Testing MySQL Replication Connection

        目前项目中使用了MySQL replication，并通过LVS对slaves进行负载均衡，数据库连接池使用的是c3p0。在使用过程中发现, LVS TCP timeout可能导致数据库连接被切断，从而应用程序中报数据库连接异常。
    ReplicationConnection内部保持了两个数据库连接，分别是masterConnection和slaveConnection。实际生效的连接取决于连接的readOnly属性,即readOnly ? currentConnection=slaveConnection : currentConnection=masterConnection。
    c3p0的提供了两种处理空闲连接的机制，对应的配置参数分别是：idleConnectionTestPeriod和maxIdleTime。但是这两种机制在默认情况下对ReplicationConnection不奏效。原因如下：

idleConnectionTestPeriod：c3p0在定期检查ReplicationConnection时只是检查了其内部的currentConnection，如果在某段时间内连接的readOnly属性为false，那么c3p0只会检查masterConnection，而不会检测slaveConnection是否仍然有效。
maxIdleTime：如果应用程序在一段内只以readOnly=false使用某个ReplicationConnection，那么c3p0不会认为该连接空闲，而实际上其内部的slaveConnection可能已经超时。

    默认情况下，c3p0使用DefaultConnectionTester（通过connectionTesterClassName配置）进行连接检查（基于Query），该类有以下两个比较重要的方法：

public int activeCheckConnection(Connection c, String query, Throwable[] rootCauseOutParamHolder)：通常用于checkin，checkout和定期的连接检查。
public int statusOnException(Connection c, Throwable t, String query, Throwable[] rootCauseOutParamHolder)：在连接抛异常的情况下，c3p0通过此方法确认连接本身是否可用（即是否需要销毁此连接）。

    需要注意的是，MySQL connector也提供了一个实现了c3p0的ConnectionTester接口的类：MysqlConnectionTester。该类使用com.mysql.jdbc.Connection的ping方法（相对于执行query，ping更轻量级）来确认连接是否正常。笔者认为该类仍然不能正确检测ReplicationConnection。
    以下是笔者实现的一个ConnectionTester，用于检查MySQL ReplicationConnection:
import java.sql.Connection;
import java.sql.ResultSet;
import java.sql.SQLException;
import java.sql.Statement;

import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

import com.mchange.v2.c3p0.AbstractConnectionTester;
import com.mysql.jdbc.CommunicationsException;


public final class MysqlReplicationConnectionTester extends AbstractConnectionTester {
//
private static final Logger LOGGER = LoggerFactory.getLogger(MysqlReplicationConnectionTester.class);

//
private static final long serialVersionUID = -7348778746126099053L;

//
private static final String DEFAULT_QUERY = "SELECT 1";

public boolean equals(Object o) {
return (o != null && o.getClass() == MysqlReplicationConnectionTester.class);
}

public int hashCode() {
return MysqlReplicationConnectionTester.class.getName().hashCode();
}
   
public int activeCheckConnection(Connection c, String query, Throwable[] outParamCause) {
//
boolean readOnly = false;
boolean needRestoreReadOnly = false;
try {
//
readOnly = c.isReadOnly();

//
int r = checkConnection(c, query, outParamCause, readOnly);
if(r == CONNECTION_IS_OKAY) {
needRestoreReadOnly = true;
r = checkConnection(c, query, outParamCause, !readOnly);
}
return r;
} catch(Exception e) {
//
LOGGER.warn("the connection: " + c + " was marked invalid", e);
if (outParamCause != null) {
outParamCause[0] = e;
}
return CONNECTION_IS_INVALID;
} finally {
try {
if(needRestoreReadOnly) {
c.setReadOnly(readOnly);
}
} catch (SQLException e) {
LOGGER.error("failed to restore read only: " + readOnly + " on connection", e);
}
}
}

public int statusOnException(Connection c, Throwable t, String query, Throwable[] outParamCause) {
//
int r = checkConnectionOnException(c, t, query, outParamCause);

//
if(r != CONNECTION_IS_OKAY) {
if (outParamCause != null) {
outParamCause[0] = t;
}
}
return r;
}

private int checkConnection(Connection c, String query, Throwable[] outParamCause, Boolean readOnly) {
//
if (query == null || query.equals("")) {
query = DEFAULT_QUERY;
}

//
ResultSet rs = null;
Statement stmt = null;
try {
//
boolean ro = c.isReadOnly();
if(readOnly != null && readOnly != ro) {
c.setReadOnly(readOnly);
}

//
if(LOGGER.isInfoEnabled()) {
LOGGER.info("testing connection: {} with query: {}, read only: {}", new Object[]{c, query, ro});
}

//
stmt = c.createStatement();
rs = stmt.executeQuery(query);
return CONNECTION_IS_OKAY;
} catch (SQLException e) {
LOGGER.warn("failed to test connection: " + c + " with query: " + query + ", state: " + e.getSQLState(), e);
if (outParamCause != null) {
outParamCause[0] = e;
}
return CONNECTION_IS_INVALID;
} catch (Exception e) {
LOGGER.warn("failed to test connection: " + c + " with query: " + query, e);
if (outParamCause != null) {
outParamCause[0] = e;
}
return CONNECTION_IS_INVALID;
} finally {
closeQuietly(rs);
closeQuietly(stmt);
}
}

private int checkConnectionOnException(Connection c, Throwable t, String query, Throwable[] outParamCause) {
//
if (t instanceof CommunicationsException) {
return CONNECTION_IS_INVALID;
}

//
if (t instanceof SQLException) {
final String sqlState = ((SQLException) t).getSQLState();
if (sqlState != null && sqlState.startsWith("08")) {
return CONNECTION_IS_INVALID;
} else {
return CONNECTION_IS_OKAY;
}
}

// Runtime/Unchecked?
return CONNECTION_IS_INVALID;
}

private void closeQuietly(ResultSet rs) {
//
if(rs == null) {
return;
}

//
try {
rs.close();
} catch (SQLException e) {
LOGGER.warn("failed to close result set", e);
}
}

private void closeQuietly(Statement stat) {
//
if(stat == null) {
return;
}

//
try {
stat.close();
} catch (SQLException e) {
LOGGER.warn("failed to close statement", e);
}
        }
}
    需要注意的是，以上代码适用于MySQL 5.0。对于MySQL 5.1，需要修改checkConnectionOnException方法，如下：
 if (t instanceof CommunicationsException || "com.mysql.jdbc.exceptions.jdbc4.CommunicationsException".equals(throwable.getClass().getName())) {
    return CONNECTION_IS_INVALID;
}
 
    此外， 由于Spring Dao会转译SQLException，因此在Spring环境中，不能使用sqlState判断连接是否正常，而是需要使用基于query的方式，如下：
public int statusOnException(Connection c, Throwable t, String query, Throwable[] outParamCause) {
return checkConnection(c, query, outParamCause, null);
}