Commit 855c4e06 authored by unknown's avatar unknown
Browse files

ndb - bug#23210 

  Fix race-condition between COPY_GCIREQ (GCP) and lcpSetActiveStatusEnd
  Solution is _not_ to copy sysfileData from COPY_GCIREQ from "self"


ndb/src/kernel/blocks/ERROR_codes.txt:
  Add error insert for dealying of copy sysfileData
ndb/src/kernel/blocks/dbdih/DbdihMain.cpp:
  1) Add error insert for delaying of sysfileData
  2) Change to that master is _not_ copying sysfileData from COPY_GCIREQ
     as it might be updating it while COPY_GCIREQ is "in the fly"
parent 311abf10

ndb/src/kernel/blocks/ERROR_codes.txt

+3 −1
Original line number Diff line number Diff line
@@ -5,7 +5,7 @@ Next DBACC 3002 
Next DBTUP 4014

Next DBLQH 5043

Next DBDICT 6006

Next DBDIH 7174

Next DBDIH 7178

Next DBTC 8038

Next CMVMI 9000

Next BACKUP 10022
@@ -66,6 +66,8 @@ Delay GCP_SAVEREQ by 10 secs 
7030: Delay in GCP_PREPARE until node has completed a node failure

7031: Delay in GCP_PREPARE and die 3s later



7177: Delay copying of sysfileData in execCOPY_GCIREQ



ERROR CODES FOR TESTING NODE FAILURE, LOCAL CHECKPOINT HANDLING:

-----------------------------------------------------------------

ndb/src/kernel/blocks/dbdih/DbdihMain.cpp

+32 −6
Original line number Diff line number Diff line
@@ -626,21 +626,47 @@ void Dbdih::execCOPY_GCIREQ(Signal* signal) 
  ndbrequire(c_copyGCISlave.m_copyReason  == CopyGCIReq::IDLE);

  ndbrequire(c_copyGCISlave.m_expectedNextWord == tstart);

  ndbrequire(reason != CopyGCIReq::IDLE);

  bool isdone = (tstart + CopyGCIReq::DATA_SIZE) >= Sysfile::SYSFILE_SIZE32;



  if (ERROR_INSERTED(7177))

  {

    jam();



    if (signal->getLength() == 3)

    {

      jam();

      goto done;

    }

  }



  arrGuard(tstart + CopyGCIReq::DATA_SIZE, sizeof(sysfileData)/4);

  for(Uint32 i = 0; i<CopyGCIReq::DATA_SIZE; i++)

    cdata[tstart+i] = copyGCI->data[i];

  

  if ((tstart + CopyGCIReq::DATA_SIZE) >= Sysfile::SYSFILE_SIZE32) {

  if (ERROR_INSERTED(7177) && isMaster() && isdone)

  {

    sendSignalWithDelay(reference(), GSN_COPY_GCIREQ, signal, 1000, 3);

    return;

  }

  

done:  

  if (isdone)

  {

    jam();

    c_copyGCISlave.m_expectedNextWord = 0;

  } else {

  } 

  else 

  {

    jam();

    c_copyGCISlave.m_expectedNextWord += CopyGCIReq::DATA_SIZE;

    return;

  }//if

  }

  

  if (cmasterdihref != reference())

  {

    jam();

    memcpy(sysfileData, cdata, sizeof(sysfileData));

  }



  c_copyGCISlave.m_copyReason = reason;

  c_copyGCISlave.m_senderRef  = signal->senderBlockRef();