本文从 Commons DBCP testOnBorrow 的作用机制着手,管中窥豹,从一点去分析数据库连接池获取的过程以及架构分层设计。
以下内容会按照每层的作用,贯穿分析整个调用流程。
The indication of whether objects will be validated before being borrowed from the pool.
If the object fails to validate, it will be dropped from the pool, and we will attempt to borrow another.
testOnBorrow 不是 dbcp 定义的,是commons-pool 定义的。commons-pool 详细的定义了资源池使用的一套规范和运行流程。
/**
* Borrow an object from the pool. get object from 资源池
* @see org.apache.commons.pool2.impl.GenericObjectPool#borrowObject(long)
*/
public T borrowObject(final long borrowMaxWaitMillis) throws Exception {
PooledObject<T> p = null;
// if validation fails, the instance is destroyed and the next available instance is examined.
// This continues until either a valid instance is returned or there are no more idle instances available.
while (p == null) {
// If there is one or more idle instance available in the pool,
// then an idle instance will be selected based on the value of getLifo(), activated and returned.
p = idleObjects.pollFirst();
if (p != null) {
// 设置 testOnBorrow 就会进行可用性校验
if (p != null && (getTestOnBorrow() || create && getTestOnCreate())) {
boolean validate = false;
Throwable validationThrowable = null;
try {
// 具体的校验实现由实现类完成。
// see org.apache.commons.dbcp2.PoolableConnectionFactory
validate = factory.validateObject(p);
} catch (final Throwable t) {
PoolUtils.checkRethrow(t);
validationThrowable = t;
}
if (!validate) {
try {
// 如果校验异常,会销毁该资源。
// obj is not valid and should be dropped from the pool
destroy(p);
destroyedByBorrowValidationCount.incrementAndGet();
} catch (final Exception e) {
// Ignore - validation failure is more important
}
p = null;
}
}
}
}
return p.getObject();
}
dbcp 是特定于管理数据库连接的资源池。
PoolableConnectionFactory is a PooledObjectFactory
PoolableConnection is a PooledObject
/**
* @see PoolableConnectionFactory#validateObject(PooledObject)
*/
@Override
public boolean validateObject(final PooledObject<PoolableConnection> p) {
try {
/**
* 检测资源池对象的创建时间,是否超过生存时间
* 如果超过 maxConnLifetimeMillis, 不再委托数据库连接进行校验,直接废弃改资源
* @see PoolableConnectionFactory#setMaxConnLifetimeMillis(long)
*/
validateLifetime(p);
// 委托数据库连接进行自我校验
validateConnection(p.getObject());
return true;
} catch (final Exception e) {
return false;
}
}
/**
* 数据库连接层的校验。具体到是否已关闭、是否与 server 连接可用
* @see Connection#isValid(int)
*/
public void validateConnection(final PoolableConnection conn) throws SQLException {
if(conn.isClosed()) {
throw new SQLException("validateConnection: connection closed");
}
conn.validate(_validationQuery, _validationQueryTimeout);
}
Returns true if the connection has not been closed and is still valid.
这个是 java.sql.Connection 定义的规范。具体实现根据对应数据库的driver 来完成。使用某种机制用来探测连接是否可用。
/**
* 调用 com.mysql.jdbc.MysqlIO, 发送ping 请求,检测是否可用
* 对比 H2 数据库,是通过获取当前事务级别来检测连接是否可以。但是忽略了 timeout 配置,毕竟是 demo 数据库 😅
*/
public synchronized boolean isValid(int timeout) throws SQLException {
if (this.isClosed()) {
return false;
} else {
try {
this.pingInternal(false, timeout * 1000);
return true;
} catch (Throwable var5) {
return false;
}
}
}
参考:MySQL 的连接时长控制--interactive_timeout和wait_timeout_翔云123456的博客-CSDN博客
在阅读 MySQL Driver 源码过程中,有个点要特别记录下。以 MySQL Driver 创建连接为例,用重试连接实现可用性,这就是高可用。🎯
高可用不是一个口号,也不是复杂的概念和公式。能够实实在在体系化的解决一类问题就是架构的目的。结合上述的架构分层,如果解决问题的方案通用性好,并且实现很优雅,就是好的架构。
// autoReconnect
public void createNewIO(boolean isForReconnect) throws SQLException {
synchronized (getConnectionMutex()) {
// jdbc.url autoReconnect 指定为 true,识别为 HighAvailability。emmm..... 🙉
if (!getHighAvailability()) {
connectOneTryOnly(isForReconnect, mergedProps);
return;
}
// maxReconnects 默认为 3,重试失败的提示就是: Attempted reconnect 3 times. Giving up.
connectWithRetries(isForReconnect, mergedProps);
}
}
作者:京东物流 杨攀
来源:京东云开发者社区