ndbeventoperation.hpp

mysql-5.0.22.tar.gz源码包

/* Copyright (C) 2003 MySQL AB

   This program is free software; you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
   the Free Software Foundation; either version 2 of the License, or
   (at your option) any later version.

   This program 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 General Public License for more details.

   You should have received a copy of the GNU General Public License
   along with this program; if not, write to the Free Software
   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA */

#ifndef NdbEventOperation_H
#define NdbEventOperation_H

class NdbGlobalEventBuffer;
class NdbEventOperationImpl;

/**
 * @class NdbEventOperation
 * @brief Class of operations for getting change events from database.  
 *
 * Brief description on how to work with events:
 *
 * - An event, represented by an NdbDictionary::Event, i created in the 
 *   Database through
 *   NdbDictionary::Dictionary::createEvent() (note that this can be done 
 *   by any application or thread and not necessarily by the "listener")
 * - To listen to events, an NdbEventOperation object is instantiated by 
 *   Ndb::createEventOperation()
 * - execute() starts the event flow. Use Ndb::pollEvents() to wait
 *   for an event to occur.  Use next() to iterate
 *   through the events that have occured.
 * - The instance is removed by Ndb::dropEventOperation()
 *
 * For more info see:
 * @ref ndbapi_event.cpp
 *
 * Known limitations:
 *
 * - Maximum number of active NdbEventOperations are now set at compile time.
 * Today 100.  This will become a configuration parameter later.
 * - Maximum number of NdbEventOperations tied to same event are maximum 16
 * per process.
 *
 * Known issues:
 *
 * - When several NdbEventOperation's are tied to the same event in the same
 * process they will share the circular buffer. The BufferLength will then
 * be the same for all and decided by the first NdbEventOperation 
 * instantiation. Just make sure to instantiate the "largest" one first.
 * - Today all events INSERT/DELETE/UPDATE and all changed attributes are
 * sent to the API, even if only specific attributes have been specified.
 * These are however hidden from the user and only relevant data is shown
 * after next().
 * - "False" exits from Ndb::pollEvents() may occur and thus
 * the subsequent next() will return zero,
 * since there was no available data. Just do Ndb::pollEvents() again.
 * - Event code does not check table schema version. Make sure to drop events
 * after table is dropped. Will be fixed in later
 * versions.
 * - If a node failure has occured not all events will be recieved
 * anymore. Drop NdbEventOperation and Create again after nodes are up
 * again. Will be fixed in later versions.
 *
 * Test status:
 *
 * - Tests have been run on 1-node and 2-node systems
 *
 * Useful API programs:
 *
 * - ndb_select_all -d sys 'NDB$EVENTS_0'
 * shows contents in the system table containing created events.
 *
 * @note this is an inteface to viewing events that is subject to change
 */
class NdbEventOperation {
public:
  /**
   * State of the NdbEventOperation object
   */
  enum State {
    EO_CREATED,   ///< Created but execute() not called
    EO_EXECUTING, ///< execute() called
    EO_ERROR      ///< An error has occurred. Object unusable.
  };
  /**
   * Retrieve current state of the NdbEventOperation object
   */
  State getState();

  /**
   * Activates the NdbEventOperation to start receiving events. The
   * changed attribute values may be retrieved after next() has returned
   * a value greater than zero. The getValue() methods must be called
   * prior to execute().
   *
   * @return 0 if successful otherwise -1.
   */
  int execute();

  // about the event operation
  // getting data
  //  NdbResultSet* getResultSet();

  /**
   * Defines a retrieval operation of an attribute value.
   * The NDB API allocate memory for the NdbRecAttr object that
   * will hold the returned attribute value. 
   *
   * @note Note that it is the applications responsibility
   *       to allocate enough memory for aValue (if non-NULL).
   *       The buffer aValue supplied by the application must be
   *       aligned appropriately.  The buffer is used directly
   *       (avoiding a copy penalty) only if it is aligned on a
   *       4-byte boundary and the attribute size in bytes
   *       (i.e. NdbRecAttr::attrSize() times NdbRecAttr::arraySize() is
   *       a multiple of 4).
   *
   * @note There are two versions, getValue() and
   *       getPreValue() for retrieving the current and
   *       previous value repectively.
   *
   * @note This method does not fetch the attribute value from 
   *       the database!  The NdbRecAttr object returned by this method 
   *       is <em>not</em> readable/printable before the 
   *       execute() has been made and
   *       next() has returned a value greater than
   *       zero. If a specific attribute has not changed the corresponding 
   *       NdbRecAttr will be in state UNDEFINED.  This is checked by 
   *       NdbRecAttr::isNULL() which then returns -1.
   *
   * @param anAttrName  Attribute name 
   * @param aValue      If this is non-NULL, then the attribute value 
   *                    will be returned in this parameter.<br>
   *                    If NULL, then the attribute value will only 
   *                    be stored in the returned NdbRecAttr object.
   * @return            An NdbRecAttr object to hold the value of 
   *                    the attribute, or a NULL pointer 
   *                    (indicating error).
   */
  NdbRecAttr *getValue(const char *anAttrName, char *aValue = 0);
  /**
   * See getValue().
   */
  NdbRecAttr *getPreValue(const char *anAttrName, char *aValue = 0);

  /**
   * Retrieves event resultset if available, inserted into the NdbRecAttrs
   * specified in getValue() and getPreValue(). To avoid polling for
   * a resultset, one can use Ndb::pollEvents()
   * which will wait on a mutex until an event occurs or the specified
   * timeout occurs.
   *
   * @return >=0 if successful otherwise -1. Return value indicates number
   * of available events. By sending pOverRun one may query for buffer
   * overflow and *pOverRun will indicate the number of events that have
   * overwritten.
   *
   * @return number of available events, -1 on failure
   */
  int next(int *pOverRun=0);

  /**
   * In the current implementation a nodefailiure may cause loss of events,
   * in which case isConsistent() will return false
   */
  bool isConsistent();

  /**
   * Query for occured event type.
   *
   * @note Only valid after next() has been called and returned value >= 0
   *
   * @return type of event
   */
  NdbDictionary::Event::TableEvent getEventType();

  /**
   * Retrieve the GCI of the latest retrieved event
   *
   * @return GCI number
   */
  Uint32 getGCI();

  /**
   * Retrieve the complete GCI in the cluster (not necessarily
   * associated with an event)
   *
   * @return GCI number
   */
  Uint32 getLatestGCI();

  /**
   * Get the latest error
   *
   * @return   Error object.
   */			     
  const struct NdbError & getNdbError() const;

#ifndef DOXYGEN_SHOULD_SKIP_INTERNAL
  /*
   *
   */
  void print();
#endif

private:
#ifndef DOXYGEN_SHOULD_SKIP_INTERNAL
  friend class NdbEventOperationImpl;
  friend class Ndb;
#endif
  NdbEventOperation(Ndb *theNdb, const char* eventName,int bufferLength);
  ~NdbEventOperation();
  static int wait(void *p, int aMillisecondNumber);
  class NdbEventOperationImpl &m_impl;
  NdbEventOperation(NdbEventOperationImpl& impl);
};

typedef void (* NdbEventCallback)(NdbEventOperation*, Ndb*, void*);
#endif