Commit e1c9dd5f authored by unknown's avatar unknown
Browse files

BUG#18966 Change in stop/shutdown behaviour 

Improvements that were discussed with Tomas.

Maintain protocol backwards/forwards compatibilty for client and server and support
the commands from ndb_mgm 'as expected'


ndb/include/mgmapi/mgmapi.h:
  Add ndb_mgm_stop3 mgmapi function.
   - This supports stopping all DB nodes, or all DB and MGM nodes.
   - It also returns to mgmapi program if it needs to disconnect to apply changes.
  
  Add ndb_mgm_restart3 mgmapi function.
   - Tells mgmapi program if it needs to disconnect to apply changes
  
  Add (internal) ndb_mgm_get_version
   - designed to be used to find out what protocol version we need to speak to the server.
ndb/src/mgmapi/mgmapi.cpp:
  Add cache of mgmd version to ndb_mgm_handle. Only filled out in functions that
  need to know the version of the mgmd we're talking to.
  Initialize these members in create handle.
  added ndb_mgm_get_version which asks the mgm server what version it is. This call
  has been supported since the dawn of time, no compatibility issues here.
  
  Add implementation of ndb_mgm_stop3
  Check what version of the protocol the server speaks, and speak it.
  Add compatibility for ndb_mgm_stop2
  
  Same for ndb_mgm_restart3.
ndb/src/mgmclient/CommandInterpreter.cpp:
  Simplify stop and restart code.
  
  Use the new ndb_mgm_(stop|restart)3 calls to find out if we need to disconnect.
ndb/src/mgmsrv/MgmtSrvr.cpp:
  Add nice call for shutting down MGM servers (like shutdownDB)
ndb/src/mgmsrv/MgmtSrvr.hpp:
  add prototype for shutdownMGM
ndb/src/mgmsrv/Services.cpp:
  Add restart node v2, stop v2, stop all v2 while maintaining protocol backwards
  compatibility.
  
  Unfortunately we can't add result lines due to protocol errors in clients :(
  Neither can we add extra things to the 'result: Ok' line due to the use of
  strcmp instead of strncmp.
ndb/src/mgmsrv/Services.hpp:
  Add prototypes for restart, stop and stopall v1 and v2
parent 564ba97b

ndb/include/mgmapi/mgmapi.h

+57 −0
Original line number Diff line number Diff line
@@ -694,6 +694,28 @@ extern "C" { 
  int ndb_mgm_stop2(NdbMgmHandle handle, int no_of_nodes,

		    const int * node_list, int abort);



  /**

   * Stops cluster nodes

   *

   * @param   handle        Management handle.

   * @param   no_of_nodes   Number of database nodes to stop<br>

   *                         -1: All database and management nodes<br>

   *                          0: All database nodes in cluster<br>

   *                          n: Stop the <var>n</var> node(s) specified in

   *                            the array node_list

   * @param   node_list     List of node IDs of database nodes to be stopped

   * @param   abort         Don't perform graceful stop,

   *                        but rather stop immediately

   * @param   disconnect    Returns true if you need to disconnect to apply

   *                        the stop command (e.g. stopping the mgm server

   *                        that handle is connected to)

   *

   * @return                Number of nodes stopped (-1 on error).

   */

  int ndb_mgm_stop3(NdbMgmHandle handle, int no_of_nodes,

		    const int * node_list, int abort, int *disconnect);





  /**

   * Restart database nodes

   *
@@ -733,6 +755,31 @@ extern "C" { 
		       const int * node_list, int initial,

		       int nostart, int abort);



  /**

   * Restart nodes

   *

   * @param   handle        Management handle.

   * @param   no_of_nodes   Number of database nodes to be restarted:<br>

   *                          0: Restart all database nodes in the cluster<br>

   *                          n: Restart the <var>n</var> node(s) specified in the

   *                            array node_list

   * @param   node_list     List of node IDs of database nodes to be restarted

   * @param   initial       Remove filesystem from restarting node(s)

   * @param   nostart       Don't actually start node(s) but leave them

   *                        waiting for start command

   * @param   abort         Don't perform graceful restart,

   *                        but rather restart immediately

   * @param   disconnect    Returns true if mgmapi client must disconnect from

   *                        server to apply the requested operation. (e.g.

   *                        restart the management server)

   *

   *

   * @return                Number of nodes stopped (-1 on error).

   */

  int ndb_mgm_restart3(NdbMgmHandle handle, int no_of_nodes,

		       const int * node_list, int initial,

		       int nostart, int abort, int *disconnect);



  /**

   * Start database nodes

   *
@@ -1023,6 +1070,16 @@ extern "C" { 
   */

  Uint32 ndb_mgm_get_mgmd_nodeid(NdbMgmHandle handle);



  /**

   * Get the version of the mgm server we're talking to.

   * Designed to allow switching of protocol depending on version

   * so that new clients can speak to old servers in a compat mode

   */

  int ndb_mgm_get_version(NdbMgmHandle handle,

                          int *major, int *minor, int* build,

                          int len, char* str);





  /**

   * Config iterator

   */

ndb/src/mgmapi/mgmapi.cpp

+174 −22
Original line number Diff line number Diff line
@@ -27,6 +27,7 @@ 
#include <mgmapi_debug.h>

#include "mgmapi_configuration.hpp"

#include <socket_io.h>

#include <version.h>



#include <NdbOut.hpp>

#include <SocketServer.hpp>
@@ -103,6 +104,9 @@ struct ndb_mgm_handle { 
#endif

  FILE *errstream;

  char *m_name;

  int mgmd_version_major;

  int mgmd_version_minor;

  int mgmd_version_build;

};



#define SET_ERROR(h, e, s) setError(h, e, __LINE__, s)
@@ -168,6 +172,10 @@ ndb_mgm_create_handle() 
  h->logfile = 0;

#endif



  h->mgmd_version_major= -1;

  h->mgmd_version_minor= -1;

  h->mgmd_version_build= -1;



  DBUG_PRINT("info", ("handle=0x%x", (UintPtr)h));

  DBUG_RETURN(h);

}
@@ -826,37 +834,75 @@ ndb_mgm_stop(NdbMgmHandle handle, int no_of_nodes, const int * node_list) 
  return ndb_mgm_stop2(handle, no_of_nodes, node_list, 0);

}





extern "C"

int

ndb_mgm_stop2(NdbMgmHandle handle, int no_of_nodes, const int * node_list,

	      int abort)

{

  SET_ERROR(handle, NDB_MGM_NO_ERROR, "Executing: ndb_mgm_stop2");

  const ParserRow<ParserDummy> stop_reply[] = {

  int disconnect;

  return ndb_mgm_stop3(handle, no_of_nodes, node_list, abort, &disconnect);

}





extern "C"

int

ndb_mgm_stop3(NdbMgmHandle handle, int no_of_nodes, const int * node_list,

	      int abort, int *disconnect)

{

  SET_ERROR(handle, NDB_MGM_NO_ERROR, "Executing: ndb_mgm_stop3");

  const ParserRow<ParserDummy> stop_reply_v1[] = {

    MGM_CMD("stop reply", NULL, ""),

    MGM_ARG("stopped", Int, Optional, "No of stopped nodes"),

    MGM_ARG("result", String, Mandatory, "Error message"),

    MGM_END()

  };

  const ParserRow<ParserDummy> stop_reply_v2[] = {

    MGM_CMD("stop reply", NULL, ""),

    MGM_ARG("stopped", Int, Optional, "No of stopped nodes"),

    MGM_ARG("result", String, Mandatory, "Error message"),

    MGM_ARG("disconnect", Int, Mandatory, "Need to disconnect"),

    MGM_END()

  };



  CHECK_HANDLE(handle, -1);

  CHECK_CONNECTED(handle, -1);



  if(no_of_nodes < 0){

  if(handle->mgmd_version_build==-1)

  {

    char verstr[50];

    ndb_mgm_get_version(handle,

                        &(handle->mgmd_version_major),

                        &(handle->mgmd_version_minor),

                        &(handle->mgmd_version_build),

                        sizeof(verstr),

                        verstr);

  }

  int use_v2= (handle->mgmd_version_major==5)

    && (

        (handle->mgmd_version_minor==0 && handle->mgmd_version_build>=21)

        ||(handle->mgmd_version_minor==1 && handle->mgmd_version_build>=12)

        );



  if(no_of_nodes < -1){

    SET_ERROR(handle, NDB_MGM_ILLEGAL_NUMBER_OF_NODES, 

	      "Negative number of nodes requested to stop");

    return -1;

  }



  Uint32 stoppedNoOfNodes = 0;

  if(no_of_nodes == 0){

  if(no_of_nodes <= 0){

    /**

     * All database nodes should be stopped

     * All nodes should be stopped (all or just db)

     */

    Properties args;

    args.put("abort", abort);

    if(use_v2)

      args.put("stop", (no_of_nodes==-1)?"mgm,db":"db");

    const Properties *reply;

    reply = ndb_mgm_call(handle, stop_reply, "stop all", &args);

    if(use_v2)

      reply = ndb_mgm_call(handle, stop_reply_v2, "stop all v2", &args);

    else

      reply = ndb_mgm_call(handle, stop_reply_v1, "stop all", &args);

    CHECK_REPLY(reply, -1);



    if(!reply->get("stopped", &stoppedNoOfNodes)){
@@ -865,6 +911,10 @@ ndb_mgm_stop2(NdbMgmHandle handle, int no_of_nodes, const int * node_list, 
      delete reply;

      return -1;

    }

    if(use_v2)

      reply->get("disconnect", (Uint32*)disconnect);

    else

      *disconnect= 0;

    BaseString result;

    reply->get("result", result);

    if(strcmp(result.c_str(), "Ok") != 0) {
@@ -890,7 +940,11 @@ ndb_mgm_stop2(NdbMgmHandle handle, int no_of_nodes, const int * node_list, 
  args.put("abort", abort);



  const Properties *reply;

  reply = ndb_mgm_call(handle, stop_reply, "stop", &args);

  if(use_v2)

    reply = ndb_mgm_call(handle, stop_reply_v2, "stop v2", &args);

  else

    reply = ndb_mgm_call(handle, stop_reply_v1, "stop", &args);



  CHECK_REPLY(reply, stoppedNoOfNodes);

  if(!reply->get("stopped", &stoppedNoOfNodes)){

    SET_ERROR(handle, NDB_MGM_STOP_FAILED,
@@ -898,6 +952,10 @@ ndb_mgm_stop2(NdbMgmHandle handle, int no_of_nodes, const int * node_list, 
    delete reply;

    return -1;

  }

  if(use_v2)

    reply->get("disconnect", (Uint32*)disconnect);

  else

    *disconnect= 0;

  BaseString result;

  reply->get("result", result);

  if(strcmp(result.c_str(), "Ok") != 0) {
@@ -909,22 +967,65 @@ ndb_mgm_stop2(NdbMgmHandle handle, int no_of_nodes, const int * node_list, 
  return stoppedNoOfNodes;

}



extern "C"

int

ndb_mgm_restart(NdbMgmHandle handle, int no_of_nodes, const int *node_list) 

{

  SET_ERROR(handle, NDB_MGM_NO_ERROR, "Executing: ndb_mgm_restart");

  return ndb_mgm_restart2(handle, no_of_nodes, node_list, 0, 0, 0);

}



extern "C"

int

ndb_mgm_restart2(NdbMgmHandle handle, int no_of_nodes, const int * node_list,

		 int initial, int nostart, int abort)

{

  SET_ERROR(handle, NDB_MGM_NO_ERROR, "Executing: ndb_mgm_restart2");

  int disconnect;



  return ndb_mgm_restart3(handle, no_of_nodes, node_list, initial, nostart,

                          abort, &disconnect);

}



extern "C"

int

ndb_mgm_restart3(NdbMgmHandle handle, int no_of_nodes, const int * node_list,

		 int initial, int nostart, int abort, int *disconnect)

{

  SET_ERROR(handle, NDB_MGM_NO_ERROR, "Executing: ndb_mgm_restart3");

  Uint32 restarted = 0;

  const ParserRow<ParserDummy> restart_reply[] = {

  const ParserRow<ParserDummy> restart_reply_v1[] = {

    MGM_CMD("restart reply", NULL, ""),

    MGM_ARG("result", String, Mandatory, "Error message"),

    MGM_ARG("restarted", Int, Optional, "No of restarted nodes"),

    MGM_END()

  };

  const ParserRow<ParserDummy> restart_reply_v2[] = {

    MGM_CMD("restart reply", NULL, ""),

    MGM_ARG("result", String, Mandatory, "Error message"),

    MGM_ARG("restarted", Int, Optional, "No of restarted nodes"),

    MGM_ARG("disconnect", Int, Optional, "Disconnect to apply"),

    MGM_END()

  };



  CHECK_HANDLE(handle, -1);

  CHECK_CONNECTED(handle, -1);



  if(handle->mgmd_version_build==-1)

  {

    char verstr[50];

    ndb_mgm_get_version(handle,

                        &(handle->mgmd_version_major),

                        &(handle->mgmd_version_minor),

                        &(handle->mgmd_version_build),

                        sizeof(verstr),

                        verstr);

  }

  int use_v2= (handle->mgmd_version_major==5)

    && (

        (handle->mgmd_version_minor==0 && handle->mgmd_version_build>=21)

        ||(handle->mgmd_version_minor==1 && handle->mgmd_version_build>=12)

        );



  if(no_of_nodes < 0){

    SET_ERROR(handle, NDB_MGM_RESTART_FAILED, 

	      "Restart requested of negative number of nodes");
@@ -939,7 +1040,7 @@ ndb_mgm_restart2(NdbMgmHandle handle, int no_of_nodes, const int * node_list, 
    const Properties *reply;

    const int timeout = handle->read_timeout;

    handle->read_timeout= 5*60*1000; // 5 minutes

    reply = ndb_mgm_call(handle, restart_reply, "restart all", &args);

    reply = ndb_mgm_call(handle, restart_reply_v1, "restart all", &args);

    handle->read_timeout= timeout;

    CHECK_REPLY(reply, -1);
@@ -975,7 +1076,10 @@ ndb_mgm_restart2(NdbMgmHandle handle, int no_of_nodes, const int * node_list, 
  const Properties *reply;

  const int timeout = handle->read_timeout;

  handle->read_timeout= 5*60*1000; // 5 minutes

  reply = ndb_mgm_call(handle, restart_reply, "restart node", &args);

  if(use_v2)

    reply = ndb_mgm_call(handle, restart_reply_v2, "restart node v2", &args);

  else

    reply = ndb_mgm_call(handle, restart_reply_v1, "restart node", &args);

  handle->read_timeout= timeout;

  if(reply != NULL) {

    BaseString result;
@@ -986,20 +1090,16 @@ ndb_mgm_restart2(NdbMgmHandle handle, int no_of_nodes, const int * node_list, 
      return -1;

    }

    reply->get("restarted", &restarted);

    if(use_v2)

      reply->get("disconnect", (Uint32*)disconnect);

    else

      *disconnect= 0;

    delete reply;

  } 

  

  return restarted;

}



extern "C"

int

ndb_mgm_restart(NdbMgmHandle handle, int no_of_nodes, const int *node_list) 

{

  SET_ERROR(handle, NDB_MGM_NO_ERROR, "Executing: ndb_mgm_restart");

  return ndb_mgm_restart2(handle, no_of_nodes, node_list, 0, 0, 0);

}



static const char *clusterlog_severity_names[]=

  { "enabled", "debug", "info", "warning", "error", "critical", "alert" };
@@ -2353,4 +2453,56 @@ int ndb_mgm_end_session(NdbMgmHandle handle) 
  DBUG_RETURN(0);

}



extern "C"

int ndb_mgm_get_version(NdbMgmHandle handle,

                        int *major, int *minor, int *build, int len, char* str)

{

  DBUG_ENTER("ndb_mgm_get_version");

  CHECK_HANDLE(handle, 0);

  CHECK_CONNECTED(handle, 0);



  Properties args;



  const ParserRow<ParserDummy> reply[]= {

    MGM_CMD("version", NULL, ""),

    MGM_ARG("id", Int, Mandatory, "ID"),

    MGM_ARG("major", Int, Mandatory, "Major"),

    MGM_ARG("minor", Int, Mandatory, "Minor"),

    MGM_ARG("string", String, Mandatory, "String"),

    MGM_END()

  };



  const Properties *prop;

  prop = ndb_mgm_call(handle, reply, "get version", &args);

  CHECK_REPLY(prop, 0);



  Uint32 id;

  if(!prop->get("id",&id)){

    fprintf(handle->errstream, "Unable to get value\n");

    return 0;

  }

  *build= getBuild(id);



  if(!prop->get("major",(Uint32*)major)){

    fprintf(handle->errstream, "Unable to get value\n");

    return 0;

  }



  if(!prop->get("minor",(Uint32*)minor)){

    fprintf(handle->errstream, "Unable to get value\n");

    return 0;

  }



  BaseString result;

  if(!prop->get("string", result)){

    fprintf(handle->errstream, "Unable to get value\n");

    return 0;

  }



  strncpy(str, result.c_str(), len);



  delete prop;

  DBUG_RETURN(1);

}



template class Vector<const ParserRow<ParserDummy>*>;

ndb/src/mgmclient/CommandInterpreter.cpp

+19 −56
Original line number Diff line number Diff line
@@ -1057,7 +1057,8 @@ CommandInterpreter::executeShutdown(char* parameters) 
  NdbAutoPtr<char> ap1((char*)state);



  int result = 0;

  result = ndb_mgm_stop(m_mgmsrv, 0, 0);

  int need_disconnect;

  result = ndb_mgm_stop3(m_mgmsrv, -1, 0, 0, &need_disconnect);

  if (result < 0) {

    ndbout << "Shutdown of NDB Cluster node(s) failed." << endl;

    printError();
@@ -1066,39 +1067,11 @@ CommandInterpreter::executeShutdown(char* parameters) 


  ndbout << result << " NDB Cluster node(s) have shutdown." << endl;



  int nodeId= 0;

  int this_mgmd= 0;

  this_mgmd= ndb_mgm_get_mgmd_nodeid(m_mgmsrv);

  while(get_next_nodeid(state, &nodeId, NDB_MGM_NODE_TYPE_MGM))

  {

    if(nodeId==this_mgmd)

      continue;

    ndbout << "Shutting down NDB Cluster management server nodeId="

           << nodeId << "...";

    result = ndb_mgm_stop(m_mgmsrv, 1, &nodeId);

    if (result <= 0) {

      ndbout << " failed." << endl;

      printError();

    }

    else

      ndbout << "Done." << endl;

  }



  ndbout << "Shutting down NDB Cluster management server nodeId="

         << this_mgmd << "...";

  result= ndb_mgm_stop(m_mgmsrv, 1, &this_mgmd);

  if (result <= 0) {

    ndbout << " failed." << endl;

    printError();

  }

  else

  {

    ndbout << "Done." << endl;

  if(need_disconnect) {

    ndbout << "Disconnecting to allow management server to shutdown."

           << endl;

    disconnect();

  }

  ndbout << "NDB Cluster management servers shutdown." << endl;

  return 0;

}
@@ -1487,6 +1460,7 @@ CommandInterpreter::executeStop(Vector<BaseString> &command_list, 
                                unsigned command_pos,

                                int *node_ids, int no_of_nodes)

{

  int need_disconnect;

  int abort= 0;

  for (; command_pos < command_list.size(); command_pos++)

  {
@@ -1501,7 +1475,8 @@ CommandInterpreter::executeStop(Vector<BaseString> &command_list, 
    return;

  }



  int result= ndb_mgm_stop2(m_mgmsrv, no_of_nodes, node_ids, abort);

  int result= ndb_mgm_stop3(m_mgmsrv, no_of_nodes, node_ids, abort,

                            &need_disconnect);

  if (result < 0)

  {

    ndbout_c("Shutdown failed.");
@@ -1513,27 +1488,19 @@ CommandInterpreter::executeStop(Vector<BaseString> &command_list, 
      ndbout_c("NDB Cluster has shutdown.");

    else

    {

      int mgm_id= 0;

      int need_reconnect= 0;

      mgm_id= ndb_mgm_get_mgmd_nodeid(m_mgmsrv);

      ndbout << "Node";

      for (int i= 0; i < no_of_nodes; i++)

      {

        if(node_ids[i] == mgm_id)

          need_reconnect= 1;

        else

          ndbout << " " << node_ids[i];

      }

      ndbout_c(" has shutdown.");

      if(need_reconnect)

      {

        ndbout << "You are connected to node " << mgm_id

               << ", disconnecting to allow it to shutdown"

               << endl;

        disconnect();

    }

  }



  if(need_disconnect)

  {

    ndbout << "Disconnecting to allow Management Server to shutdown" << endl;

    disconnect();

  }



}



void
@@ -1624,6 +1591,7 @@ CommandInterpreter::executeRestart(Vector<BaseString> &command_list, 
  int nostart= 0;

  int initialstart= 0;

  int abort= 0;

  int need_disconnect= 0;



  for (; command_pos < command_list.size(); command_pos++)

  {
@@ -1648,8 +1616,8 @@ CommandInterpreter::executeRestart(Vector<BaseString> &command_list, 
    return;

  }



  result= ndb_mgm_restart2(m_mgmsrv, no_of_nodes, node_ids,

                           initialstart, nostart, abort);

  result= ndb_mgm_restart3(m_mgmsrv, no_of_nodes, node_ids,

                           initialstart, nostart, abort, &need_disconnect);



  if (result <= 0) {

    ndbout_c("Restart failed.");
@@ -1661,18 +1629,13 @@ CommandInterpreter::executeRestart(Vector<BaseString> &command_list, 
      ndbout_c("NDB Cluster is being restarted.");

    else

    {

      int mgm_id= 0;

      mgm_id= ndb_mgm_get_mgmd_nodeid(m_mgmsrv);



      ndbout << "Node";

      for (int i= 0; i < no_of_nodes; i++)

      {

        if(node_ids[i] == mgm_id)

          disconnect();

        ndbout << " " << node_ids[i];

      }

      ndbout_c(" is being restarted");

    }

    if(need_disconnect)

      disconnect();

  }

}

ndb/src/mgmsrv/MgmtSrvr.cpp

+21 −0
Original line number Diff line number Diff line
@@ -1196,6 +1196,27 @@ int MgmtSrvr::stopNodes(const Vector<NodeId> &node_ids, 
  return ret;

}



int MgmtSrvr::shutdownMGM(int *stopCount, bool abort, int *stopSelf)

{

  NodeId nodeId = 0;

  int error;



  while(getNextNodeId(&nodeId, NDB_MGM_NODE_TYPE_MGM))

  {

    if(nodeId==getOwnNodeId())

      continue;

    error= sendStopMgmd(nodeId, abort, true, false,

                        false, false);

    if (error == 0)

      *stopCount++;

  }



  *stopSelf= 1;

  *stopCount++;



  return 0;

}



/*

 * Perform DB nodes shutdown.

 * MGM servers are left in their current state

ndb/src/mgmsrv/MgmtSrvr.hpp

+2 −0
Original line number Diff line number Diff line
@@ -256,6 +256,8 @@ public: 
  int stopNodes(const Vector<NodeId> &node_ids, int *stopCount, bool abort,

                int *stopSelf);



  int shutdownMGM(int *stopCount, bool abort, int *stopSelf);



  /**

   * shutdown the DB nodes

   */