コード例 #1
0
ファイル: BatchTestCase.java プロジェクト: sohu001/mondrian
  /**
   * During MDX query parse and execution, checks that the query results (or does not result) in a
   * particular SQL statement being generated.
   *
   * <p>Parses and executes the MDX query once for each SQL pattern in the current dialect. If there
   * are multiple patterns, runs the MDX query multiple times, and expects to see each SQL statement
   * appear. If there are no patterns in this dialect, the test trivially succeeds.
   *
   * @param testContext non-default test context if required
   * @param mdxQuery MDX query
   * @param patterns Set of patterns
   * @param negative false to assert if SQL is generated; true to assert if SQL is NOT generated
   * @param bypassSchemaCache whether to grab a new connection and bypass the schema cache before
   *     parsing the MDX query
   * @param clearCache whether to clear cache before executing the MDX query
   */
  protected void assertQuerySqlOrNot(
      TestContext testContext,
      String mdxQuery,
      SqlPattern[] patterns,
      boolean negative,
      boolean bypassSchemaCache,
      boolean clearCache) {
    Connection connection = testContext.getConnection();

    mdxQuery = testContext.upgradeQuery(mdxQuery);

    // Run the test once for each pattern in this dialect.
    // (We could optimize and run it once, collecting multiple queries, and
    // comparing all queries at the end.)
    Dialect dialect = testContext.getDialect();
    Dialect.DatabaseProduct d = dialect.getDatabaseProduct();
    boolean patternFound = false;
    for (SqlPattern sqlPattern : patterns) {
      if (!sqlPattern.hasDatabaseProduct(d)) {
        // If the dialect is not one in the pattern set, skip the
        // test. If in the end no pattern is located, print a warning
        // message if required.
        continue;
      }

      patternFound = true;

      String sql = sqlPattern.getSql();
      String trigger = sqlPattern.getTriggerSql();

      sql = dialectize(d, sql);
      trigger = dialectize(d, trigger);

      // Create a dummy DataSource which will throw a 'bomb' if it is
      // asked to execute a particular SQL statement, but will otherwise
      // behave exactly the same as the current DataSource.
      RolapUtil.setHook(new TriggerHook(trigger));
      Bomb bomb = null;
      try {
        if (bypassSchemaCache) {
          connection = testContext.withSchemaPool(false).getConnection();
        }
        final Query query = connection.parseQuery(mdxQuery);
        if (clearCache) {
          clearCache((RolapCube) query.getCube());
        }
        final Result result = connection.execute(query);
        Util.discard(result);
        bomb = null;
      } catch (Bomb e) {
        bomb = e;
      } catch (RuntimeException e) {
        // Walk up the exception tree and see if the root cause
        // was a SQL bomb.
        bomb = Util.getMatchingCause(e, Bomb.class);
        if (bomb == null) {
          throw e;
        }
      } finally {
        RolapUtil.setHook(null);
      }
      if (negative) {
        if (bomb != null) {
          fail("forbidden query [" + sql + "] detected");
        }
      } else {
        if (bomb == null) {
          fail("expected query [" + sql + "] did not occur");
        }
        assertEquals(
            replaceQuotes(sql.replaceAll("\r\n", "\n")),
            replaceQuotes(bomb.sql.replaceAll("\r\n", "\n")));
      }
    }

    // Print warning message that no pattern was specified for the current
    // dialect.
    if (!patternFound) {
      String warnDialect = MondrianProperties.instance().WarnIfNoPatternForDialect.get();

      if (warnDialect.equals(d.toString())) {
        System.out.println(
            "[No expected SQL statements found for dialect \""
                + dialect.toString()
                + "\" and test not run]");
      }
    }
  }
コード例 #2
0
ファイル: SqlStatement.java プロジェクト: sohu001/mondrian
/**
 * SqlStatement contains a SQL statement and associated resources throughout its lifetime.
 *
 * <p>The goal of SqlStatement is to make tracing, error-handling and resource-management easier.
 * None of the methods throws a SQLException; if an error occurs in one of the methods, the method
 * wraps the exception in a {@link RuntimeException} describing the high-level operation, logs that
 * the operation failed, and throws that RuntimeException.
 *
 * <p>If methods succeed, the method generates lifecycle logging such as the elapsed time and number
 * of rows fetched.
 *
 * <p>There are a few obligations on the caller. The caller must:
 *
 * <ul>
 *   <li>call the {@link #handle(Throwable)} method if one of the contained objects (say the {@link
 *       java.sql.ResultSet}) gives an error;
 *   <li>call the {@link #close()} method if all operations complete successfully.
 *   <li>increment the {@link #rowCount} field each time a row is fetched.
 * </ul>
 *
 * <p>The {@link #close()} method is idempotent. You are welcome to call it more than once.
 *
 * <p>SqlStatement is not thread-safe.
 *
 * @author jhyde
 * @since 2.3
 */
public class SqlStatement {
  private static final String TIMING_NAME = "SqlStatement-";

  // used for SQL logging, allows for a SQL Statement UID
  private static final AtomicLong ID_GENERATOR = new AtomicLong();

  private static final RolapUtil.Semaphore querySemaphore = RolapUtil.getQuerySemaphore();

  private final DataSource dataSource;
  private Connection jdbcConnection;
  private ResultSet resultSet;
  private final String sql;
  private final List<Type> types;
  private final int maxRows;
  private final int firstRowOrdinal;
  private final Locus locus;
  private final int resultSetType;
  private final int resultSetConcurrency;
  private boolean haveSemaphore;
  public int rowCount;
  private long startTimeNanos;
  private long startTimeMillis;
  private final List<Accessor> accessors = new ArrayList<Accessor>();
  private State state = State.FRESH;
  private final long id;

  /**
   * Creates a SqlStatement.
   *
   * @param dataSource Data source
   * @param sql SQL
   * @param types Suggested types of columns, or null; if present, must have one element for each
   *     SQL column; each not-null entry overrides deduced JDBC type of the column
   * @param maxRows Maximum rows; <= 0 means no maximum
   * @param firstRowOrdinal Ordinal of first row to skip to; <= 0 do not skip
   * @param locus Execution context of this statement
   * @param resultSetType Result set type
   * @param resultSetConcurrency Result set concurrency
   */
  public SqlStatement(
      DataSource dataSource,
      String sql,
      List<Type> types,
      int maxRows,
      int firstRowOrdinal,
      Locus locus,
      int resultSetType,
      int resultSetConcurrency) {
    this.id = ID_GENERATOR.getAndIncrement();
    this.dataSource = dataSource;
    this.sql = sql;
    this.types = types;
    this.maxRows = maxRows;
    this.firstRowOrdinal = firstRowOrdinal;
    this.locus = locus;
    this.resultSetType = resultSetType;
    this.resultSetConcurrency = resultSetConcurrency;
  }

  /** Executes the current statement, and handles any SQLException. */
  public void execute() {
    assert state == State.FRESH : "cannot re-execute";
    state = State.ACTIVE;
    String status = "failed";
    Statement statement = null;
    try {
      this.jdbcConnection = dataSource.getConnection();
      querySemaphore.enter();
      haveSemaphore = true;
      // Trace start of execution.
      if (RolapUtil.SQL_LOGGER.isDebugEnabled()) {
        StringBuilder sqllog = new StringBuilder();
        sqllog.append(id).append(": ").append(locus.component).append(": executing sql [");
        if (sql.indexOf('\n') >= 0) {
          // SQL appears to be formatted as multiple lines. Make it
          // start on its own line.
          sqllog.append("\n");
        }
        sqllog.append(sql);
        sqllog.append(']');
        RolapUtil.SQL_LOGGER.debug(sqllog.toString());
      }

      // Execute hook.
      RolapUtil.ExecuteQueryHook hook = RolapUtil.getHook();
      if (hook != null) {
        hook.onExecuteQuery(sql);
      }
      startTimeNanos = System.nanoTime();
      startTimeMillis = System.currentTimeMillis();
      if (resultSetType < 0 || resultSetConcurrency < 0) {
        statement = jdbcConnection.createStatement();
      } else {
        statement = jdbcConnection.createStatement(resultSetType, resultSetConcurrency);
      }
      if (maxRows > 0) {
        statement.setMaxRows(maxRows);
      }

      // First make sure to register with the execution instance.
      locus.execution.registerStatement(locus, statement);

      locus
          .getServer()
          .getMonitor()
          .sendEvent(
              new SqlStatementStartEvent(
                  startTimeMillis, id, locus, sql, getPurpose(), getCellRequestCount()));

      this.resultSet = statement.executeQuery(sql);

      // skip to first row specified in request
      this.state = State.ACTIVE;
      if (firstRowOrdinal > 0) {
        if (resultSetType == ResultSet.TYPE_FORWARD_ONLY) {
          for (int i = 0; i < firstRowOrdinal; ++i) {
            if (!this.resultSet.next()) {
              this.state = State.DONE;
              break;
            }
          }
        } else {
          if (!this.resultSet.absolute(firstRowOrdinal)) {
            this.state = State.DONE;
          }
        }
      }

      long timeMillis = System.currentTimeMillis();
      long timeNanos = System.nanoTime();
      final long executeNanos = timeNanos - startTimeNanos;
      final long executeMillis = executeNanos / 1000000;
      Util.addDatabaseTime(executeMillis);
      status = ", exec " + executeMillis + " ms";

      locus
          .getServer()
          .getMonitor()
          .sendEvent(
              new SqlStatementExecuteEvent(timeMillis, id, locus, sql, getPurpose(), executeNanos));

      // Compute accessors. They ensure that we use the most efficient
      // method (e.g. getInt, getDouble, getObject) for the type of the
      // column. Even if you are going to box the result into an object,
      // it is better to use getInt than getObject; the latter might
      // return something daft like a BigDecimal (does, on the Oracle JDBC
      // driver).
      accessors.clear();
      for (Type type : guessTypes()) {
        accessors.add(createAccessor(accessors.size(), type));
      }
    } catch (Throwable e) {
      status = ", failed (" + e + ")";
      Util.close(resultSet, statement, jdbcConnection);

      if (haveSemaphore) {
        haveSemaphore = false;
        querySemaphore.leave();
      }
      if (e instanceof Error) {
        throw (Error) e;
      } else {
        throw handle(e);
      }
    } finally {
      RolapUtil.SQL_LOGGER.debug(id + ": " + status);

      if (RolapUtil.LOGGER.isDebugEnabled()) {
        RolapUtil.LOGGER.debug(locus.component + ": executing sql [" + sql + "]" + status);
      }
    }
  }

  /**
   * Closes all resources (statement, result set) held by this SqlStatement.
   *
   * <p>If any of them fails, wraps them in a {@link RuntimeException} describing the high-level
   * operation which this statement was performing. No further error-handling is required to produce
   * a descriptive stack trace, unless you want to absorb the error.
   */
  public void close() {
    if (haveSemaphore) {
      haveSemaphore = false;
      querySemaphore.leave();
    }

    // According to the JDBC spec, closing a statement automatically closes
    // its result sets, and closing a connection automatically closes its
    // statements. But let's be conservative and close everything
    // explicitly.
    Statement statement = null;
    SQLException ex = null;
    if (resultSet != null) {
      try {
        statement = resultSet.getStatement();
        resultSet.close();
      } catch (SQLException e) {
        ex = e;
      } finally {
        resultSet = null;
      }
    }
    if (statement != null) {
      try {
        statement.close();
      } catch (SQLException e) {
        if (ex != null) {
          ex = e;
        }
      }
    }
    if (jdbcConnection != null) {
      try {
        jdbcConnection.close();
      } catch (SQLException e) {
        if (ex != null) {
          ex = e;
        }
      } finally {
        jdbcConnection = null;
      }
    }

    if (ex != null) {
      throw Util.newError(ex, locus.message + "; sql=[" + sql + "]");
    }

    long endTime = System.currentTimeMillis();
    long totalMs = endTime - startTimeMillis;
    String status = ", exec+fetch " + totalMs + " ms, " + rowCount + " rows";

    locus.execution.getQueryTiming().markFull(TIMING_NAME + locus.component, totalMs);

    RolapUtil.SQL_LOGGER.debug(id + ": " + status);

    if (RolapUtil.LOGGER.isDebugEnabled()) {
      RolapUtil.LOGGER.debug(locus.component + ": done executing sql [" + sql + "]" + status);
    }

    locus
        .getServer()
        .getMonitor()
        .sendEvent(
            new SqlStatementEndEvent(endTime, id, locus, sql, getPurpose(), rowCount, false, null));
  }

  public ResultSet getResultSet() {
    return resultSet;
  }

  /**
   * Handles an exception thrown from the ResultSet, implicitly calls {@link #close}, and returns an
   * exception which includes the full stack, including a description of the high-level operation.
   *
   * @param e Exception
   * @return Runtime exception
   */
  public RuntimeException handle(Throwable e) {
    RuntimeException runtimeException = Util.newError(e, locus.message + "; sql=[" + sql + "]");
    try {
      close();
    } catch (RuntimeException re) {
      // ignore
    }
    return runtimeException;
  }

  private static Type getDecimalType(int precision, int scale) {
    if ((scale == 0 || scale == -127) && (precision <= 9 || precision == 38)) {
      // An int (up to 2^31 = 2.1B) can hold any NUMBER(10, 0) value
      // (up to 10^9 = 1B). NUMBER(38, 0) is conventionally used in
      // Oracle for integers of unspecified precision, so let's be
      // bold and assume that they can fit into an int.
      //
      // Oracle also seems to sometimes represent integers as
      // (type=NUMERIC, precision=0, scale=-127) for reasons unknown.
      return Type.INT;
    } else {
      return Type.DOUBLE;
    }
  }

  /**
   * Chooses the most appropriate type for accessing the values of a column in a result set.
   *
   * <p>NOTE: It is possible that this method is driver-dependent. If this is the case, move it to
   * {@link mondrian.spi.Dialect}.
   *
   * @param suggestedType Type suggested by Level.internalType attribute
   * @param metaData Result set metadata
   * @param i Column ordinal (0-based)
   * @return Best client type
   * @throws SQLException on error
   */
  public static Type guessType(Type suggestedType, ResultSetMetaData metaData, int i)
      throws SQLException {
    if (suggestedType != null) {
      return suggestedType;
    }
    final String typeName = metaData.getColumnTypeName(i + 1);
    final int columnType = metaData.getColumnType(i + 1);
    int precision;
    int scale;
    switch (columnType) {
      case Types.SMALLINT:
      case Types.INTEGER:
      case Types.BOOLEAN:
        return Type.INT;
      case Types.NUMERIC:
        precision = metaData.getPrecision(i + 1);
        scale = metaData.getScale(i + 1);
        if (precision == 0
            && (scale == 0 || scale == -127)
            && (typeName.equalsIgnoreCase("NUMBER") || (typeName.equalsIgnoreCase("NUMERIC")))) {
          // In Oracle and Greenplum the NUMBER/NUMERIC datatype with no
          // precision or scale (not NUMBER(p) or NUMBER(p, s)) means
          // floating point. Some drivers represent this with scale 0,
          // others scale -127.
          //
          // There is a further problem. In GROUPING SETS queries, Oracle
          // loosens the type of columns compared to mere GROUP BY
          // queries. We need integer GROUP BY columns to remain integers,
          // otherwise the segments won't be found; but if we convert
          // measure (whose column names are like "m0", "m1") to integers,
          // data loss will occur.
          final String columnName = metaData.getColumnName(i + 1);
          if (columnName.startsWith("m")) {
            return Type.OBJECT;
          } else {
            return Type.INT;
          }
        }
        return getDecimalType(precision, scale);
      case Types.DECIMAL:
        precision = metaData.getPrecision(i + 1);
        scale = metaData.getScale(i + 1);
        return getDecimalType(precision, scale);
      case Types.DOUBLE:
      case Types.FLOAT:
        return Type.DOUBLE;
      default:
        return Type.OBJECT;
    }
  }

  private Accessor createAccessor(int column, Type type) {
    final int columnPlusOne = column + 1;
    switch (type) {
      case OBJECT:
        return new Accessor() {
          public Object get() throws SQLException {
            return resultSet.getObject(columnPlusOne);
          }
        };
      case STRING:
        return new Accessor() {
          public Object get() throws SQLException {
            return resultSet.getString(columnPlusOne);
          }
        };
      case INT:
        return new Accessor() {
          public Object get() throws SQLException {
            final int val = resultSet.getInt(columnPlusOne);
            if (val == 0 && resultSet.wasNull()) {
              return null;
            }
            return val;
          }
        };
      case LONG:
        return new Accessor() {
          public Object get() throws SQLException {
            final long val = resultSet.getLong(columnPlusOne);
            if (val == 0 && resultSet.wasNull()) {
              return null;
            }
            return val;
          }
        };
      case DOUBLE:
        return new Accessor() {
          public Object get() throws SQLException {
            final double val = resultSet.getDouble(columnPlusOne);
            if (val == 0 && resultSet.wasNull()) {
              return null;
            }
            return val;
          }
        };
      default:
        throw Util.unexpected(type);
    }
  }

  public List<Type> guessTypes() throws SQLException {
    final ResultSetMetaData metaData = resultSet.getMetaData();
    final int columnCount = metaData.getColumnCount();
    assert this.types == null || this.types.size() == columnCount;
    List<Type> types = new ArrayList<Type>();
    for (int i = 0; i < columnCount; i++) {
      final Type suggestedType = this.types == null ? null : this.types.get(i);
      types.add(guessType(suggestedType, metaData, i));
    }
    return types;
  }

  public List<Accessor> getAccessors() throws SQLException {
    return accessors;
  }

  /**
   * Returns the result set in a proxy which automatically closes this SqlStatement (and hence also
   * the statement and result set) when the result set is closed.
   *
   * <p>This helps to prevent connection leaks. The caller still has to remember to call
   * ResultSet.close(), of course.
   *
   * @return Wrapped result set
   */
  public ResultSet getWrappedResultSet() {
    return (ResultSet)
        Proxy.newProxyInstance(
            null, new Class<?>[] {ResultSet.class}, new MyDelegatingInvocationHandler(this));
  }

  private SqlStatementEvent.Purpose getPurpose() {
    if (locus instanceof StatementLocus) {
      return ((StatementLocus) locus).purpose;
    } else {
      return SqlStatementEvent.Purpose.OTHER;
    }
  }

  private int getCellRequestCount() {
    if (locus instanceof StatementLocus) {
      return ((StatementLocus) locus).cellRequestCount;
    } else {
      return 0;
    }
  }

  /**
   * The approximate JDBC type of a column.
   *
   * <p>This type affects which {@link ResultSet} method we use to get values of this column: the
   * default is {@link java.sql.ResultSet#getObject(int)}, but we'd prefer to use native values
   * {@code getInt} and {@code getDouble} if possible.
   */
  public enum Type {
    OBJECT,
    DOUBLE,
    INT,
    LONG,
    STRING;

    public Object get(ResultSet resultSet, int column) throws SQLException {
      switch (this) {
        case OBJECT:
          return resultSet.getObject(column + 1);
        case STRING:
          return resultSet.getString(column + 1);
        case INT:
          return resultSet.getInt(column + 1);
        case LONG:
          return resultSet.getLong(column + 1);
        case DOUBLE:
          return resultSet.getDouble(column + 1);
        default:
          throw Util.unexpected(this);
      }
    }
  }

  public interface Accessor {
    Object get() throws SQLException;
  }

  /**
   * Reflectively implements the {@link ResultSet} interface by routing method calls to the result
   * set inside a {@link mondrian.rolap.SqlStatement}. When the result set is closed, so is the
   * SqlStatement, and hence the JDBC connection and statement also.
   */
  // must be public for reflection to work
  public static class MyDelegatingInvocationHandler extends DelegatingInvocationHandler {
    private final SqlStatement sqlStatement;

    /**
     * Creates a MyDelegatingInvocationHandler.
     *
     * @param sqlStatement SQL statement
     */
    MyDelegatingInvocationHandler(SqlStatement sqlStatement) {
      this.sqlStatement = sqlStatement;
    }

    protected Object getTarget() throws InvocationTargetException {
      final ResultSet resultSet = sqlStatement.getResultSet();
      if (resultSet == null) {
        throw new InvocationTargetException(
            new SQLException("Invalid operation. Statement is closed."));
      }
      return resultSet;
    }

    /**
     * Helper method to implement {@link java.sql.ResultSet#close()}.
     *
     * @throws SQLException on error
     */
    public void close() throws SQLException {
      sqlStatement.close();
    }
  }

  private enum State {
    FRESH,
    ACTIVE,
    DONE
  }

  public static class StatementLocus extends Locus {
    private final SqlStatementEvent.Purpose purpose;
    private final int cellRequestCount;

    public StatementLocus(
        Execution execution,
        String component,
        String message,
        SqlStatementEvent.Purpose purpose,
        int cellRequestCount) {
      super(execution, component, message);
      this.purpose = purpose;
      this.cellRequestCount = cellRequestCount;
    }
  }
}
コード例 #3
0
ファイル: BatchTestCase.java プロジェクト: sohu001/mondrian
  /**
   * Checks that a given sequence of cell requests results in a particular SQL statement being
   * generated.
   *
   * <p>Always clears the cache before running the requests.
   *
   * <p>Runs the requests once for each SQL pattern in the current dialect. If there are multiple
   * patterns, runs the MDX query multiple times, and expects to see each SQL statement appear. If
   * there are no patterns in this dialect, the test trivially succeeds.
   *
   * @param requests Sequence of cell requests
   * @param patterns Set of patterns
   * @param negative Set to false in order to 'expect' a query or true to 'forbid' a query.
   */
  protected void assertRequestSql(CellRequest[] requests, SqlPattern[] patterns, boolean negative) {
    final RolapStar star = requests[0].getMeasure().getStar();
    final String cubeName = requests[0].getMeasure().getCubeName();
    final RolapCube cube = lookupCube(cubeName);
    final Dialect sqlDialect = star.getSqlQueryDialect();
    Dialect.DatabaseProduct d = sqlDialect.getDatabaseProduct();
    SqlPattern sqlPattern = SqlPattern.getPattern(d, patterns);
    if (d == Dialect.DatabaseProduct.UNKNOWN) {
      // If the dialect is not one in the pattern set, do not run the
      // test. We do not print any warning message.
      return;
    }

    boolean patternFound = false;
    for (SqlPattern pattern : patterns) {
      if (!pattern.hasDatabaseProduct(d)) {
        continue;
      }

      patternFound = true;

      clearCache(cube);

      String sql = sqlPattern.getSql();
      String trigger = sqlPattern.getTriggerSql();
      switch (d) {
        case ORACLE:
          sql = sql.replaceAll(" =as= ", " ");
          trigger = trigger.replaceAll(" =as= ", " ");
          break;
        case TERADATA:
          sql = sql.replaceAll(" =as= ", " as ");
          trigger = trigger.replaceAll(" =as= ", " as ");
          break;
      }

      // Create a dummy DataSource which will throw a 'bomb' if it is
      // asked to execute a particular SQL statement, but will otherwise
      // behave exactly the same as the current DataSource.
      RolapUtil.setHook(new TriggerHook(trigger));
      Bomb bomb;
      final Execution execution =
          new Execution(((RolapConnection) getConnection()).getInternalStatement(), 1000);
      final AggregationManager aggMgr =
          execution
              .getMondrianStatement()
              .getMondrianConnection()
              .getServer()
              .getAggregationManager();
      final Locus locus = new Locus(execution, "BatchTestCase", "BatchTestCase");
      try {
        FastBatchingCellReader fbcr =
            new FastBatchingCellReader(execution, getCube(cubeName), aggMgr);
        for (CellRequest request : requests) {
          fbcr.recordCellRequest(request);
        }
        // The FBCR will presume there is a current Locus in the stack,
        // so let's create a mock one.
        Locus.push(locus);
        fbcr.loadAggregations();
        bomb = null;
      } catch (Bomb e) {
        bomb = e;
      } finally {
        RolapUtil.setHook(null);
        Locus.pop(locus);
      }
      if (!negative && bomb == null) {
        fail("expected query [" + sql + "] did not occur");
      } else if (negative && bomb != null) {
        fail("forbidden query [" + sql + "] detected");
      }
      TestContext.assertEqualsVerbose(replaceQuotes(sql), replaceQuotes(bomb.sql));
    }

    // Print warning message that no pattern was specified for the current
    // dialect.
    if (!patternFound) {
      String warnDialect = MondrianProperties.instance().WarnIfNoPatternForDialect.get();

      if (warnDialect.equals(d.toString())) {
        System.out.println(
            "[No expected SQL statements found for dialect \""
                + sqlDialect.toString()
                + "\" and test not run]");
      }
    }
  }
コード例 #4
0
ファイル: SqlStatement.java プロジェクト: sohu001/mondrian
  /** Executes the current statement, and handles any SQLException. */
  public void execute() {
    assert state == State.FRESH : "cannot re-execute";
    state = State.ACTIVE;
    String status = "failed";
    Statement statement = null;
    try {
      this.jdbcConnection = dataSource.getConnection();
      querySemaphore.enter();
      haveSemaphore = true;
      // Trace start of execution.
      if (RolapUtil.SQL_LOGGER.isDebugEnabled()) {
        StringBuilder sqllog = new StringBuilder();
        sqllog.append(id).append(": ").append(locus.component).append(": executing sql [");
        if (sql.indexOf('\n') >= 0) {
          // SQL appears to be formatted as multiple lines. Make it
          // start on its own line.
          sqllog.append("\n");
        }
        sqllog.append(sql);
        sqllog.append(']');
        RolapUtil.SQL_LOGGER.debug(sqllog.toString());
      }

      // Execute hook.
      RolapUtil.ExecuteQueryHook hook = RolapUtil.getHook();
      if (hook != null) {
        hook.onExecuteQuery(sql);
      }
      startTimeNanos = System.nanoTime();
      startTimeMillis = System.currentTimeMillis();
      if (resultSetType < 0 || resultSetConcurrency < 0) {
        statement = jdbcConnection.createStatement();
      } else {
        statement = jdbcConnection.createStatement(resultSetType, resultSetConcurrency);
      }
      if (maxRows > 0) {
        statement.setMaxRows(maxRows);
      }

      // First make sure to register with the execution instance.
      locus.execution.registerStatement(locus, statement);

      locus
          .getServer()
          .getMonitor()
          .sendEvent(
              new SqlStatementStartEvent(
                  startTimeMillis, id, locus, sql, getPurpose(), getCellRequestCount()));

      this.resultSet = statement.executeQuery(sql);

      // skip to first row specified in request
      this.state = State.ACTIVE;
      if (firstRowOrdinal > 0) {
        if (resultSetType == ResultSet.TYPE_FORWARD_ONLY) {
          for (int i = 0; i < firstRowOrdinal; ++i) {
            if (!this.resultSet.next()) {
              this.state = State.DONE;
              break;
            }
          }
        } else {
          if (!this.resultSet.absolute(firstRowOrdinal)) {
            this.state = State.DONE;
          }
        }
      }

      long timeMillis = System.currentTimeMillis();
      long timeNanos = System.nanoTime();
      final long executeNanos = timeNanos - startTimeNanos;
      final long executeMillis = executeNanos / 1000000;
      Util.addDatabaseTime(executeMillis);
      status = ", exec " + executeMillis + " ms";

      locus
          .getServer()
          .getMonitor()
          .sendEvent(
              new SqlStatementExecuteEvent(timeMillis, id, locus, sql, getPurpose(), executeNanos));

      // Compute accessors. They ensure that we use the most efficient
      // method (e.g. getInt, getDouble, getObject) for the type of the
      // column. Even if you are going to box the result into an object,
      // it is better to use getInt than getObject; the latter might
      // return something daft like a BigDecimal (does, on the Oracle JDBC
      // driver).
      accessors.clear();
      for (Type type : guessTypes()) {
        accessors.add(createAccessor(accessors.size(), type));
      }
    } catch (Throwable e) {
      status = ", failed (" + e + ")";
      Util.close(resultSet, statement, jdbcConnection);

      if (haveSemaphore) {
        haveSemaphore = false;
        querySemaphore.leave();
      }
      if (e instanceof Error) {
        throw (Error) e;
      } else {
        throw handle(e);
      }
    } finally {
      RolapUtil.SQL_LOGGER.debug(id + ": " + status);

      if (RolapUtil.LOGGER.isDebugEnabled()) {
        RolapUtil.LOGGER.debug(locus.component + ": executing sql [" + sql + "]" + status);
      }
    }
  }