Merge branch 'master' of github.com:mongodb/mongo

23 files changed, 200 insertions, 94 deletions

diff --git a/client/dbclient.h b/client/dbclient.h
index 5d6d87cc129..03030a3c6be 100644
--- a/client/dbclient.h
+++ b/client/dbclient.h
@@ -884,7 +884,7 @@ namespace mongo {
          */
         bool isFailed() const { return _failed; }
 
-        MessagingPort& port() { return *p; }
+        MessagingPort& port() { assert(p); return *p; }
 
         string toStringLong() const {
             stringstream ss;
diff --git a/client/dbclient_rs.cpp b/client/dbclient_rs.cpp
index 1fb8e755e47..217c38804a5 100644
--- a/client/dbclient_rs.cpp
+++ b/client/dbclient_rs.cpp
@@ -74,7 +74,7 @@ namespace mongo {
 
 
     ReplicaSetMonitor::ReplicaSetMonitor( const string& name , const vector<HostAndPort>& servers )
-        : _lock( "ReplicaSetMonitor instance" ) , _checkConnectionLock( "ReplicaSetMonitor check connection lock" ), _name( name ) , _master(-1) {
+        : _lock( "ReplicaSetMonitor instance" ) , _checkConnectionLock( "ReplicaSetMonitor check connection lock" ), _name( name ) , _master(-1), _nextSlave(0) {
         
         uassert( 13642 , "need at least 1 node for a replica set" , servers.size() > 0 );
 
@@ -221,19 +221,19 @@ namespace mongo {
     }
 
     HostAndPort ReplicaSetMonitor::getSlave() {
-        int x = rand() % _nodes.size();
+
         {
             scoped_lock lk( _lock );
             for ( unsigned i=0; i<_nodes.size(); i++ ) {
-                int p = ( i + x ) % _nodes.size();
-                if ( p == _master )
+                _nextSlave = ( _nextSlave + 1 ) % _nodes.size();
+                if ( _nextSlave == _master )
                     continue;
-                if ( _nodes[p].ok )
-                    return _nodes[p].addr;
+                if ( _nodes[ _nextSlave ].ok )
+                    return _nodes[ _nextSlave ].addr;
             }
         }
 
-        return _nodes[0].addr;
+        return _nodes[ 0 ].addr;
     }
 
     /**
@@ -535,7 +535,7 @@ namespace mongo {
                     return checkSlave()->query(ns,query,nToReturn,nToSkip,fieldsToReturn,queryOptions,batchSize);
                 }
                 catch ( DBException &e ) {
-                    log() << "can't query replica set slave: " << _slaveHost << causedBy( e ) << endl;
+                    LOG(1) << "can't query replica set slave " << i << " : " << _slaveHost << causedBy( e ) << endl;
                 }
             }
         }
@@ -552,8 +552,8 @@ namespace mongo {
                 try {
                     return checkSlave()->findOne(ns,query,fieldsToReturn,queryOptions);
                 }
-                catch ( DBException & ) {
-                    LOG(1) << "can't query replica set slave: " << _slaveHost << endl;
+                catch ( DBException &e ) {
+                	LOG(1) << "can't findone replica set slave " << i << " : " << _slaveHost << causedBy( e ) << endl;
                 }
             }
         }
@@ -585,8 +585,8 @@ namespace mongo {
                     try {
                         return checkSlave()->callLazy( toSend );
                     }
-                    catch ( DBException & ) {
-                        log(1) << "can't query replica set slave: " << _slaveHost << endl;
+                    catch ( DBException &e ) {
+                    	LOG(1) << "can't callLazy replica set slave " << i << " : " << _slaveHost << causedBy( e ) << endl;
                     }
                 }
             }
@@ -608,8 +608,8 @@ namespace mongo {
                             *actualServer = s->getServerAddress();
                         return s->call( toSend , response , assertOk );
                     }
-                    catch ( DBException & ) {
-                        log(1) << "can't query replica set slave: " << _slaveHost << endl;
+                    catch ( DBException &e ) {
+                    	LOG(1) << "can't call replica set slave " << i << " : " << _slaveHost << causedBy( e ) << endl;
                         if ( actualServer )
                             *actualServer = "";
                     }
diff --git a/client/dbclient_rs.h b/client/dbclient_rs.h
index f65fe3d3ad4..f0e855ae8c4 100644
--- a/client/dbclient_rs.h
+++ b/client/dbclient_rs.h
@@ -147,7 +147,7 @@ namespace mongo {
         vector<Node> _nodes;
 
         int _master; // which node is the current master.  -1 means no master is known
-
+        int _nextSlave; // which node is the current slave
 
         static mongo::mutex _setsLock; // protects _sets
         static map<string,ReplicaSetMonitorPtr> _sets; // set name to Monitor
diff --git a/client/examples/clientTest.cpp b/client/examples/clientTest.cpp
index 96c014e245c..aaea6bd1bdf 100644
--- a/client/examples/clientTest.cpp
+++ b/client/examples/clientTest.cpp
@@ -246,5 +246,34 @@ int main( int argc, const char **argv ) {
         //MONGO_PRINT(out);
     }
 
+    { 
+        // test timeouts
+
+        DBClientConnection conn( true , 0 , 2 );
+        if ( ! conn.connect( string( "127.0.0.1:" ) + port , errmsg ) ) {
+            cout << "couldn't connect : " << errmsg << endl;
+            throw -11;
+        }
+        conn.insert( "test.totest" , BSON( "x" << 1 ) );
+        BSONObj res;
+        
+        bool gotError = false;
+        assert( conn.eval( "test" , "return db.totest.findOne().x" , res ) );
+        try {
+            conn.eval( "test" , "sleep(5000); return db.totest.findOne().x" , res );
+        }
+        catch ( std::exception& e ) {
+            gotError = true;
+            log() << e.what() << endl;
+        }
+        assert( gotError );
+        // sleep so the server isn't locked anymore
+        sleepsecs( 4 );
+        
+        assert( conn.eval( "test" , "return db.totest.findOne().x" , res ) );
+        
+        
+    }
+
     cout << "client test finished!" << endl;
 }
diff --git a/db/db.cpp b/db/db.cpp
index 72f5631bf43..18ffa36cc14 100644
--- a/db/db.cpp
+++ b/db/db.cpp
@@ -284,7 +284,9 @@ namespace mongo {
             if ( !h->isCurrentVersion() || forceRepair ) {
 
                 if( h->version <= 0 ) {
-                    uasserted(14026, str::stream() << "db " << dbName << " appears corrupt pdfile version: " << h->version << " info: " << h->versionMinor << ' ' << h->fileLength);
+                    uasserted(14026, 
+                      str::stream() << "db " << dbName << " appears corrupt pdfile version: " << h->version 
+							        << " info: " << h->versionMinor << ' ' << h->fileLength);
                 }
 
                 log() << "****" << endl;
@@ -416,7 +418,7 @@ namespace mongo {
             l << "MongoDB starting : pid=" << pid << " port=" << cmdLine.port << " dbpath=" << dbpath;
             if( replSettings.master ) l << " master=" << replSettings.master;
             if( replSettings.slave )  l << " slave=" << (int) replSettings.slave;
-            l << ( is32bit ? " 32" : " 64" ) << "-bit " << endl;
+            l << ( is32bit ? " 32" : " 64" ) << "-bit host=" << getHostNameCached() << endl;
         }
         DEV log() << "_DEBUG build (which is slower)" << endl;
         show_warnings();
@@ -748,13 +750,10 @@ int main(int argc, char* argv[]) {
         if (params.count("repairpath")) {
             repairpath = params["repairpath"].as<string>();
             if (!repairpath.size()) {
-                out() << "repairpath has to be non-zero" << endl;
+                out() << "repairpath is empty" << endl;
                 dbexit( EXIT_BADOPTIONS );
             }
         }
-        else {
-            repairpath = dbpath;
-        }
         if (params.count("nocursors")) {
             useCursors = false;
         }
@@ -933,6 +932,10 @@ int main(int argc, char* argv[]) {
             dbexit( EXIT_BADOPTIONS );
         }
 
+        // needs to be after things like --configsvr parsing, thus here.
+        if( repairpath.empty() )
+            repairpath = dbpath;
+
         Module::configAll( params );
         dataFileSync.go();
 
diff --git a/db/dur_journal.cpp b/db/dur_journal.cpp
index 1738884006c..b8c259a6763 100644
--- a/db/dur_journal.cpp
+++ b/db/dur_journal.cpp
@@ -486,6 +486,11 @@ namespace mongo {
                 File f;
                 f.open(lsnPath().string().c_str());
                 assert(f.is_open());
+                if( f.len() == 0 ) { 
+                    // this could be 'normal' if we crashed at the right moment
+                    log() << "info lsn file is zero bytes long" << endl;
+                    return 0;
+                }
                 f.read(0,(char*)&L, sizeof(L));
                 unsigned long long lsn = L.get();
                 return lsn;
diff --git a/db/repl/rs.cpp b/db/repl/rs.cpp
index eea51e3f2e2..84ef3935a3b 100644
--- a/db/repl/rs.cpp
+++ b/db/repl/rs.cpp
@@ -465,9 +465,14 @@ namespace mongo {
         forgetPrimary();
 
         setSelfTo(0);
+
+        // For logging
+        string members = "";
+
         for( vector<ReplSetConfig::MemberCfg>::iterator i = _cfg->members.begin(); i != _cfg->members.end(); i++ ) {
             const ReplSetConfig::MemberCfg& m = *i;
             Member *mi;
+            members += ( members == "" ? "" : ", " ) + m.h.toString();
             if( m.h.isSelf() ) {
                 assert( _self == 0 );
                 mi = new Member(m.h, m._id, &m, true);
@@ -484,6 +489,11 @@ namespace mongo {
                     box.setOtherPrimary(mi);
             }
         }
+
+        if( ! _self ){
+            log() << "replSet warning did not detect own host in full reconfig, members " << members << " config: " << c << rsLog;
+        }
+
         return true;
     }
 
diff --git a/db/repl/rs_sync.cpp b/db/repl/rs_sync.cpp
index 815a224a79b..535796b5a25 100644
--- a/db/repl/rs_sync.cpp
+++ b/db/repl/rs_sync.cpp
@@ -34,7 +34,7 @@ namespace mongo {
         if ( *ns == '.' || *ns == 0 ) {
             if( *o.getStringField("op") == 'n' )
                 return;
-            log() << "replSet skipping bad op in oplog: " << o.toString() << endl;
+            log() << "replSet skipping bad op in oplog: " << o.toString() << rsLog;
             return;
         }
 
@@ -141,7 +141,7 @@ namespace mongo {
                     time_t now = time(0);
                     if (now - start > 10) {
                         // simple progress metering
-                        log() << "initialSyncOplogApplication applied " << n << " operations, synced to "
+                        log() << "replSet initialSyncOplogApplication applied " << n << " operations, synced to "
                               << ts.toStringPretty() << rsLog;
                         start = now;
                     }
@@ -322,8 +322,8 @@ namespace mongo {
             OpTime ts = o["ts"]._opTime();
             long long h = o["h"].numberLong();
             if( ts != lastOpTimeWritten || h != lastH ) {
-                log() << "replSet our last op time written: " << lastOpTimeWritten.toStringPretty() << endl;
-                log() << "replset source's GTE: " << ts.toStringPretty() << endl;
+                log() << "replSet our last op time written: " << lastOpTimeWritten.toStringPretty() << rsLog;
+                log() << "replset source's GTE: " << ts.toStringPretty() << rsLog;
                 syncRollback(r);
                 return;
             }
@@ -467,8 +467,13 @@ namespace mongo {
         */
 
         while( 1 ) {
-            if( myConfig().arbiterOnly )
+            // After a reconfig, we may not be in the replica set anymore, so
+            // check that we are in the set (and not an arbiter) before
+            // trying to sync with other replicas.
+            if( ! _self || myConfig().arbiterOnly ){
+            	if( ! _self ) log() << "replSet warning did not detect own host and port, not syncing, config: " << theReplSet->config() << rsLog;
                 return;
+            }
 
             try {
                 _syncThread();
@@ -489,9 +494,9 @@ namespace mongo {
                member has done a stepDown() and needs to come back up.
                */
             OCCASIONALLY {
-            	log() << "default rs heartbeat starting..." << endl;
+            	log() << "replSet default heartbeat starting..." << rsLog;
             	mgr->send( boost::bind(&Manager::msgCheckNewState, theReplSet->mgr) );
-            	log() << "rs heartbeat finished" << endl;
+            	log() << "replSet heartbeat finished" << rsLog;
             }
         }
     }
diff --git a/jstests/replsets/sync1.js b/jstests/replsets/sync1.js
index de824d0989b..5e862106efa 100644
--- a/jstests/replsets/sync1.js
+++ b/jstests/replsets/sync1.js
@@ -80,12 +80,17 @@ doTest = function (signal) {
 
     assert(count < 300);
     
+    // Need to be careful here, allocating datafiles for the slaves can take a *long* time on slow systems
+    sleep(7000);
+
     print("\nsync1.js ********************************************************************** part 6");
     dbs[0].getSisterDB("admin").runCommand({ replSetTest: 1, blind: true });
 
     print("\nsync1.js ********************************************************************** part 7");
 
     sleep(5000);
+    // If we start getting error hasNext: false with done alloc datafile msgs - may need to up the sleep again in part 5
+
 
     var max1;
     var max2;
diff --git a/jstests/sharding/migrateBig.js b/jstests/sharding/migrateBig.js
index f6ba18a7c02..917f152a4d2 100644
--- a/jstests/sharding/migrateBig.js
+++ b/jstests/sharding/migrateBig.js
@@ -40,6 +40,6 @@ for ( i=0; i<20; i+= 2 )
 
 db.printShardingStatus()
 
-assert.soon( function(){ var x = s.chunkDiff( "foo" , "test" ); print( "chunk diff: " + x ); return x < 2; } , "no balance happened" , 120 * 1000 , 2000 ) 
+assert.soon( function(){ var x = s.chunkDiff( "foo" , "test" ); print( "chunk diff: " + x ); return x < 2; } , "no balance happened" , 8 * 60 * 1000 , 2000 ) 
 
 s.stop()
diff --git a/jstests/sharding/shard3.js b/jstests/sharding/shard3.js
index 71325637baf..0a4bc2ff75f 100644
--- a/jstests/sharding/shard3.js
+++ b/jstests/sharding/shard3.js
@@ -4,9 +4,18 @@ s = new ShardingTest( "shard3" , 2 , 1 , 2 );
 
 s2 = s._mongos[1];
 
+db = s.getDB( "test" )
 s.adminCommand( { enablesharding : "test" } );
 s.adminCommand( { shardcollection : "test.foo" , key : { num : 1 } } );
 
+assert( sh.getBalancerState() , "A1" )
+sh.setBalancerState( false ) 
+assert( ! sh.getBalancerState() , "A2" )
+sh.setBalancerState( true ) 
+assert( sh.getBalancerState() , "A3" )
+sh.setBalancerState( false )
+assert( ! sh.getBalancerState() , "A4" )
+
 s.config.databases.find().forEach( printjson )
 
 a = s.getDB( "test" ).foo;
diff --git a/jstests/sharding/sync7.js b/jstests/sharding/sync7.js
index 0badd9ebeb7..a8ff0944a00 100644
--- a/jstests/sharding/sync7.js
+++ b/jstests/sharding/sync7.js
@@ -7,7 +7,7 @@ s._connections[1].getDB( "admin" ).runCommand( { _skewClockCommand : 1, skew : -
 
 // We need to start another mongos after skewing the clock, since the first mongos will have already
 // tested the config servers (via the balancer) before we manually skewed them
-otherMongos = startMongos( { port : 30020, v : 0, configdb : s._configDB, logpath : "/dev/null" } );
+otherMongos = startMongos( { port : 30020, v : 0, configdb : s._configDB } );
 
 // Initialize DB data
 initDB = function(name) {
diff --git a/jstests/slowNightly/sharding_balance1.js b/jstests/slowNightly/sharding_balance1.js
index 9379c4fa310..c50148c0400 100644
--- a/jstests/slowNightly/sharding_balance1.js
+++ b/jstests/slowNightly/sharding_balance1.js
@@ -41,7 +41,8 @@ print( diff() )
 assert.soon( function(){
     var d = diff();
     return d < 5;
-} , "balance didn't happen" , 1000 * 60 * 3 , 5000 );
+// Make sure there's enough time here, since balancing can sleep for 15s or so between balances.
+} , "balance didn't happen" , 1000 * 60 * 5 , 5000 );
     
 var chunkCount = sum();
 s.adminCommand( { removeshard: "shard0000" } );
diff --git a/s/balance.cpp b/s/balance.cpp
index 40d542b9111..fba325ed5ae 100644
--- a/s/balance.cpp
+++ b/s/balance.cpp
@@ -275,47 +275,51 @@ namespace mongo {
 
             try {
                 
-                // first make sure we should even be running
+                ScopedDbConnection conn( config );
+                
+                // ping has to be first so we keep things in the config server in sync
+                _ping( conn.conn() );
+
+                // now make sure we should even be running
                 if ( ! grid.shouldBalance() ) {
                     log(1) << "skipping balancing round because balancing is disabled" << endl;
+                    conn.done();
+                    
                     sleepsecs( 30 );
                     continue;
                 }
                 
-
-                ScopedDbConnection conn( config );
-
-                _ping( conn.conn() );
-                if ( ! _checkOIDs() ) {
-                    uassert( 13258 , "oids broken after resetting!" , _checkOIDs() );
-                }
+                uassert( 13258 , "oids broken after resetting!" , _checkOIDs() );
 
                 // use fresh shard state
                 Shard::reloadShardInfo();
 
-                dist_lock_try lk( &balanceLock , "doing balance round" );
-                if ( ! lk.got() ) {
-                    log(1) << "skipping balancing round because another balancer is active" << endl;
-                    conn.done();
-
-                    sleepsecs( 30 ); // no need to wake up soon
-                    continue;
-                }
-
-                log(1) << "*** start balancing round" << endl;
-
-                vector<CandidateChunkPtr> candidateChunks;
-                _doBalanceRound( conn.conn() , &candidateChunks );
-                if ( candidateChunks.size() == 0 ) {
-                    log(1) << "no need to move any chunk" << endl;
+                {
+                    dist_lock_try lk( &balanceLock , "doing balance round" );
+                    if ( ! lk.got() ) {
+                        log(1) << "skipping balancing round because another balancer is active" << endl;
+                        conn.done();
+                        
+                        sleepsecs( 30 ); // no need to wake up soon
+                        continue;
+                    }
+                    
+                    log(1) << "*** start balancing round" << endl;
+                    
+                    vector<CandidateChunkPtr> candidateChunks;
+                    _doBalanceRound( conn.conn() , &candidateChunks );
+                    if ( candidateChunks.size() == 0 ) {
+                        log(1) << "no need to move any chunk" << endl;
+                    }
+                    else {
+                        _balancedLastTime = _moveChunks( &candidateChunks );
+                    }
+                    
+                    log(1) << "*** end of balancing round" << endl;
                 }
-                else {
-                    _balancedLastTime = _moveChunks( &candidateChunks );
-                }
-
-                log(1) << "*** end of balancing round" << endl;
+                
                 conn.done();
-
+                    
                 sleepsecs( _balancedLastTime ? 5 : 10 );
             }
             catch ( std::exception& e ) {
diff --git a/s/config.cpp b/s/config.cpp
index c1410c9e33f..86b169afe40 100644
--- a/s/config.cpp
+++ b/s/config.cpp
@@ -456,9 +456,6 @@ namespace mongo {
     }
 
     bool ConfigServer::checkConfigServersConsistent( string& errmsg , int tries ) const {
-        if ( _config.size() == 1 )
-            return true;
-
         if ( tries <= 0 )
             return false;
 
@@ -485,6 +482,9 @@ namespace mongo {
             res.push_back(x);
         }
 
+        if ( _config.size() == 1 )
+            return true;
+
         if ( up == 0 ) {
             errmsg = "no config servers reachable";
             return false;
diff --git a/s/d_migrate.cpp b/s/d_migrate.cpp
index fe4c0efe355..22b600a8db1 100644
--- a/s/d_migrate.cpp
+++ b/s/d_migrate.cpp
@@ -1008,6 +1008,7 @@ namespace mongo {
                     conn.done();
                 }
                 catch ( DBException& e ) {
+                    warning() << e << endl;
                     ok = false;
                     BSONObjBuilder b;
                     e.getInfo().append( b );
diff --git a/s/strategy_shard.cpp b/s/strategy_shard.cpp
index 4e0c76312cd..2b36025d4a2 100644
--- a/s/strategy_shard.cpp
+++ b/s/strategy_shard.cpp
@@ -166,8 +166,8 @@ namespace mongo {
                         gotThrough = true;
                         break;
                     }
-                    catch ( StaleConfigException& ) {
-                        log( i < ( maxTries / 2 ) ) << "retrying insert because of StaleConfigException: " << o << endl;
+                    catch ( StaleConfigException& e ) {
+                        log( i < ( maxTries / 2 ) ) << "retrying insert because of StaleConfigException: " << e << " object: " << o << endl;
                         r.reset();
                         manager = r.getChunkManager();
                     }
diff --git a/shell/mongo_vstudio.cpp b/shell/mongo_vstudio.cpp
index 175ac004e00..01fa3ad54ea 100644
--- a/shell/mongo_vstudio.cpp
+++ b/shell/mongo_vstudio.cpp
@@ -1534,7 +1534,10 @@ const StringData _jscode_raw_utils_sh =
 "print( \"\\tsh.enableSharding(dbname)                 enables sharding on the database dbname\" )\n" 
 "print( \"\\tsh.shardCollection(fullName,key,unique)   shards the collection\" );\n" 
 "print( \"\\tsh.splitFind(fullName,find)               splits the chunk that find is in at the median\" );\n" 
-"print( \"\\tsh.splitAt(fullName,middle)           splits the chunk that middle is in at middle\" );\n" 
+"print( \"\\tsh.splitAt(fullName,middle)               splits the chunk that middle is in at middle\" );\n" 
+"\n" 
+"print( \"\\tsh.setBalancerState( <bool on or not> )   turns the balancer on or off true=on, false=off\" );\n" 
+"print( \"\\tsh.getBalancerState()   return true if on, off if not\" );\n" 
 "\n" 
 "print( \"\\tsh.status()                               prints a general overview of the cluster\" )\n" 
 "}\n" 
@@ -1575,6 +1578,17 @@ const StringData _jscode_raw_utils_sh =
 "sh._checkFullName( fullName )\n" 
 "sh._adminCommand( { split : fullName , middle : middle } )\n" 
 "}\n" 
+"\n" 
+"sh.setBalancerState = function( onOrNot ) {\n" 
+"db.getSisterDB( \"config\" ).settings.update({ _id: \"balancer\" }, { $set : { stopped: onOrNot ? false : true } }, true );\n" 
+"}\n" 
+"\n" 
+"sh.getBalancerState = function() {\n" 
+"var x = db.getSisterDB( \"config\" ).settings.findOne({ _id: \"balancer\" } )\n" 
+"if ( x == null )\n" 
+"return true;\n" 
+"return ! x.stopped;\n" 
+"}\n" 
 ;
 extern const JSFile utils_sh;
 const JSFile utils_sh = { "shell/utils_sh.js" , _jscode_raw_utils_sh };
diff --git a/shell/utils_sh.js b/shell/utils_sh.js
index 8644fcbe6ca..b66929ced05 100644
--- a/shell/utils_sh.js
+++ b/shell/utils_sh.js
@@ -28,7 +28,10 @@ sh.help = function() {
     print( "\tsh.enableSharding(dbname)                 enables sharding on the database dbname" )
     print( "\tsh.shardCollection(fullName,key,unique)   shards the collection" );
     print( "\tsh.splitFind(fullName,find)               splits the chunk that find is in at the median" );
-    print( "\tsh.splitAt(fullName,middle)           splits the chunk that middle is in at middle" );
+    print( "\tsh.splitAt(fullName,middle)               splits the chunk that middle is in at middle" );
+
+    print( "\tsh.setBalancerState( <bool on or not> )   turns the balancer on or off true=on, false=off" );
+    print( "\tsh.getBalancerState()   return true if on, off if not" );
     
     print( "\tsh.status()                               prints a general overview of the cluster" )
 }
@@ -69,3 +72,14 @@ sh.splitAt = function( fullName , middle ) {
     sh._checkFullName( fullName )
     sh._adminCommand( { split : fullName , middle : middle } )
 }
+
+sh.setBalancerState = function( onOrNot ) { 
+    db.getSisterDB( "config" ).settings.update({ _id: "balancer" }, { $set : { stopped: onOrNot ? false : true } }, true );
+}
+
+sh.getBalancerState = function() {
+    var x = db.getSisterDB( "config" ).settings.findOne({ _id: "balancer" } )
+    if ( x == null )
+        return true;
+    return ! x.stopped;
+}
diff --git a/tools/sniffer.cpp b/tools/sniffer.cpp
index 4e9cfd21bf3..024a37475cc 100644
--- a/tools/sniffer.cpp
+++ b/tools/sniffer.cpp
@@ -104,6 +104,9 @@ struct sniff_ip {
 #define IP_V(ip)                (((ip)->ip_vhl) >> 4)
 
 /* TCP header */
+#ifdef _WIN32
+typedef unsigned __int32 uint32_t;
+#endif
 typedef uint32_t tcp_seq;
 
 struct sniff_tcp {
diff --git a/util/file.h b/util/file.h
index aed5d52fdaa..704dc536211 100644
--- a/util/file.h
+++ b/util/file.h
@@ -38,6 +38,8 @@ namespace mongo {
     typedef boost::uint64_t fileofs;
 #endif
 
+    /* NOTE: not thread-safe. (at least the windows implementation isn't. */
+
     class FileInterface {
     public:
         void open(const char *fn) {}
@@ -183,7 +185,11 @@ namespace mongo {
         bool bad() { return _bad; }
         bool is_open() { return fd > 0; }
         fileofs len() {
-            return lseek(fd, 0, SEEK_END);
+            off_t o = lseek(fd, 0, SEEK_END);
+            if( o != (off_t) -1 )
+                return o;
+            err(false);
+            return 0;
         }
         void fsync() { ::fsync(fd); }
         static boost::intmax_t freeSpace ( const string &path ) {
diff --git a/util/message.cpp b/util/message.cpp
index c58d0ddca63..892d64c7bdb 100644
--- a/util/message.cpp
+++ b/util/message.cpp
@@ -364,7 +364,7 @@ namespace mongo {
         ConnectBG(int sock, SockAddr farEnd) : _sock(sock), _farEnd(farEnd) { }
 
         void run() { _res = ::connect(_sock, _farEnd.raw(), _farEnd.addressSize); }
-        string name() const { return ""; /* too short lived to need to name */ }
+        string name() const { return "ConnectBG"; }
         int inError() const { return _res; }
 
     private:
@@ -634,12 +634,20 @@ again:
         unsigned retries = 0;
         while( len > 0 ) {
             int ret = ::recv( sock , buf , len , portRecvFlags );
-            if ( ret == 0 ) {
+            if ( ret > 0 ) {
+                if ( len <= 4 && ret != len )
+                    log(_logLevel) << "MessagingPort recv() got " << ret << " bytes wanted len=" << len << endl;
+                assert( ret <= len );
+                len -= ret;
+                buf += ret;
+            }
+            else if ( ret == 0 ) {
                 log(3) << "MessagingPort recv() conn closed? " << farEnd.toString() << endl;
                 throw SocketException( SocketException::CLOSED );
             }
-            if ( ret < 0 ) {
+            else { /* ret < 0  */
                 int e = errno;
+                
 #if defined(EINTR) && !defined(_WIN32)
                 if( e == EINTR ) {
                     if( ++retries == 1 ) {
@@ -648,29 +656,18 @@ again:
                     }
                 }
 #endif
-                if ( e != EAGAIN || _timeout == 0 ) {
-                    SocketException::Type t = SocketException::RECV_ERROR;
-#if defined(_WINDOWS)
-                    if( e == WSAETIMEDOUT ) t = SocketException::RECV_TIMEOUT;
-#else
-                    /* todo: what is the error code on an SO_RCVTIMEO on linux? EGAIN? EWOULDBLOCK? */
+                if ( ( e == EAGAIN 
+#ifdef _WINDOWS
+                       || e == WSAETIMEDOUT
 #endif
-                    log(_logLevel) << "MessagingPort recv() " << errnoWithDescription(e) << " " << farEnd.toString() <<endl;
-                    throw SocketException(t);
-                }
-                else {
-                    if ( !serverAlive( farEnd.toString() ) ) {
-                        log(_logLevel) << "MessagingPort recv() remote dead " << farEnd.toString() << endl;
-                        throw SocketException( SocketException::RECV_ERROR );
-                    }
+                       ) && _timeout > 0 ) {
+                    // this is a timeout
+                    log(_logLevel) << "MessagingPort recv() timeout  " << farEnd.toString() <<endl;
+                    throw SocketException(SocketException::RECV_TIMEOUT);                    
                 }
-            }
-            else {
-                if ( len <= 4 && ret != len )
-                    log(_logLevel) << "MessagingPort recv() got " << ret << " bytes wanted len=" << len << endl;
-                assert( ret <= len );
-                len -= ret;
-                buf += ret;
+
+                log(_logLevel) << "MessagingPort recv() " << errnoWithDescription(e) << " " << farEnd.toString() <<endl;
+                throw SocketException(SocketException::RECV_ERROR);
             }
         }
     }
diff --git a/util/version.cpp b/util/version.cpp
index 7919b0be790..f763c7f4119 100644
--- a/util/version.cpp
+++ b/util/version.cpp
@@ -110,7 +110,7 @@ namespace mongo {
 #endif
 
     void printSysInfo() {
-        log() << "build sys info: " << sysInfo() << endl;
+        log() << "build info: " << sysInfo() << endl;
     }
 
     //