iBatis batch处置那些事
iBatis batch处理那些事
昨天应同事要求在框架中（Spring+iBatis2.3.4）加入Batch处理，于是满足之，由于需要更灵活并且不想为批量插入、批量更新、批量删除等操作单独写对应的方法，于是写了这样的一个方法
不想将SqlMapExecutor侵入到业务代码中，于是又有了如下3个类，在今天的主题中不是关键，可以忽略，只是为了将代码贴完整
然后执行了一个类似以下伪代码行为的操作：
再然后...尼玛不但速度没变快还异常了，原因竟然是生成了太多的PreparedStatement，你妹不是批处理吗？不是应该一个sql只有一个PreparedStatement吗？
跟踪iBatis代码，发现了iBatis是这样处理的
不细解释了，只看currentSql这个实例变量就知道了，如果sql与前一次不同那么会新建一个PreparedStatement，所以刚才的伪代码应该这样写：
很不爽是不？反正我是决了一个定，改iBatis的源码
改源码最好这么来：
1.复制对应类的源码到工程下，本例中要复制的是com.ibatis.sqlmap.engine.execution.SqlExecutor
注意要保持包名，其实是类的全限定名称要一样哇，这样根据ClassLoader的类加载机制会先加载你工程中的SqlExecutor而不加载iBatis jar包中的对应SqlExecutor
如图：
2.改之，只改static class Batch这个内部类即可，策略是去掉currentSql，将PreparedStatement放入HashMap
如下：
下面贴出修改后完整的代码，方便有同样需求的同学修改，只贴出内部类com.ibatis.sqlmap.engine.execution.SqlExecutor$Batch，com.ibatis.sqlmap.engine.execution.SqlExecutor没有做出任何修改
昨天应同事要求在框架中（Spring+iBatis2.3.4）加入Batch处理，于是满足之，由于需要更灵活并且不想为批量插入、批量更新、批量删除等操作单独写对应的方法，于是写了这样的一个方法
public Object batchExecute(final CallBack callBack) {
Object result = getSqlMapClientTemplate().execute(new SqlMapClientCallback<Object>() {
@Override
public Object doInSqlMapClient(SqlMapExecutor executor) throws SQLException {
executor.startBatch();
Object obj = callBack.execute(new IbatisSqlExecutor(executor));
executor.executeBatch();
return obj;
}
});
return result;
}
不想将SqlMapExecutor侵入到业务代码中，于是又有了如下3个类，在今天的主题中不是关键，可以忽略，只是为了将代码贴完整
public interface SqlExecutor {
Object insert(String id, Object parameterObject) throws SQLException;
Object insert(String id) throws SQLException;
int update(String id, Object parameterObject) throws SQLException;
int update(String id) throws SQLException;
int delete(String id, Object parameterObject) throws SQLException;
int delete(String id) throws SQLException;
Object queryForObject(String id, Object parameterObject) throws SQLException;
Object queryForObject(String id) throws SQLException;
Object queryForObject(String id, Object parameterObject, Object resultObject) throws SQLException;
}
class IbatisSqlExecutor implements SqlExecutor {
private SqlMapExecutor executor;
IbatisSqlExecutor(SqlMapExecutor executor) {
this.executor = executor;
}
@Override
public Object insert(String id, Object parameterObject) throws SQLException {
return executor.insert(id, parameterObject);
}
// 剩下的就省略了，和insert都类似
}
public interface CallBack {
Object execute(SqlExecutor executor);
}
然后执行了一个类似以下伪代码行为的操作：
getDao().batchExecute(new CallBack() {
@Override
public Object execute(SqlExecutor executor) {
for (int i = 0; i < 10000; ++i) {
// 注意每个sql_id的sql语句都是不相同的
executor.insert("id1", obj1);
executor.insert("id2", obj2);
// ...
executor.insert("idn", objn);
}
return null;
}
});
再然后...尼玛不但速度没变快还异常了，原因竟然是生成了太多的PreparedStatement，你妹不是批处理吗？不是应该一个sql只有一个PreparedStatement吗？
跟踪iBatis代码，发现了iBatis是这样处理的
// 以下代码来自com.ibatis.sqlmap.engine.execution.SqlExecutor$Batch
public void addBatch(StatementScope statementScope, Connection conn, String sql, Object[] parameters) throws SQLException {
PreparedStatement ps = null;
// 就是它：currentSql
if (currentSql != null && currentSql.equals(sql)) {
int last = statementList.size() - 1;
ps = (PreparedStatement) statementList.get(last);
} else {
ps = prepareStatement(statementScope.getSession(), conn, sql);
setStatementTimeout(statementScope.getStatement(), ps);
// 就是它：currentSql
currentSql = sql;
statementList.add(ps);
batchResultList.add(new BatchResult(statementScope.getStatement().getId(), sql));
}
statementScope.getParameterMap().setParameters(statementScope, ps, parameters);
ps.addBatch();
size++;
}
不细解释了，只看currentSql这个实例变量就知道了，如果sql与前一次不同那么会新建一个PreparedStatement，所以刚才的伪代码应该这样写：
getDao().batchExecute(new CallBack() {
@Override
public Object execute(SqlExecutor executor) {
for (int i = 0; i < 10000; ++i) {
executor.insert("id1", obj1);
}
for (int i = 0; i < 10000; ++i) {
executor.insert("id2", obj2);
}
// ...你就反复写for循环吧
return null;
}
});
很不爽是不？反正我是决了一个定，改iBatis的源码
改源码最好这么来：
1.复制对应类的源码到工程下，本例中要复制的是com.ibatis.sqlmap.engine.execution.SqlExecutor
注意要保持包名，其实是类的全限定名称要一样哇，这样根据ClassLoader的类加载机制会先加载你工程中的SqlExecutor而不加载iBatis jar包中的对应SqlExecutor
如图：
2.改之，只改static class Batch这个内部类即可，策略是去掉currentSql，将PreparedStatement放入HashMap
如下：
public void addBatch(StatementScope statementScope, Connection conn, String sql, Object[] parameters) throws SQLException {
PreparedStatement ps = statementMap.get(sql);
if (ps == null) {
ps = prepareStatement(statementScope.getSession(), conn, sql);
setStatementTimeout(statementScope.getStatement(), ps);
statementMap.put(sql, ps);
batchResultMap.put(sql, new BatchResult(statementScope.getStatement().getId(), sql));
}
statementScope.getParameterMap().setParameters(statementScope, ps, parameters);
ps.addBatch();
++size;
}
下面贴出修改后完整的代码，方便有同样需求的同学修改，只贴出内部类com.ibatis.sqlmap.engine.execution.SqlExecutor$Batch，com.ibatis.sqlmap.engine.execution.SqlExecutor没有做出任何修改
private static class Batch {
private Map<String, PreparedStatement> statementMap = new HashMap<String, PreparedStatement>();
private Map<String, BatchResult> batchResultMap = new HashMap<String, BatchResult>();
private int size;
/**
* Create a new batch
*/
public Batch() {
size = 0;
}
/**
* Getter for the batch size
* @return - the batch size
*/
@SuppressWarnings("unused")
public int getSize() {
return size;
}
/**
* Add a prepared statement to the batch
* @param statementScope - the request scope
* @param conn - the database connection
* @param sql - the SQL to add
* @param parameters - the parameters for the SQL
* @throws SQLException - if the prepare for the SQL fails
*/
public void addBatch(StatementScope statementScope, Connection conn, String sql, Object[] parameters) throws SQLException {
PreparedStatement ps = statementMap.get(sql);
if (ps == null) {
ps = prepareStatement(statementScope.getSession(), conn, sql);
setStatementTimeout(statementScope.getStatement(), ps);
statementMap.put(sql, ps);
batchResultMap.put(sql, new BatchResult(statementScope.getStatement().getId(), sql));
}
statementScope.getParameterMap().setParameters(statementScope, ps, parameters);
ps.addBatch();
++size;
}
/**
* TODO (Jeff Butler) - maybe this method should be deprecated in some release,
* and then removed in some even later release. executeBatchDetailed gives
* much more complete information.
* <p/>
* Execute the current session's batch
* @return - the number of rows updated
* @throws SQLException - if the batch fails
*/
public int executeBatch() throws SQLException {
int totalRowCount = 0;
for (Map.Entry<String, PreparedStatement> iter : statementMap.entrySet()) {
PreparedStatement ps = iter.getValue();
int[] rowCounts = ps.executeBatch();
for (int j = 0; j < rowCounts.length; j++) {
if (rowCounts[j] == Statement.SUCCESS_NO_INFO) {
// do nothing
} else if (rowCounts[j] == Statement.EXECUTE_FAILED) {
throw new SQLException("The batched statement at index " + j + " failed to execute.");
} else {
totalRowCount += rowCounts[j];
}
}
}
return totalRowCount;
}
/**
* Batch execution method that returns all the information
* the driver has to offer.
* @return a List of BatchResult objects
* @throws BatchException (an SQLException sub class) if any nested
* batch fails
* @throws SQLException if a database access error occurs, or the drive
* does not support batch statements
* @throws BatchException if the driver throws BatchUpdateException
*/
@SuppressWarnings({ "rawtypes", "unchecked" })
public List executeBatchDetailed() throws SQLException, BatchException {
List answer = new ArrayList();
int i = 0;
for (String sql : statementMap.keySet()) {
BatchResult br = batchResultMap.get(sql);
PreparedStatement ps = statementMap.get(sql);
try {
br.setUpdateCounts(ps.executeBatch());
} catch (BatchUpdateException e) {
StringBuffer message = new StringBuffer();
message.append("Sub batch number ");
message.append(i + 1);
message.append(" failed.");
if (i > 0) {
message.append(" ");
message.append(i);
message.append(" prior sub batch(s) completed successfully, but will be rolled back.");
}
throw new BatchException(message.toString(), e, answer, br.getStatementId(), br.getSql());
}
++i;
answer.add(br);
}
return answer;
}
/**
* Close all the statements in the batch and clear all the statements
* @param sessionScope
*/
public void cleanupBatch(SessionScope sessionScope) {
for (Map.Entry<String, PreparedStatement> iter : statementMap.entrySet()) {
PreparedStatement ps = iter.getValue();
closeStatement(sessionScope, ps);
}
statementMap.clear();
batchResultMap.clear();
size = 0;
}
}