，引用的1.4.jar版本与源码的1.4版本存在差异
3、 注意当mysql-connector-java 版本小于5.1.45就会出现这个bug

---

自猿其说Tech-JDL京东物流技术发展部

作者：供应链技术部-仓配研发组 彭大伟

参考阅读：

技术原创及架构实践文章，欢迎通过公众号菜单「联系我们」进行投稿。

高可用架构

改变互联网的构建方式