fix replica set cursor cleaning through mongos SERVER-2505

fix cursor cleaning in mongos generally by not doing lazy cleaning
mongodb · Feb 10, 2011 · 37b11e4 · 37b11e4
1 parent e0c8d9a
commit 37b11e4
Show file tree

Hide file tree

Showing 7 changed files with 37 additions and 4 deletions.
diff --git a/client/dbclient.cpp b/client/dbclient.cpp
@@ -902,11 +902,15 @@ namespace mongo {
 
         Message m;
         m.setData( dbKillCursors , b.buf() , b.len() );
-
-        sayPiggyBack( m );
+
+        if ( _lazyKillCursor )
+            sayPiggyBack( m );
+        else
+            say(m);
     }
 
     AtomicUInt DBClientConnection::_numConnections;
+    bool DBClientConnection::_lazyKillCursor = true;
 
 
     bool serverAlive( const string &uri ) {

diff --git a/client/dbclient.h b/client/dbclient.h
@@ -900,6 +900,9 @@ namespace mongo {
         static int getNumConnections() {
             return _numConnections;
         }
+
+        static void setLazyKillCursor( bool lazy ) { _lazyKillCursor = lazy; }
+        static bool getLazyKillCursor() { return _lazyKillCursor; }
 
     protected:
         friend class SyncClusterConnection;
@@ -924,6 +927,7 @@ namespace mongo {
         bool _connect( string& errmsg );
 
         static AtomicUInt _numConnections;
+        static bool _lazyKillCursor; // lazy means we piggy back kill cursors on next op
     };
 
     /** pings server to check if it's up

diff --git a/client/dbclient_rs.cpp b/client/dbclient_rs.cpp
@@ -522,6 +522,10 @@ namespace mongo {
     }
 
     void DBClientReplicaSet::killCursor( long long cursorID ) {
+        // we should neve call killCursor on a replica set conncetion
+        // since we don't know which server it belongs to
+        // can't assume master because of slave ok
+        // and can have a cursor survive a master change
         assert(0);
     }
 
@@ -547,7 +551,11 @@ namespace mongo {
                 }
             }
         }
-        return checkMaster()->call( toSend , response , assertOk );
+
+        DBClientConnection* m = checkMaster();
+        if ( actualServer )
+            *actualServer = m->getServerAddress();
+        return m->call( toSend , response , assertOk );
     }
 
 }
diff --git a/jstests/slowNightly/sharding_rs2.js b/jstests/slowNightly/sharding_rs2.js
@@ -95,6 +95,9 @@ assert.eq( 100 , ts.count() , "B4" )
 assert.eq( 100 , ts.find().itcount() , "B5" )
 assert.eq( 100 , ts.find().batchSize(5).itcount() , "B6" )
 
+t.find().batchSize(3).next();
+gc(); gc(); gc();
+
 // --- sharded ----
 
 assert.eq( 100 , db.foo.count() , "C1" )

diff --git a/s/cursors.cpp b/s/cursors.cpp
@@ -163,12 +163,14 @@ namespace mongo {
     }
 
     void CursorCache::storeRef( const string& server , long long id ) {
+        LOG(_myLogLevel) << "CursorCache::storeRef server: " << server << " id: " << id << endl;
         assert( id );
         scoped_lock lk( _mutex );
         _refs[id] = server;
     }
 
     string CursorCache::getRef( long long id ) const {
+        LOG(_myLogLevel) << "CursorCache::getRef id: " << id << endl;
         assert( id );
         scoped_lock lk( _mutex );
         MapNormal::const_iterator i = _refs.find( id );
@@ -215,6 +217,8 @@ namespace mongo {
         long long * cursors = (long long *)x;
         for ( int i=0; i<n; i++ ) {
             long long id = cursors[i];
+            LOG(_myLogLevel) << "CursorCache::gotKillCursors id: " << id << endl;
+
             if ( ! id ) {
                 log( LL_WARNING ) << " got cursor id of 0 to kill" << endl;
                 continue;
@@ -239,6 +243,8 @@ namespace mongo {
                 _refs.erase( j );
             }
 
+            LOG(_myLogLevel) << "CursorCache::found gotKillCursors id: " << id << " server: " << server << endl;
+
             assert( server.size() );
             ScopedDbConnection conn( server );
             conn->killCursor( id );
@@ -269,6 +275,8 @@ namespace mongo {
 
     CursorCache cursorCache;
 
+    int CursorCache::_myLogLevel = 3;
+
     class CursorTimeoutTask : public task::Task {
     public:
         virtual string name() const { return "cursorTimeout"; }

diff --git a/s/cursors.h b/s/cursors.h
@@ -98,6 +98,8 @@ namespace mongo {
         MapNormal _refs;
 
         long long _shardedTotal;
+
+        static int _myLogLevel;
     };
 
     extern CursorCache cursorCache;

diff --git a/s/server.cpp b/s/server.cpp
@@ -276,12 +276,16 @@ int _main(int argc, char* argv[]) {
             return 9;
         }
     }
+
+    // set some global state
 
     pool.addHook( &shardingConnectionHook );
     pool.setName( "mongos connectionpool" );
+
+    DBClientConnection::setLazyKillCursor( false );
 
     ReplicaSetMonitor::setConfigChangeHook( boost::bind( &ConfigServer::replicaSetChange , &configServer , _1 ) );
-
+    
     if ( argc <= 1 ) {
         usage( argv );
         return 3;