/* MariaDB Client for Java Copyright (c) 2012-2014 Monty Program Ab. This library is free software; you can redistribute it and/or modify it under the terms of the GNU Lesser General Public License as published by the Free Software Foundation; either version 2.1 of the License, or (at your option) any later version. This library is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more details. You should have received a copy of the GNU Lesser General Public License along with this library; if not, write to Monty Program Ab info@montyprogram.com. This particular MariaDB Client for Java file is work derived from a Drizzle-JDBC. Drizzle-JDBC file which is covered by subject to the following copyright and notice provisions: Copyright (c) 2009-2011, Marcus Eriksson Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. Neither the name of the driver nor the names of its contributors may not be used to endorse or promote products derived from this software without specific prior written permission. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ package org.mariadb.jdbc; import org.mariadb.jdbc.internal.com.read.resultset.SelectResultSet; import org.mariadb.jdbc.internal.logging.Logger; import org.mariadb.jdbc.internal.logging.LoggerFactory; import org.mariadb.jdbc.internal.protocol.Protocol; import org.mariadb.jdbc.internal.com.read.dao.*; import org.mariadb.jdbc.internal.util.exceptions.ExceptionMapper; import org.mariadb.jdbc.internal.util.Utils; import org.mariadb.jdbc.internal.util.Options; import org.mariadb.jdbc.internal.util.dao.ClientPrepareResult; import org.mariadb.jdbc.internal.util.scheduler.SchedulerServiceProviderHolder; import java.io.IOException; import java.io.InputStream; import java.nio.charset.Charset; import java.sql.*; import java.util.ArrayList; import java.util.Arrays; import java.util.List; import java.util.concurrent.ExecutionException; import java.util.concurrent.Future; import java.util.concurrent.ScheduledExecutorService; import java.util.concurrent.TimeUnit; import java.util.concurrent.locks.ReentrantLock; public class MariaDbStatement implements Statement, Cloneable { //timeout scheduler private static final ScheduledExecutorService timeoutScheduler = SchedulerServiceProviderHolder.getTimeoutScheduler(); protected static Logger logger = LoggerFactory.getLogger(MariaDbStatement.class); protected final ReentrantLock lock; /** * the protocol used to talk to the server. */ protected Protocol protocol; /** * the Connection object. */ protected MariaDbConnection connection; protected Future<?> timerTaskFuture; protected volatile boolean closed = false; protected List<String> batchQueries; protected int queryTimeout; protected long maxRows; protected Results results; protected int resultSetScrollType; protected boolean mustCloseOnCompletion = false; protected Options options; boolean isTimedout; volatile boolean executing; //are warnings cleared? protected boolean warningsCleared; protected int fetchSize; private boolean canUseServerTimeout; private int maxFieldSize; /** * Creates a new Statement. * * @param connection the connection to return in getConnection. * @param resultSetScrollType one of the following <code>ResultSet</code> constants: <code>ResultSet.TYPE_FORWARD_ONLY</code>, * <code>ResultSet.TYPE_SCROLL_INSENSITIVE</code>, or <code>ResultSet.TYPE_SCROLL_SENSITIVE</code> */ public MariaDbStatement(MariaDbConnection connection, int resultSetScrollType) { this.protocol = connection.getProtocol(); this.connection = connection; this.canUseServerTimeout = connection.canUseServerTimeout(); this.resultSetScrollType = resultSetScrollType; this.lock = this.connection.lock; this.options = this.protocol.getOptions(); this.results = new Results(this, protocol.getAutoIncrementIncrement()); } /** * Provide a "cleanup" method that can be called after unloading driver, to fix Tomcat's obscure classpath handling. */ public static void unloadDriver() { // nothing to do here, as scheduler is already daemon thread } /** * Clone statement. * @param connection connection * @return Clone statement. * @throws CloneNotSupportedException if any error occur. */ public MariaDbStatement clone(MariaDbConnection connection) throws CloneNotSupportedException { MariaDbStatement clone = (MariaDbStatement) super.clone(); clone.connection = connection; clone.protocol = connection.getProtocol(); clone.timerTaskFuture = null; clone.batchQueries = new ArrayList<>(); clone.results = new Results(clone, clone.protocol.getAutoIncrementIncrement()); clone.closed = false; clone.warningsCleared = true; clone.fetchSize = 0; clone.maxRows = 0; return clone; } /** * returns the protocol. * * @return the protocol used. */ public Protocol getProtocol() { return protocol; } // Part of query prolog - setup timeout timer protected void setTimerTask() { assert (timerTaskFuture == null); timerTaskFuture = timeoutScheduler.schedule(new Runnable() { @Override public void run() { try { isTimedout = true; protocol.cancelCurrentQuery(); } catch (Throwable e) { } } }, queryTimeout, TimeUnit.SECONDS); } protected void executeQueryPrologue() throws SQLException { executing = true; if (closed) { throw new SQLException("execute() is called on closed statement"); } protocol.prolog(maxRows, protocol.getProxy() != null, connection, this); if (queryTimeout != 0 && !canUseServerTimeout) { setTimerTask(); } } protected void stopTimeoutTask() { if (timerTaskFuture != null) { if (!timerTaskFuture.cancel(true)) { // could not cancel, task either started or already finished // we must now wait for task to finish to ensure state modifications are done try { timerTaskFuture.get(); } catch (InterruptedException e) { // reset interrupt status Thread.currentThread().interrupt(); } catch (ExecutionException e) { // ignore error, likely due to interrupting during cancel } // we don't catch the exception if already canceled, that would indicate we tried // to cancel in parallel (which this code currently is not designed for) } timerTaskFuture = null; } } /** * Reset timeout after query, re-throw SQL exception * * @param sqle current exception * @return SQLException exception with new message in case of timer timeout. */ protected SQLException executeExceptionEpilogue(SQLException sqle) { //if has a failover, closing the statement if (sqle.getSQLState() != null && sqle.getSQLState().startsWith("08")) { try { close(); } catch (SQLException sqlee) { //eat exception } } if (isTimedout) { return new SQLTimeoutException("(conn:" + getServerThreadId() + ") Query timed out", "JZ0002", 1317, sqle); } SQLException sqlException = ExceptionMapper.getException(sqle, connection, this, queryTimeout != 0); logger.error("error executing query", sqlException); return sqlException; } protected void executeEpilogue() { stopTimeoutTask(); isTimedout = false; executing = false; } protected void executeBatchEpilogue() { executing = false; stopTimeoutTask(); isTimedout = false; clearBatch(); } protected SQLException handleFailoverAndTimeout(SQLException sqle) { //if has a failover, closing the statement if (sqle.getSQLState() != null && sqle.getSQLState().startsWith("08")) { try { close(); } catch (SQLException sqlee) { //eat exception } } if (isTimedout) { sqle = new SQLTimeoutException("(conn:" + getServerThreadId() + ") Query timed out", "JZ0002", 1317, sqle); } return sqle; } protected BatchUpdateException executeBatchExceptionEpilogue(SQLException sqle, CmdInformation cmdInformation, int size, boolean rewritten) { sqle = handleFailoverAndTimeout(sqle); int[] ret; if (cmdInformation == null) { ret = new int[size]; Arrays.fill(ret, Statement.EXECUTE_FAILED); } else if (rewritten) { ret = cmdInformation.getRewriteUpdateCounts(); } else { ret = cmdInformation.getUpdateCounts(); } sqle = ExceptionMapper.getException(sqle, connection, this, queryTimeout != 0); logger.error("error executing query", sqle); return new BatchUpdateException(sqle.getMessage(), sqle.getSQLState(), sqle.getErrorCode(), ret, sqle); } /** * Handle Exception for large batch update (return BatchUpdateException with long[]. * * @param exception initial exception * @param cmdInformation command return information (to indicate output that have been executed) * @param size initial batch length * @return a BatchUpdateException */ private BatchUpdateException executeLargeBatchExceptionEpilogue(SQLException exception, CmdInformation cmdInformation, int size) { exception = handleFailoverAndTimeout(exception); long[] ret; if (cmdInformation == null) { ret = new long[size]; Arrays.fill(ret, Statement.EXECUTE_FAILED); } else ret = cmdInformation.getLargeUpdateCounts(); exception = ExceptionMapper.getException(exception, connection, this, getQueryTimeout() != 0); logger.error("error executing query", exception); return new BatchUpdateException(exception.getMessage(), exception.getSQLState(), exception.getErrorCode(), ret, exception); } /** * executes a query. * * @param sql the query * @param fetchSize fetch size * @return true if there was a result set, false otherwise. * @throws SQLException the error description */ protected boolean executeInternal(String sql, int fetchSize) throws SQLException { lock.lock(); try { executeQueryPrologue(); results.reset(fetchSize, false, 1, false, resultSetScrollType); protocol.executeQuery(protocol.isMasterConnection(), results, getTimeoutSql(Utils.nativeSql(sql, connection.noBackslashEscapes))); results.commandEnd(); return results.getResultSet() != null; } catch (SQLException exception) { throw executeExceptionEpilogue(exception); } finally { executeEpilogue(); lock.unlock(); } } private String getTimeoutSql(String sql) { if (queryTimeout != 0 && canUseServerTimeout) { return "SET STATEMENT max_statement_time=" + queryTimeout + " FOR " + sql; } return sql; } /** * ! This method is for test only ! * This permit sending query using specific charset. * * @param sql sql * @param charset charset * @return boolean if execution went well * @throws SQLException if any exception occur */ public boolean testExecute(String sql, Charset charset) throws SQLException { lock.lock(); try { executeQueryPrologue(); results.reset(fetchSize, false, 1, false, resultSetScrollType); protocol.executeQuery(protocol.isMasterConnection(), results, getTimeoutSql(Utils.nativeSql(sql, connection.noBackslashEscapes)),charset); results.commandEnd(); return results.getResultSet() != null; } catch (SQLException exception) { throw executeExceptionEpilogue(exception); } finally { executeEpilogue(); lock.unlock(); } } /** * executes a query. * * @param sql the query * @return true if there was a result set, false otherwise. * @throws SQLException if the query could not be sent to server */ public boolean execute(String sql) throws SQLException { return executeInternal(sql, fetchSize); } /** * <p>Executes the given SQL statement, which may return multiple results, and signals the driver that any * auto-generated keys should be made available for retrieval. The driver will ignore this signal if the SQL * statement is not an <code>INSERT</code> statement, or an SQL statement able to return auto-generated keys * (the list of such statements is vendor-specific).</p> * <p>In some (uncommon) situations, a single SQL statement may return multiple result sets and/or update counts. * Normally you can ignore this unless you are (1) executing a stored procedure * that you know may return multiple results or (2) you are dynamically executing an unknown SQL string. </p> * The <code>execute</code> method * executes an SQL statement and indicates the form of the first result. You must then use the methods * <code>getResultSet</code> or * <code>getUpdateCount</code> to retrieve the result, and <code>getInternalMoreResults</code> to move to any * subsequent result(s). * * @param sql any SQL statement * @param autoGeneratedKeys a constant indicating whether auto-generated keys should be made available for retrieval * using the method<code>getGeneratedKeys</code>; one of the following constants: * <code>Statement.RETURN_GENERATED_KEYS</code> or * <code>Statement.NO_GENERATED_KEYS</code> * @return <code>true</code> if the first result is a <code>ResultSet</code> object; <code>false</code> if it is an * update count or there are no results * @throws SQLException if a database access error occurs, this method is called on a closed * <code>Statement</code> or the second parameter supplied to this method is * not <code>Statement.RETURN_GENERATED_KEYS</code> or * <code>Statement.NO_GENERATED_KEYS</code>. * @throws SQLFeatureNotSupportedException if the JDBC driver does not support this method with a constant of * Statement.RETURN_GENERATED_KEYS * @see #getResultSet * @see #getUpdateCount * @see #getMoreResults * @see #getGeneratedKeys */ public boolean execute(final String sql, final int autoGeneratedKeys) throws SQLException { return executeInternal(sql, fetchSize); } /** * Executes the given SQL statement, which may return multiple results, and signals the driver that the * auto-generated keys indicated in the given array should be made available for retrieval. This array contains * the indexes of the columns in the target table that contain the auto-generated keys that should be made * available. The driver will ignore the array if the SQL statement is not an <code>INSERT</code> * statement, or an SQL statement able to return auto-generated keys (the list of such statements is * vendor-specific). * <p>Under some (uncommon) situations, a single SQL statement may return multiple result sets and/or update counts. * Normally you can ignore this unless you are * (1) executing a stored procedure that you know may return multiple results or * (2) you are dynamically executing an unknown SQL string.</p> * The <code>execute</code> method executes an SQL statement and indicates the form of the first result. * You must then use the methods <code>getResultSet</code> or <code>getUpdateCount</code> to retrieve the result, * and <code>getInternalMoreResults</code> to move to any subsequent result(s). * * @param sql any SQL statement * @param columnIndexes an array of the indexes of the columns in the inserted row that should be made available * for retrieval by a call to the * method <code>getGeneratedKeys</code> * @return <code>true</code> if the first result is a <code>ResultSet</code> object; <code>false</code> if it is an * update count or there are no results * @throws SQLException if a database access error occurs, this method is called on a closed * <code>Statement</code> or the elements in * the <code>int</code> array passed to this method are not valid column * indexes * @throws SQLFeatureNotSupportedException if the JDBC driver does not support this method * @see #getResultSet * @see #getUpdateCount * @see #getMoreResults * @since 1.4 */ public boolean execute(final String sql, final int[] columnIndexes) throws SQLException { return executeInternal(sql, fetchSize); } /** * <p>Executes the given SQL statement, which may return multiple results, and signals the driver that the * auto-generated keys indicated in the given array should be made available for retrieval. This array contains * the names of the columns in the target table that contain the auto-generated keys that should be made available. * The driver will ignore the array if the SQL statement is not an <code>INSERT</code> * statement, or an SQL statement able to return auto-generated keys (the list of such statements is * vendor-specific). </p> * <p> In some (uncommon) situations, a single SQL statement may return multiple result sets and/or update counts. * Normally you can ignore this unless you are (1) executing a stored procedure that you know may return multiple * results or (2) you are dynamically executing an unknown SQL string. </p> * <p>The <code>execute</code> method executes an SQL statement and indicates the form of the first result. * You must then use the methods <code>getResultSet</code> or <code>getUpdateCount</code> to retrieve the result, * and <code>getInternalMoreResults</code> to move to any subsequent result(s).</p> * * @param sql any SQL statement * @param columnNames an array of the names of the columns in the inserted row that should be made available * for retrieval by a call to the method <code>getGeneratedKeys</code> * @return <code>true</code> if the next result is a <code>ResultSet</code> object; <code>false</code> if it is an * update count or there are no more results * @throws SQLException if a database access error occurs, this method is called on a closed * <code>Statement</code> or the elements of * the <code>String</code> array passed to this method are not valid column * names * @throws SQLFeatureNotSupportedException if the JDBC driver does not support this method * @see #getResultSet * @see #getUpdateCount * @see #getMoreResults * @see #getGeneratedKeys * @since 1.4 */ public boolean execute(final String sql, final String[] columnNames) throws SQLException { return executeInternal(sql, fetchSize); } /** * executes a select query. * * @param sql the query to send to the server * @return a result set * @throws SQLException if something went wrong */ public ResultSet executeQuery(String sql) throws SQLException { if (executeInternal(sql, fetchSize)) { return results.getResultSet(); } //throw new SQLException("executeQuery() with query '" + query +"' did not return a result set"); return SelectResultSet.createEmptyResultSet(); } /** * Executes an update. * * @param sql the update query. * @return update count * @throws SQLException if the query could not be sent to server. */ public int executeUpdate(String sql) throws SQLException { if (executeInternal(sql, fetchSize)) { return 0; } return getUpdateCount(); } /** * Executes the given SQL statement and signals the driver with the given flag about whether the auto-generated keys * produced by this <code>Statement</code> object should be made available for retrieval. The driver will ignore * the flag if the SQL statement is not an * <code>INSERT</code> statement, or an SQL statement able to return auto-generated keys (the list of such * statements is vendor-specific). * * @param sql an SQL Data Manipulation Language (DML) statement, such as <code>INSERT</code>, * <code>UPDATE</code> or <code>DELETE</code>; or an * SQL statement that returns nothing, such as a DDL statement. * @param autoGeneratedKeys a flag indicating whether auto-generated keys should be made available for retrieval; * one of the following constants: * <code>Statement.RETURN_GENERATED_KEYS</code> <code>Statement.NO_GENERATED_KEYS</code> * @return either (1) the row count for SQL Data Manipulation Language (DML) statements or (2) 0 for SQL statements * that return nothing * @throws SQLException if a database access error occurs, this method is called on a closed * <code>Statement</code>, the given SQL * statement returns a <code>ResultSet</code> object, or the given constant * is not one of those allowed * @throws SQLFeatureNotSupportedException if the JDBC driver does not support this method with a constant of * Statement.RETURN_GENERATED_KEYS * @since 1.4 */ public int executeUpdate(final String sql, final int autoGeneratedKeys) throws SQLException { if (executeInternal(sql, fetchSize)) { return 0; } return getUpdateCount(); } /** * Executes the given SQL statement and signals the driver that the auto-generated keys indicated in the given array * should be made available for retrieval. This array contains the indexes of the columns in the target table that * contain the auto-generated keys that should be made available. The driver will ignore the array if the SQL * statement is not an <code>INSERT</code> statement, or an SQL statement able to return auto-generated keys (the * list of such statements is vendor-specific). * * @param sql an SQL Data Manipulation Language (DML) statement, such as <code>INSERT</code>, * <code>UPDATE</code> or <code>DELETE</code>; or an SQL statement that returns nothing, such * as a DDL statement. * @param columnIndexes an array of column indexes indicating the columns that should be returned from the * inserted row * @return either (1) the row count for SQL Data Manipulation Language (DML) statements or (2) 0 for SQL statements * that return nothing * @throws SQLException if a database access error occurs, this method is called on a closed * <code>Statement</code>, the SQL statement returns a <code>ResultSet</code> object, or the second argument * supplied to this method is not an <code>int</code> array whose elements are valid column indexes * @throws SQLFeatureNotSupportedException if the JDBC driver does not support this method * @since 1.4 */ public int executeUpdate(final String sql, final int[] columnIndexes) throws SQLException { return executeUpdate(sql); } /** * Executes the given SQL statement and signals the driver that the auto-generated keys indicated in the given * array should be made available for retrieval. This array contains the names of the columns in the target table * that contain the auto-generated keys that should be made available. The driver will ignore the array if the SQL * statement is not an <code>INSERT</code> statement, or an SQL statement able to return * auto-generated keys (the list of such statements is vendor-specific). * * @param sql an SQL Data Manipulation Language (DML) statement, such as <code>INSERT</code>, * <code>UPDATE</code> or <code>DELETE</code>; or an * SQL statement that returns nothing, such as a DDL statement. * @param columnNames an array of the names of the columns that should be returned from the inserted row * @return either the row count for <code>INSERT</code>, <code>UPDATE</code>, or <code>DELETE</code> statements, * or 0 for SQL statements that return nothing * @throws SQLException if a database access error occurs, this method is called on a closed * <code>Statement</code>, the SQL statement returns a <code>ResultSet</code> object, or the second argument * supplied to this method is not a <code>String</code> array whose elements are valid column names * @throws SQLFeatureNotSupportedException if the JDBC driver does not support this method * @since 1.4 */ public int executeUpdate(final String sql, final String[] columnNames) throws SQLException { return executeUpdate(sql); } /** * Executes the given SQL statement, which may be an INSERT, UPDATE, or DELETE statement or an SQL statement that returns nothing, * such as an SQL DDL statement. * This method should be used when the returned row count may exceed Integer.MAX_VALUE. * * @param sql sql command * @return update counts * @throws SQLException if any error occur during execution */ @Override public long executeLargeUpdate(String sql) throws SQLException { if (executeInternal(sql, fetchSize)) { return 0; } return getLargeUpdateCount(); } /** * Identical to executeLargeUpdate(String sql), with a flag that indicate that autoGeneratedKeys (primary key fields with "auto_increment") * generated id's must be retrieved. * * Those id's will be available using getGeneratedKeys() method. * * @param sql sql command * @param autoGeneratedKeys a flag indicating whether auto-generated keys should be made available for retrieval; * one of the following constants: * Statement.RETURN_GENERATED_KEYS * Statement.NO_GENERATED_KEYS * @return update counts * @throws SQLException if any error occur during execution */ @Override public long executeLargeUpdate(String sql, int autoGeneratedKeys) throws SQLException { //driver always get generated keys. return executeLargeUpdate(sql); } /** * Identical to executeLargeUpdate(String sql, int autoGeneratedKeys) with autoGeneratedKeys = Statement.RETURN_GENERATED_KEYS set. * * @param sql sql command * @param columnIndexes column Indexes * @return update counts * @throws SQLException if any error occur during execution */ @Override public long executeLargeUpdate(String sql, int[] columnIndexes) throws SQLException { //driver always get generated keys. no need for columnIndexes indication return executeLargeUpdate(sql); } /** * Identical to executeLargeUpdate(String sql, int autoGeneratedKeys) with autoGeneratedKeys = Statement.RETURN_GENERATED_KEYS set. * * @param sql sql command * @param columnNames columns names * @return update counts * @throws SQLException if any error occur during execution */ @Override public long executeLargeUpdate(String sql, String[] columnNames) throws SQLException { //driver always get generated keys. no need for columnNames indication return executeLargeUpdate(sql); } /** * Releases this <code>Statement</code> object's database and JDBC resources immediately instead of waiting for this to happen when it is * automatically closed. It is generally good practice to release resources as soon as you are finished with them to avoid tying up database * resources. Calling the method <code>close</code> on a <code>Statement</code> object that is already closed has no effect. <B>Note:</B>When a * <code>Statement</code> object is closed, its current <code>ResultSet</code> object, if one exists, is also closed. * * @throws SQLException if a database access error occurs */ public void close() throws SQLException { lock.lock(); try { closed = true; if (results.getFetchSize() != 0) { if (options.killFetchStmtOnClose) { try { protocol.cancelCurrentQuery(); skipMoreResults(); } catch (SQLException | IOException sqle) { //eat exception } } else skipMoreResults(); } results.close(); protocol = null; if (connection == null || connection.pooledConnection == null || connection.pooledConnection.statementEventListeners.isEmpty()) { return; } connection.pooledConnection.fireStatementClosed(this); } finally { lock.unlock(); } } /** * Retrieves the maximum number of bytes that can be returned for character and binary column values in a <code>ResultSet</code> object produced * by this <code>Statement</code> object. * This limit applies only to * <code>BINARY</code>, * <code>VARBINARY</code>, * <code>LONGVARBINARY</code>, * <code>CHAR</code>, * <code>VARCHAR</code>, * <code>NCHAR</code>, * <code>NVARCHAR</code>, * <code>LONGNVARCHAR</code> and * <code>LONGVARCHAR</code> * columns. If the limit is exceeded, the excess data is silently discarded. * * @return the current column size limit for columns storing character and binary values; zero means there is no limit * @see #setMaxFieldSize */ public int getMaxFieldSize() { return maxFieldSize; } /** * Sets the limit for the maximum number of bytes that can be returned for character and binary column values in a <code>ResultSet</code> object * produced by this <code>Statement</code> object. This limit applies only to <code>BINARY</code>, <code>VARBINARY</code>, * <code>LONGVARBINARY</code>, <code>CHAR</code>, <code>VARCHAR</code>, <code>NCHAR</code>, <code>NVARCHAR</code>, <code>LONGNVARCHAR</code> and * <code>LONGVARCHAR</code> fields. If the limit is exceeded, the excess data is silently discarded. For maximum portability, use values greater * than 256. * * @param max the new column size limit in bytes; zero means there is no limit * @throws SQLException if a database access error occurs, this method is called on a closed <code>Statement</code> or the condition max * >= 0 is not satisfied * @see #getMaxFieldSize */ public void setMaxFieldSize(final int max) throws SQLException { maxFieldSize = max; } /** * Retrieves the maximum number of rows that a <code>ResultSet</code> object produced by this <code>Statement</code> object can contain. If this * limit is exceeded, the excess rows are silently dropped. * * @return the current maximum number of rows for a <code>ResultSet</code> object produced by this <code>Statement</code> object; zero means there * is no limit * @throws SQLException if a database access error occurs or this method is called on a closed <code>Statement</code> * @see #setMaxRows */ public int getMaxRows() throws SQLException { return (int) maxRows; } /** * Retrieves the maximum number of rows that a ResultSet object produced by this Statement object can contain. * If this limit is exceeded, the excess rows are silently dropped. * @return the current maximum number of rows for a ResultSet object produced by this Statement object; zero means there is no limit */ @Override public long getLargeMaxRows() { return maxRows; } /** * Sets the limit for the maximum number of rows that any ResultSet object generated by this Statement object can contain to the given number. * If the limit is exceeded, the excess rows are silently dropped. * * @param max the new max rows limit; zero means there is no limit * @throws SQLException if the condition max >= 0 is not satisfied */ @Override public void setLargeMaxRows(long max) throws SQLException { if (max < 0) { throw new SQLException("max rows cannot be negative : setLargeMaxRows value is " + max); } maxRows = max; } /** * Sets the limit for the maximum number of rows that any <code>ResultSet</code> object generated by this <code>Statement</code> object can * contain to the given number. If the limit is exceeded, the excess rows are silently dropped. * * @param max the new max rows limit; zero means there is no limit * @throws SQLException if the condition max >= 0 is not satisfied * @see #getMaxRows */ public void setMaxRows(final int max) throws SQLException { if (max < 0) { throw new SQLException("max rows cannot be negative : asked for " + max); } maxRows = max; } /** * Sets escape processing on or off. If escape scanning is on (the default), the driver will do escape substitution before sending the SQL * statement to the database. Note: Since prepared statements have usually been parsed prior to making this call, disabling escape processing for * <code>PreparedStatements</code> objects will have no effect. * * @param enable <code>true</code> to enable escape processing; <code>false</code> to disable it * @throws SQLException if a database access error occurs or this method is called on a closed <code>Statement</code> */ public void setEscapeProcessing(final boolean enable) throws SQLException { } /** * Retrieves the number of seconds the driver will wait for a <code>Statement</code> object to execute. If the limit is exceeded, a * <code>SQLException</code> is thrown. * * @return the current query timeout limit in seconds; zero means there is no limit * @see #setQueryTimeout */ public int getQueryTimeout() { return queryTimeout; } /** * Sets the number of seconds the driver will wait for a <code>Statement</code> object to execute to the given number of seconds. If the limit is * exceeded, an <code>SQLException</code> is thrown. A JDBC driver must apply this limit to the <code>execute</code>, * <code>executeQuery</code> and <code>executeUpdate</code> methods. * * @param seconds the new query timeout limit in seconds; zero means there is no limit * @throws SQLException if a database access error occurs, this method is called on a closed <code>Statement</code> or the condition * seconds >= 0 is not satisfied * @see #getQueryTimeout */ public void setQueryTimeout(final int seconds) throws SQLException { if (seconds < 0) { throw new SQLException("Query timeout rows cannot be negative : asked for " + seconds); } this.queryTimeout = seconds; } /** * Sets the inputStream that will be used for the next execute that uses "LOAD DATA LOCAL INFILE". The name specified as local file/URL will be * ignored. * * @param inputStream inputStream instance, that will be used to send data to server * @throws SQLException if statement is closed */ public void setLocalInfileInputStream(InputStream inputStream) throws SQLException { checkClose(); protocol.setLocalInfileInputStream(inputStream); } /** * Cancels this <code>Statement</code> object if both the DBMS and driver support aborting an SQL statement. This method can be used by one thread * to cancel a statement that is being executed by another thread. * * @throws SQLException if a database access error occurs or this method is called on a closed <code>Statement</code> * @throws SQLFeatureNotSupportedException if the JDBC driver does not support this method */ public void cancel() throws SQLException { checkClose(); try { if (!executing) return; protocol.cancelCurrentQuery(); } catch (SQLException e) { logger.error("error cancelling query", e); ExceptionMapper.throwException(e, connection, this); } catch (IOException e) { // connection gone, query is definitely canceled } } /** * Retrieves the first warning reported by calls on this <code>Statement</code> object. Subsequent <code>Statement</code> object warnings will be * chained to this <code>SQLWarning</code> object. <p>The warning chain is automatically cleared each time a statement is (re)executed. This * method may not be called on a closed <code>Statement</code> object; doing so will cause an <code>SQLException</code> to be thrown.</p> * <p><B>Note:</B> If you are processing a <code>ResultSet</code> object, any warnings associated with reads on that <code>ResultSet</code> object * will be chained on it rather than on the <code>Statement</code> object that produced it.</p> * * @return the first <code>SQLWarning</code> object or <code>null</code> if there are no warnings * @throws SQLException if a database access error occurs or this method is called on a closed <code>Statement</code> */ public SQLWarning getWarnings() throws SQLException { checkClose(); if (!warningsCleared) { return this.connection.getWarnings(); } return null; } /** * Clears all the warnings reported on this <code>Statement</code> object. After a call to this method, the method <code>getWarnings</code> will * return <code>null</code> until a new warning is reported for this <code>Statement</code> object. * * @throws SQLException if a database access error occurs or this method is called on a closed <code>Statement</code> */ public void clearWarnings() throws SQLException { warningsCleared = true; } /** * Sets the SQL cursor name to the given <code>String</code>, which will be used by subsequent <code>Statement</code> object <code>execute</code> * methods. This name can then be used in SQL positioned update or delete statements to identify the current row in the <code>ResultSet</code> * object generated by this statement. If the database does not support positioned update/delete, this method is a noop. To insure that a cursor * has the proper isolation level to support updates, the cursor's <code>SELECT</code> statement should have the form <code>SELECT FOR * UPDATE</code>. If <code>FOR UPDATE</code> is not present, positioned updates may fail. <p><B>Note:</B> By definition, the execution of * positioned updates and deletes must be done by a different <code>Statement</code> object than the one that generated the <code>ResultSet</code> * object being used for positioning. Also, cursor names must be unique within a connection.</p> * * @param name the new cursor name, which must be unique within a connection * @throws SQLException if a database access error occurs or this method is called on a closed <code>Statement</code> * @throws SQLFeatureNotSupportedException if the JDBC driver does not support this method */ public void setCursorName(final String name) throws SQLException { throw ExceptionMapper.getFeatureNotSupportedException("Cursors are not supported"); } /** * Gets the connection that created this statement. * * @return the connection * @throws SQLException if connection is invalid */ public MariaDbConnection getConnection() throws SQLException { return this.connection; } /** * Retrieves any auto-generated keys created as a result of executing this <code>Statement</code> object. If this <code>Statement</code> object * did not generate any keys, an empty <code>ResultSet</code> object is returned. <p><B>Note:</B>If the columns which represent the auto-generated * keys were not specified, the JDBC driver implementation will determine the columns which best represent the auto-generated keys.</p> * * @return a <code>ResultSet</code> object containing the auto-generated key(s) generated by the execution of this <code>Statement</code> object * @throws SQLException if a database access error occurs or this method is called on a closed <code>Statement</code> * @throws SQLFeatureNotSupportedException if the JDBC driver does not support this method * @since 1.4 */ public ResultSet getGeneratedKeys() throws SQLException { if (results != null) { return results.getGeneratedKeys(protocol); } return SelectResultSet.createEmptyResultSet(); } /** * Retrieves the result set holdability for <code>ResultSet</code> objects generated by this <code>Statement</code> object. * * @return either <code>ResultSet.HOLD_CURSORS_OVER_COMMIT</code> or <code>ResultSet.CLOSE_CURSORS_AT_COMMIT</code> * @throws SQLException if a database access error occurs or this method is called on a closed <code>Statement</code> * @since 1.4 */ public int getResultSetHoldability() throws SQLException { return ResultSet.HOLD_CURSORS_OVER_COMMIT; } /** * Retrieves whether this <code>Statement</code> object has been closed. A <code>Statement</code> is closed if the method close has been called on * it, or if it is automatically closed. * * @return true if this <code>Statement</code> object is closed; false if it is still open * @throws SQLException if a database access error occurs * @since 1.6 */ public boolean isClosed() throws SQLException { return closed; } /** * Returns a value indicating whether the <code>Statement</code> is poolable or not. * * @return <code>true</code> if the <code>Statement</code> is poolable; <code>false</code> otherwise * @throws SQLException if this method is called on a closed <code>Statement</code> * @see Statement#setPoolable(boolean) setPoolable(boolean) * @since 1.6 */ @Override public boolean isPoolable() throws SQLException { return false; } /** * <p>Requests that a <code>Statement</code> be pooled or not pooled. The value specified is a hint to the statement pool implementation * indicating whether the applicaiton wants the statement to be pooled. It is up to the statement pool manager as to whether the hint is * used.</p> <p> The poolable value of a statement is applicable to both internal statement caches implemented by the driver and external * statement caches implemented by application servers and other applications. </p> <p>By default, a <code>Statement</code> is not poolable when * created, and a <code>PreparedStatement</code> and <code>CallableStatement</code> are poolable when created.</p> * * @param poolable requests that the statement be pooled if true and that the statement not be pooled if false * @throws SQLException if this method is called on a closed <code>Statement</code> * @since 1.6 */ @Override public void setPoolable(final boolean poolable) throws SQLException { } /** * Retrieves the current result as a ResultSet object. This method should be called only once per result. * * @return the current result as a ResultSet object or null if the result is an update count or there are no more results * @throws SQLException if a database access error occurs or this method is called on a closed Statement */ public ResultSet getResultSet() throws SQLException { checkClose(); return results.getResultSet(); } /** * Retrieves the current result as an update count; if the result is a ResultSet object or there are no more * results, -1 is returned. This method should be called only once per result. * * @return the current result as an update count; -1 if the current result is a ResultSet object or there are no * more results * @throws SQLException if a database access error occurs or this method is called on a closed Statement */ public int getUpdateCount() throws SQLException { if (results != null && results.getCmdInformation() != null && !results.isBatch()) { return results.getCmdInformation().getUpdateCount(); } return -1; } /** * Retrieves the current result as an update count; if the result is a ResultSet object or there are no more results, -1 is returned. * * @return last update count */ @Override public long getLargeUpdateCount() { if (results.getCmdInformation() != null) { return results.getCmdInformation().getLargeUpdateCount(); } return -1; } protected void skipMoreResults() throws SQLException { try { protocol.skip(); warningsCleared = false; connection.reenableWarnings(); } catch (SQLException e) { logger.debug("error skipMoreResults", e); ExceptionMapper.throwException(e, connection, this); } } /** * <p>Moves to this <code>Statement</code> object's next result, returns <code>true</code> if it is a * <code>ResultSet</code> object, and implicitly closes any current <code>ResultSet</code> object(s) obtained with * the method <code>getResultSet</code>.</p> * There are no more results when the following is true: <pre> // stmt is a Statement object * ((stmt.getInternalMoreResults() == false) && (stmt.getUpdateCount() == -1)) </pre> * * @return <code>true</code> if the next result is a <code>ResultSet</code> object; <code>false</code> if it is an * update count or there are no more results * @throws SQLException if a database access error occurs or this method is called on a closed * <code>Statement</code> * @see #execute */ public boolean getMoreResults() throws SQLException { return getMoreResults(Statement.CLOSE_CURRENT_RESULT); } /** * <p>Moves to this <code>Statement</code> object's next result, deals with any current <code>ResultSet</code> * object(s) according to the instructions specified by the given flag, and returns <code>true</code> if the next * result is a <code>ResultSet</code> object.</p> * There are no more results when the following is true: <pre> // stmt is a Statement object * ((stmt.getInternalMoreResults(current) == false) && (stmt.getUpdateCount() == -1))</pre> * * @param current one of the following <code>Statement</code> constants indicating what should happen to current * <code>ResultSet</code> objects obtained using the method <code>getResultSet</code>: * <code>Statement.CLOSE_CURRENT_RESULT</code>, <code>Statement.KEEP_CURRENT_RESULT</code>, * or <code>Statement.CLOSE_ALL_RESULTS</code> * @return <code>true</code> if the next result is a <code>ResultSet</code> object; <code>false</code> if it is an * update count or there are no more results * @throws SQLException if a database access error occurs, this method is called on a closed <code>Statement</code> * or the argument supplied is not one of the following: <code>Statement.CLOSE_CURRENT_RESULT</code>, * <code>Statement.KEEP_CURRENT_RESULT</code> or <code>Statement.CLOSE_ALL_RESULTS</code> * @throws SQLFeatureNotSupportedException if <code>DatabaseMetaData.supportsMultipleOpenResults</code> returns * <code>false</code> and either <code>Statement.KEEP_CURRENT_RESULT</code> * or <code>Statement.CLOSE_ALL_RESULTS</code> are supplied as the argument. * @see #execute * @since 1.4 */ public boolean getMoreResults(final int current) throws SQLException { //if fetch size is set to read fully, other resultSet are put in cache checkClose(); return results.getMoreResults(current, protocol); } /** * Retrieves the direction for fetching rows from database tables that is the default for result sets generated from this <code>Statement</code> * object. If this <code>Statement</code> object has not set a fetch direction by calling the method <code>setFetchDirection</code>, the return * value is implementation-specific. * * @return the default fetch direction for result sets generated from this <code>Statement</code> object * @throws SQLException if a database access error occurs or this method is called on a closed <code>Statement</code> * @see #setFetchDirection * @since 1.2 */ public int getFetchDirection() throws SQLException { return ResultSet.FETCH_FORWARD; } /** * <p>Gives the driver a hint as to the direction in which rows will be processed in <code>ResultSet</code> objects created using this * <code>Statement</code> object. The default value is <code>ResultSet.FETCH_FORWARD</code>.</p> * <p> Note that this method sets the default fetch * direction for result sets generated by this <code>Statement</code> object. Each result set has its own methods for getting and setting its own * fetch direction. </p> * * @param direction the initial direction for processing rows * @throws SQLException if a database access error occurs, this method is called on a closed <code>Statement</code> * or the given direction is not one of <code>ResultSet.FETCH_FORWARD</code>, * <code>ResultSet.FETCH_REVERSE</code>, or <code>ResultSet.FETCH_UNKNOWN</code> * @see #getFetchDirection * @since 1.2 */ public void setFetchDirection(final int direction) throws SQLException { return; } /** * Retrieves the number of result set rows that is the default fetch size for <code>ResultSet</code> objects generated from this * <code>Statement</code> object. If this <code>Statement</code> object has not set a fetch size by calling the method <code>setFetchSize</code>, * the return value is implementation-specific. * * @return the default fetch size for result sets generated from this <code>Statement</code> object * @throws SQLException if a database access error occurs or this method is called on a closed <code>Statement</code> * @see #setFetchSize * @since 1.2 */ public int getFetchSize() throws SQLException { return this.fetchSize; } /** * Gives the JDBC driver a hint as to the number of rows that should be fetched from the database when more rows are needed for * <code>ResultSet</code> objects genrated by this <code>Statement</code>. If the value specified is zero, then the hint is ignored. The default * value is zero. * * @param rows the number of rows to fetch * @throws SQLException if a database access error occurs, this method is called on a closed <code>Statement</code> or the condition * <code>rows >= 0</code> is not satisfied. * @see #getFetchSize * @since 1.2 */ public void setFetchSize(final int rows) throws SQLException { if (rows < 0 && rows != Integer.MIN_VALUE) { throw new SQLException("invalid fetch size"); } else if (rows == Integer.MIN_VALUE) { //for compatibility Integer.MIN_VALUE is transform to 0 => streaming this.fetchSize = 1; return; } this.fetchSize = rows; } /** * Retrieves the result set concurrency for <code>ResultSet</code> objects generated by this <code>Statement</code> object. * * @return either <code>ResultSet.CONCUR_READ_ONLY</code> or <code>ResultSet.CONCUR_UPDATABLE</code> * @throws SQLException if a database access error occurs or this method is called on a closed <code>Statement</code> * @since 1.2 */ public int getResultSetConcurrency() throws SQLException { return ResultSet.CONCUR_READ_ONLY; } /** * Retrieves the result set type for <code>ResultSet</code> objects generated by this <code>Statement</code> object. * * @return one of <code>ResultSet.TYPE_FORWARD_ONLY</code>, <code>ResultSet.TYPE_SCROLL_INSENSITIVE</code>, or * <code>ResultSet.TYPE_SCROLL_SENSITIVE</code> * @throws SQLException if a database access error occurs or this method is called on a closed <code>Statement</code> * @since 1.2 */ public int getResultSetType() throws SQLException { return resultSetScrollType; } /** * Adds the given SQL command to the current list of commmands for this <code>Statement</code> object. The send in this list can be executed * as a batch by calling the method <code>executeBatch</code>. * * @param sql typically this is a SQL <code>INSERT</code> or <code>UPDATE</code> statement * @throws SQLException if a database access error occurs, this method is called on a closed <code>Statement</code> or the driver does * not support batch updates * @see #executeBatch * @see DatabaseMetaData#supportsBatchUpdates * @since 1.2 */ public void addBatch(final String sql) throws SQLException { if (batchQueries == null) batchQueries = new ArrayList<>(); if (sql == null) throw ExceptionMapper.getSqlException("null cannot be set to addBatch( String sql)"); batchQueries.add(sql); } /** * Empties this <code>Statement</code> object's current list of SQL send. * * @see #addBatch * @see DatabaseMetaData#supportsBatchUpdates * @since 1.2 */ public void clearBatch() { if (batchQueries != null) batchQueries.clear(); } /** * Execute statements. depending on option, queries mays be rewritten : * <p> * those queries will be rewritten if possible to * INSERT INTO ... VALUES (...) ; INSERT INTO ... VALUES (...); * <p> * if option rewriteBatchedStatements is set to true, rewritten to * INSERT INTO ... VALUES (...), (...); * * @return an array of update counts containing one element for each command in the batch. The elements of the * array are ordered according to the order in which send were added to the batch. * @throws SQLException if a database access error occurs, this method is called on a closed <code>Statement</code> * or the driver does not support batch statements. Throws {@link BatchUpdateException} * (a subclass of <code>SQLException</code>) if one of the send * sent to the database fails to execute properly or attempts to return a result set. * @see #addBatch * @see DatabaseMetaData#supportsBatchUpdates * @since 1.3 */ public int[] executeBatch() throws SQLException { checkClose(); int size; if (batchQueries == null || (size = batchQueries.size()) == 0) return new int[0]; lock.lock(); try { internalBatchExecution(size); return results.getCmdInformation().getUpdateCounts(); } catch (SQLException initialSqlEx) { throw executeBatchExceptionEpilogue(initialSqlEx, results.getCmdInformation(), size, false); } finally { executeBatchEpilogue(); lock.unlock(); } } /** * Execute batch, like executeBatch(), with returning results with long[]. * For when row count may exceed Integer.MAX_VALUE. * * @return an array of update counts (one element for each command in the batch) * @throws SQLException if a database error occur. */ @Override public long[] executeLargeBatch() throws SQLException { checkClose(); int size; if (batchQueries == null || (size = batchQueries.size()) == 0) return new long[0]; lock.lock(); try { internalBatchExecution(size); return results.getCmdInformation().getLargeUpdateCounts(); } catch (SQLException initialSqlEx) { throw executeLargeBatchExceptionEpilogue(initialSqlEx, results.getCmdInformation(), size); } finally { executeBatchEpilogue(); lock.unlock(); } } /** * Internal batch execution. * @param size expected result-set size * @throws SQLException throw exception if batch error occur */ protected void internalBatchExecution(int size) throws SQLException { executeQueryPrologue(); results.reset(0, true, size, false, resultSetScrollType); if (this.options.rewriteBatchedStatements) { //check that queries are rewritable boolean batchQueryMultiRewritable = true; for (String query : batchQueries) { if (!ClientPrepareResult.isRewritableBatch(query, connection.noBackslashEscapes)) { batchQueryMultiRewritable = false; break; } } if (batchQueryMultiRewritable) { protocol.executeBatchMultiple(protocol.isMasterConnection(), results, batchQueries); } else { protocol.executeBatch(protocol.isMasterConnection(), results, batchQueries); } } else { protocol.executeBatch(protocol.isMasterConnection(), results, batchQueries); } results.commandEnd(); } /** * <p>Returns an object that implements the given interface to allow access to non-standard methods, or standard methods not exposed by the * proxy.</p> * <p>If the receiver implements the interface then the result is the receiver or a proxy for the receiver. If the receiver is a wrapper and * the wrapped object implements the interface then the result is the wrapped object or a proxy for the wrapped object. Otherwise return the the * result of calling <code>unwrap</code> recursively on the wrapped object or a proxy for that result. If the receiver is not a wrapper and does * not implement the interface, then an <code>SQLException</code> is thrown. </p> * * @param iface A Class defining an interface that the result must implement. * @return an object that implements the interface. May be a proxy for the actual implementing object. * @throws SQLException If no object found that implements the interface * @since 1.6 */ @SuppressWarnings("unchecked") public <T> T unwrap(final Class<T> iface) throws SQLException { try { if (isWrapperFor(iface)) { return (T) this; } else { throw new SQLException("The receiver is not a wrapper and does not implement the interface"); } } catch (Exception e) { throw new SQLException("The receiver is not a wrapper and does not implement the interface"); } } /** * Returns true if this either implements the interface argument or is directly or indirectly a wrapper for an object that does. Returns false * otherwise. If this implements the interface then return true, else if this is a wrapper then return the result of recursively calling * <code>isWrapperFor</code> on the wrapped object. If this does not implement the interface and is not a wrapper, return false. This method * should be implemented as a low-cost operation compared to <code>unwrap</code> so that callers can use this method to avoid expensive * <code>unwrap</code> calls that may fail. If this method returns true then calling <code>unwrap</code> with the same argument should succeed. * * @param interfaceOrWrapper a Class defining an interface. * @return true if this implements the interface or directly or indirectly wraps an object that does. * @throws SQLException if an error occurs while determining whether this is a wrapper for an object with the given interface. * @since 1.6 */ public boolean isWrapperFor(final Class<?> interfaceOrWrapper) throws SQLException { return interfaceOrWrapper.isInstance(this); } public void closeOnCompletion() throws SQLException { mustCloseOnCompletion = true; } public boolean isCloseOnCompletion() throws SQLException { return mustCloseOnCompletion; } /** * Check that close on completion is asked, and close if so. * * @param resultSet resultSet * @throws SQLException if close has error */ public void checkCloseOnCompletion(ResultSet resultSet) throws SQLException { if (mustCloseOnCompletion && !closed && resultSet.equals(results.getResultSet())) { close(); } } /** * Check if statement is closed, and throw exception if so. * * @throws SQLException if statement close */ protected void checkClose() throws SQLException { if (closed) { throw new SQLException("Cannot do an operation on a closed statement"); } } /** * Permit to retrieve current connection thread id, or -1 if unknown. * * @return current connection thread id. */ public long getServerThreadId() { return (protocol != null) ? protocol.getServerThreadId() : -1; } }