Commit fe9934c9 authored by tomas@poseidon.ndb.mysql.com's avatar tomas@poseidon.ndb.mysql.com
Browse files

Bug #16875 Using stale MySQLD FRM files can cause restored cluster to fail 

- invalidate ndb dict cache on cluster disconnect (ClusterMgr.cpp)
- add check for correct frm on external lock when table cache is found invalid
parent 133a5d17

mysql-test/r/ndb_autodiscover3.result

0 → 100644
+9 −0
Original line number Diff line number Diff line 
drop table if exists t1;

create table t1 (a int key) engine=ndbcluster;

begin;

insert into t1 values (1);

insert into t1 values (2);

ERROR HY000: Got temporary error 4025 'Node failure caused abort of transaction' from ndbcluster

commit;

ERROR HY000: Got error 4350 'Transaction already aborted' from ndbcluster

drop table t1;

mysql-test/t/ndb_autodiscover3.test

0 → 100644
+65 −0
Original line number Diff line number Diff line 
-- source include/have_ndb.inc

-- source include/have_multi_ndb.inc

-- source include/not_embedded.inc





--disable_warnings

drop table if exists t1, t2;

--enable_warnings



#

# Transaction ongoing while cluster is restarted

#

--connection server1

create table t1 (a int key) engine=ndbcluster;



begin;

insert into t1 values (1);



--exec $NDB_MGM --no-defaults -e "all restart" >> $NDB_TOOLS_OUTPUT

--exec $NDB_TOOLS_DIR/ndb_waiter --no-defaults >> $NDB_TOOLS_OUTPUT



--error 1297

insert into t1 values (2);

--error 1296

commit;



drop table t1;



#

# Stale cache after restart -i

#

--connection server1

create table t2 (a int, b int, primary key(a,b)) engine=ndbcluster;

insert into t2 values (1,1),(2,1),(3,1),(4,1),(5,1),(6,1),(7,1),(8,1),(9,1),(10,1);

select * from t2 order by a limit 3;



--exec $NDB_MGM --no-defaults -e "all restart -i" >> $NDB_TOOLS_OUTPUT

--exec $NDB_TOOLS_DIR/ndb_waiter --no-defaults >> $NDB_TOOLS_OUTPUT



--connection server2

create table t2 (a int key) engine=ndbcluster;

insert into t2 values (1),(2),(3),(4),(5),(6),(7),(8),(9),(10);

select * from t2 order by a limit 3;



# server 1 should have a stale cache, and in this case wrong frm, transaction must be retried

--connection server1

--error 1015

select * from t2 order by a limit 3;

select * from t2 order by a limit 3;



--exec $NDB_MGM --no-defaults -e "all restart -i" >> $NDB_TOOLS_OUTPUT

--exec $NDB_TOOLS_DIR/ndb_waiter --no-defaults >> $NDB_TOOLS_OUTPUT



--connection server1

show tables;

create table t2 (a int key) engine=ndbcluster;

insert into t2 values (1),(2),(3),(4),(5),(6),(7),(8),(9),(10);

select * from t2 order by a limit 3;



# server 2 should have a stale cache, but with right frm, transaction need not be retried

--connection server2

select * from t2 order by a limit 3;



drop table t2;

# End of 4.1 tests

ndb/include/ndbapi/ndb_cluster_connection.hpp

+1 −0
Original line number Diff line number Diff line
@@ -83,6 +83,7 @@ public: 
  void set_optimized_node_selection(int val);



  unsigned no_db_nodes();

  unsigned get_connect_count() const;

#endif



private:

ndb/src/ndbapi/ClusterMgr.cpp

+5 −0
Original line number Diff line number Diff line
@@ -70,6 +70,7 @@ ClusterMgr::ClusterMgr(TransporterFacade & _facade): 
  noOfAliveNodes= 0;

  noOfConnectedNodes= 0;

  theClusterMgrThread= 0;

  m_connect_count = 0;

  DBUG_VOID_RETURN;

}
@@ -456,6 +457,10 @@ ClusterMgr::reportNodeFailed(NodeId nodeId){ 
  theNode.nfCompleteRep = false;

  

  if(noOfAliveNodes == 0){

    theFacade.m_globalDictCache.lock();

    theFacade.m_globalDictCache.invalidate_all();

    theFacade.m_globalDictCache.unlock();

    m_connect_count ++;

    NFCompleteRep rep;

    for(Uint32 i = 1; i<MAX_NODES; i++){

      if(theNodes[i].defined && theNodes[i].nfCompleteRep == false){

ndb/src/ndbapi/ClusterMgr.hpp

+1 −0
Original line number Diff line number Diff line
@@ -78,6 +78,7 @@ public: 
  

  const Node &  getNodeInfo(NodeId) const;

  Uint32        getNoOfConnectedNodes() const;

  Uint32        m_connect_count;

  

private:

  Uint32        noOfAliveNodes;