BUG#15425 Small window for NF during backup failing without error

ndb/src/kernel/blocks/backup/Backup.cpp:
  If crash 10018 is inserted on a 2 node cluster with fast CPU and slower disk,
  all nodes can respond with BACKUP_FRAGMENT_CONF for all fragments before the
  error in FSAPPENDCONF is hit.
  
  This would mean that no error code was set for the backup and that it would
  be incomplete as not all IO had been written to disk before the node crash.
  This would not be reported to the user.
  
  So the backup would appear to succeed but it really didn't.
  
  The window for this is rather small though.
ndb/src/kernel/blocks/dbdict/Dbdict.cpp:
  Assert is incorrect for testBackup NFSlave test (causes crash insert 10015 to fail)

2 files changed, 10 insertions, 6 deletions

diff --git a/ndb/src/kernel/blocks/backup/Backup.cpp b/ndb/src/kernel/blocks/backup/Backup.cpp
index 56af24c5cf0..9997d365fa4 100644
--- a/ndb/src/kernel/blocks/backup/Backup.cpp
+++ b/ndb/src/kernel/blocks/backup/Backup.cpp
@@ -786,13 +786,17 @@ Backup::checkNodeFail(Signal* signal,
       pos= &ref->nodeId - signal->getDataPtr();
       break;
     }
+    case GSN_WAIT_GCP_REQ:
+    case GSN_DROP_TRIG_REQ:
     case GSN_CREATE_TRIG_REQ:
     case GSN_ALTER_TRIG_REQ:
-    case GSN_WAIT_GCP_REQ:
+      ptr.p->setErrorCode(AbortBackupOrd::BackupFailureDueToNodeFail);
+      return;
     case GSN_UTIL_SEQUENCE_REQ:
     case GSN_UTIL_LOCK_REQ:
-    case GSN_DROP_TRIG_REQ:
       return;
+    default:
+      ndbrequire(false);
     }
     
     for(Uint32 i = 0; (i = mask.find(i+1)) != NdbNodeBitmask::NotFound; )
@@ -1803,7 +1807,7 @@ Backup::execBACKUP_FRAGMENT_CONF(Signal* signal)
   const Uint32 nodeId = refToNode(signal->senderBlockRef());
   const Uint32 noOfBytes = conf->noOfBytes;
   const Uint32 noOfRecords = conf->noOfRecords;
-  
+
   BackupRecordPtr ptr;
   c_backupPool.getPtr(ptr, ptrI);
 
@@ -1880,7 +1884,7 @@ Backup::execBACKUP_FRAGMENT_REF(Signal* signal)
       }
     }
   }
-  ndbrequire(false);
+  goto err;
 
 done:
   ptr.p->masterData.sendCounter--;
@@ -1892,7 +1896,8 @@ done:
     masterAbort(signal, ptr);
     return;
   }//if
-  
+
+err:
   AbortBackupOrd *ord = (AbortBackupOrd*)signal->getDataPtrSend();
   ord->backupId = ptr.p->backupId;
   ord->backupPtr = ptr.i;
diff --git a/ndb/src/kernel/blocks/dbdict/Dbdict.cpp b/ndb/src/kernel/blocks/dbdict/Dbdict.cpp
index 5bd35812b47..6564963f61a 100644
--- a/ndb/src/kernel/blocks/dbdict/Dbdict.cpp
+++ b/ndb/src/kernel/blocks/dbdict/Dbdict.cpp
@@ -11694,7 +11694,6 @@ Dbdict::alterTrigger_toDropLocal(Signal* signal, OpAlterTriggerPtr opPtr)
     // broken trigger allowed if force
     if (! (triggerPtr.p->triggerLocal & TriggerRecord::TL_CREATED_LQH)) {
       jam();
-      ndbrequire(opPtr.p->m_requestFlag & RequestFlag::RF_FORCE);
       alterTrigger_sendReply(signal, opPtr, false);
       return;
     }