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

20 files changed, 151 insertions, 26 deletions

diff --git a/client/dbclient.h b/client/dbclient.h
index 76c5d3014b8..5da4f84af22 100644
--- a/client/dbclient.h
+++ b/client/dbclient.h
@@ -847,6 +847,7 @@ namespace mongo {
         }
 
         /** uses QueryOption_Exhaust 
+            use DBClientCursorBatchIterator if you want to do items in large blocks, perhpas to avoid granular locking and such.
          */
         unsigned long long query( boost::function<void(const BSONObj&)> f, const string& ns, Query query, const BSONObj *fieldsToReturn = 0, int queryOptions = 0);
         unsigned long long query( boost::function<void(DBClientCursorBatchIterator&)> f, const string& ns, Query query, const BSONObj *fieldsToReturn = 0, int queryOptions = 0);
diff --git a/client/parallel.cpp b/client/parallel.cpp
index d605ebd857e..a7568b2d83d 100644
--- a/client/parallel.cpp
+++ b/client/parallel.cpp
@@ -87,6 +87,8 @@ namespace mongo {
         
         auto_ptr<DBClientCursor> cursor = 
             conn->query( _ns , q , num , 0 , ( _fields.isEmpty() ? 0 : &_fields ) , _options , _batchSize == 0 ? 0 : _batchSize + skipLeft );
+
+        assert( cursor.get() );
         
         if ( cursor->hasResultFlag( ResultFlag_ShardConfigStale ) ){
             conn.done();
diff --git a/db/client.cpp b/db/client.cpp
index af83653c5f2..65c467a2dc0 100644
--- a/db/client.cpp
+++ b/db/client.cpp
@@ -387,7 +387,13 @@ namespace mongo {
                     
                     tablecell( ss , co.opNum() );
                     tablecell( ss , co.active() );
-                    tablecell( ss , co.getLockType() );
+                    {
+                        int lt = co.getLockType();
+                        if( lt == -1 ) tablecell(ss, "R");
+                        else if( lt == 1 ) tablecell(ss, "W");
+                        else
+                            tablecell( ss ,  lt);
+                    }
                     tablecell( ss , co.isWaitingForLock() );
                     if ( co.active() )
                         tablecell( ss , co.elapsedSeconds() );
diff --git a/db/cloner.cpp b/db/cloner.cpp
index 79d0c082e87..96890bf1417 100644
--- a/db/cloner.cpp
+++ b/db/cloner.cpp
@@ -172,9 +172,10 @@ namespace mongo {
             f.context = r._context;
             DBClientConnection *remote = dynamic_cast< DBClientConnection* >( conn.get() );
             if ( remote ) {
-                remote->query( boost::function<void(DBClientCursorBatchIterator &)>( f ), from_collection, query, 0, options );                
+                remote->query( boost::function<void(DBClientCursorBatchIterator &)>( f ), from_collection, query, 0, options );
             } else { // no exhaust mode for direct client, so we have this hack
                 auto_ptr<DBClientCursor> c = conn->query( from_collection, query, 0, 0, 0, options );
+                assert( c.get() );
                 while( c->more() ) {
                     DBClientCursorBatchIterator i( *c );
                     f( i );
diff --git a/db/dbcommands_generic.cpp b/db/dbcommands_generic.cpp
index 338dbd0f6ee..2d16c9f232b 100644
--- a/db/dbcommands_generic.cpp
+++ b/db/dbcommands_generic.cpp
@@ -67,7 +67,7 @@ namespace mongo {
     /* for diagnostic / testing purposes. */
     class CmdSleep : public Command { 
     public:
-        virtual LockType locktype() const { return READ; } 
+        virtual LockType locktype() const { return NONE; } 
         virtual bool adminOnly() const { return true; }
         virtual bool logTheOp() {
             return false;
@@ -76,11 +76,19 @@ namespace mongo {
             return true;
         }
         virtual void help( stringstream& help ) const {
-            help << "internal testing command.  Makes db block (in a read lock) for 100 seconds";
+            help << "internal testing command.  Makes db block (in a read lock) for 100 seconds\n";
+            help << "w:true write lock";
         }
-        CmdSleep() : Command("sleep") {}
+        CmdSleep() : Command("sleep") { }
         bool run(const string& ns, BSONObj& cmdObj, string& errmsg, BSONObjBuilder& result, bool fromRepl) {
-            sleepsecs(100);
+            if( cmdObj.getBoolField("w") ) { 
+                writelock lk("");
+                sleepsecs(100);
+            }
+            else {
+                readlock lk("");
+                sleepsecs(100);
+            }
             return true;
         }
     } cmdSleep;
diff --git a/db/repl/health.cpp b/db/repl/health.cpp
index 5c233a6a7e7..35da9edcdc6 100644
--- a/db/repl/health.cpp
+++ b/db/repl/health.cpp
@@ -28,6 +28,7 @@
 #include "../helpers/dblogger.h"
 #include "connections.h"
 #include "../../util/unittest.h"
+#include "../dbhelpers.h"
 
 namespace mongo {
     /* decls for connections.h */
@@ -63,7 +64,7 @@ namespace mongo {
         return s.str();
     }
 
-    void Member::summarizeAsHtml(stringstream& s) const { 
+    void Member::summarizeMember(stringstream& s) const { 
         s << tr();
         {
             stringstream u;
@@ -184,6 +185,10 @@ namespace mongo {
         ScopedConn conn(m->fullName());        
 
         auto_ptr<DBClientCursor> c = conn->query(rsoplog, Query().sort("$natural",1), 20, 0, &fields);
+        if( c.get() == 0 ) { 
+            ss << "couldn't query " << rsoplog;
+            return;
+        }
         static const char *h[] = {"ts","optime", "h","op","ns","rest",0};
 
         ss << "<style type=\"text/css\" media=\"screen\">"
@@ -211,6 +216,10 @@ namespace mongo {
         }
         else { 
             auto_ptr<DBClientCursor> c = conn->query(rsoplog, Query().sort("$natural",-1), 20, 0, &fields);
+            if( c.get() == 0 ) { 
+                ss << "couldn't query [2] " << rsoplog;
+                return;
+            }
             string x;
             bo o = c->next();
             otEnd = o["ts"]._opTime();
@@ -290,7 +299,7 @@ namespace mongo {
         Member *m = head();
         while( m ) {
 			stringstream s;
-            m->summarizeAsHtml(s);
+            m->summarizeMember(s);
 			mp[m->hbinfo().id()] = s.str();
             m = m->next();
         }
@@ -298,6 +307,20 @@ namespace mongo {
         for( map<int,string>::const_iterator i = mp.begin(); i != mp.end(); i++ )
             s << i->second;
         s << _table();
+
+        try {
+            readlocktry lk("local.replset.minvalid", 300);
+            if( lk.got() ) {
+                BSONObj mv;
+                if( Helpers::getSingleton("local.replset.minvalid", mv) ) { 
+                    s << p( str::stream() << "minvalid: " << mv["ts"]._opTime().toString() );
+                }
+            }
+            else s << p(".");
+        }
+        catch(...) { 
+            s << p("exception fetching minvalid?");
+        }
     }
 
 
diff --git a/db/repl/manager.cpp b/db/repl/manager.cpp
index e372ec4a6f8..82a0f01e1ef 100644
--- a/db/repl/manager.cpp
+++ b/db/repl/manager.cpp
@@ -48,7 +48,7 @@ namespace mongo {
     task::Server("rs Manager"), rs(_rs), busyWithElectSelf(false), _primary(NOPRIMARY)
     { 
     }
-
+    
     Manager::~Manager() { 
         log() << "should never be called?" << rsLog;
         rs->mgr = 0;
diff --git a/db/repl/rs.h b/db/repl/rs.h
index a6924fc6e23..09b766ccf2f 100644
--- a/db/repl/rs.h
+++ b/db/repl/rs.h
@@ -52,7 +52,7 @@ namespace mongo {
         unsigned id() const { return _hbinfo.id(); }
         bool potentiallyHot() const { return _config->potentiallyHot(); } // not arbiter, not priority 0
 
-        void summarizeAsHtml(stringstream& s) const;
+        void summarizeMember(stringstream& s) const;
         friend class ReplSetImpl;
     private:
         const ReplSetConfig::MemberCfg *_config; /* todo: when this changes??? */
diff --git a/db/repl/rs_initiate.cpp b/db/repl/rs_initiate.cpp
index 19a345bf4d1..70a1d2de5f8 100644
--- a/db/repl/rs_initiate.cpp
+++ b/db/repl/rs_initiate.cpp
@@ -166,7 +166,7 @@ namespace mongo {
                 bob members;
                 members.append("0", BSON( "_id" << 0 << "host" << HostAndPort::Me().toString() ));
                 for( unsigned i = 0; i < seeds.size(); i++ )
-                    members.append(bob::numStr(i), BSON( "_id" << i << "host" << seeds[i].toString()));
+                    members.append(bob::numStr(i+1), BSON( "_id" << i+1 << "host" << seeds[i].toString()));
                 b.appendArray("members", members.obj());
                 configObj = b.obj();
                 log() << "replSet created this configuration for initiation : " << configObj.toString() << rsLog;
diff --git a/db/repl/rs_rollback.cpp b/db/repl/rs_rollback.cpp
index 42814101f76..dc0b89302d1 100644
--- a/db/repl/rs_rollback.cpp
+++ b/db/repl/rs_rollback.cpp
@@ -140,10 +140,10 @@ namespace mongo {
 
         //auto_ptr<DBClientCursor> u = us->query(rsoplog, q, 0, 0, &fields, 0, 0);
 
-	h.rbid = getRBID(them);
+        h.rbid = getRBID(them);
         auto_ptr<DBClientCursor> t = them->query(rsoplog, q, 0, 0, &fields, 0, 0);
 
-        if( !t->more() ) throw "remote oplog empty or unreadable";
+        if( t.get() == 0 || !t->more() ) throw "remote oplog empty or unreadable";
 
         BSONObj ourObj = u.current();
         OpTime ourTime = ourObj["ts"]._opTime();
diff --git a/jstests/replsets/replset1.js b/jstests/replsets/replset1.js
index bc6c20e874a..61f3d99d549 100644
--- a/jstests/replsets/replset1.js
+++ b/jstests/replsets/replset1.js
@@ -59,7 +59,6 @@ doTest = function( signal ) {
 
     assert( master_id != new_master_id, "Old master shouldn't be equal to new master." );
 
-    
     { 
         // this may fail since it has to reconnect
         try {
@@ -74,8 +73,44 @@ doTest = function( signal ) {
     // Here's how to restart a node:
     replTest.restart( master_id );
 
+    // Now let's write some documents to the new master
+    for(var i=0; i<1000; i++) {
+        new_master.getDB("bar").bar.save({a: i});
+    }
+    new_master.getDB("admin").runCommand({getlasterror: 1});
+
+    // Here's how to restart the old master node:
+    slave = replTest.restart( master_id );
+
+    // Now, let's make sure that the old master comes up as a slave
+    assert.soon(function() {
+        var res = slave.getDB("admin").runCommand({ismaster: 1});
+        printjson(res);
+        return res['ok'] == 1 && res['ismaster'] == false;
+    });
+
+    // And we need to make sure that the replset comes back up
+    assert.soon(function() {
+        var res = new_master.getDB("admin").runCommand({replSetGetStatus: 1});
+        printjson( res );
+        return res.myState == 1;
+    });
+
+    // And that both slave nodes have all the updates
+    new_master = replTest.getMaster();
+    replTest.awaitReplication();
+
+    slaves = replTest.liveNodes.slaves;
+    assert( slaves.length == 2, "Expected 2 slaves but length was " + slaves.length );
+    slaves.forEach(function(slave) {
+        slave.setSlaveOk();
+        var count = slave.getDB("bar").runCommand({count: "bar"});
+        printjson( count );
+        assert( count.n == 1000 , slave + " expected 1000 but count was " + count.n);
+    });
+
     // Shut down the set and finish the test.
     replTest.stopSet( signal );
 }
 
-doTest( 15 );
+// doTest( 15 );
diff --git a/jstests/replsets/replset2.js b/jstests/replsets/replset2.js
index da39662221e..a5f4c7f75f5 100644
--- a/jstests/replsets/replset2.js
+++ b/jstests/replsets/replset2.js
@@ -33,16 +33,40 @@ doTest = function( signal ) {
 
     var testDB = "repl-test";
 
+    var failed = false;
     var callGetLastError = function(w, timeout, db) {
         var result = master.getDB(db).runCommand({getlasterror: 1, w: w, wtimeout: timeout});
         printjson( result );
-        assert( result['ok'] == 1, "getLastError with w=" + w + " failed");
+        if(result['ok'] != 1) {
+          print("FAILURE");
+          failed = true;
+        }
     }
 
-    // Test getlasterror with a simple insert
+    // Test getlasterror with multiple inserts
     // TEST FAILS HERE
+    print("**** Try inserting a multiple records -- first insert ****")
     master.getDB(testDB).foo.insert({n: 1});
-    callGetLastError(3, 60000, testDB);
+    master.getDB(testDB).foo.insert({n: 2});
+    master.getDB(testDB).foo.insert({n: 3});
+    callGetLastError(3, 10000, testDB);
+
+    m1 = master.getDB(testDB).foo.findOne({n: 1});
+    printjson( m1 );
+    assert( m1['n'] == 1 , "Failed to save to master on multiple inserts");
+
+    var s0 = slaves[0].getDB(testDB).foo.findOne({n: 1});
+    assert( s0['n'] == 1 , "Failed to replicate to slave 0 on multiple inserts");
+
+    var s1 = slaves[1].getDB(testDB).foo.findOne({n: 1});
+    assert( s1['n'] == 1 , "Failed to replicate to slave 1 on multiple inserts");
+
+
+    // Test getlasterror with a simple insert
+    print("**** Try inserting a single record ****")
+    master.getDB(testDB).dropDatabase();
+    master.getDB(testDB).foo.insert({n: 1});
+    callGetLastError(3, 10000, testDB);
 
     m1 = master.getDB(testDB).foo.findOne({n: 1});
     printjson( m1 );
@@ -74,6 +98,8 @@ doTest = function( signal ) {
     verifyReplication("slave 0", slaves[0].getDB(testDB).baz);
     verifyReplication("slave 1", slaves[1].getDB(testDB).baz);
 
+    assert( failed == false, "Replication with getLastError failed. See errors." );
+
     replTest.stopSet( signal );
 }
 
diff --git a/s/chunk.cpp b/s/chunk.cpp
index cee7b0ea6cc..870e7aaa916 100644
--- a/s/chunk.cpp
+++ b/s/chunk.cpp
@@ -451,8 +451,11 @@ namespace mongo {
         
         // not using regular count as this is more flexible and supports $min/$max
         Query q = Query().minKey(_min).maxKey(_max);
-        int n = conn->query(_manager->getns(), q, maxCount, 0, &fields)->itcount();
-        
+        int n;
+        {
+            auto_ptr<DBClientCursor> c = conn->query(_manager->getns(), q, maxCount, 0, &fields);
+            n = c->itcount();
+        }        
         conn.done();
         return n;
     }
diff --git a/s/config.cpp b/s/config.cpp
index e6bb48870c4..50682c67cb2 100644
--- a/s/config.cpp
+++ b/s/config.cpp
@@ -228,6 +228,7 @@ namespace mongo {
         
 
         auto_ptr<DBClientCursor> cursor = conn->query( ShardNS::collection ,b.obj() );
+        assert( cursor.get() );
         while ( cursor->more() ){
             BSONObj o = cursor->next();
             _collections[o["_id"].String()] = CollectionInfo( this , o );
@@ -488,6 +489,7 @@ namespace mongo {
         
         ScopedDbConnection conn( _primary );
         auto_ptr<DBClientCursor> c = conn->query( ShardNS::settings , BSONObj() );
+        assert( c.get() );
         while ( c->more() ){
             BSONObj o = c->next();
             string name = o["_id"].valuestrsafe();
diff --git a/s/d_migrate.cpp b/s/d_migrate.cpp
index 12b6ee3ae0c..f027f8bc611 100644
--- a/s/d_migrate.cpp
+++ b/s/d_migrate.cpp
@@ -788,6 +788,7 @@ namespace mongo {
             { // 3. initial bulk clone
                 state = CLONE;
                 auto_ptr<DBClientCursor> cursor = conn->query( ns , Query().minKey( min ).maxKey( max ) , /* QueryOption_Exhaust */ 0 );
+                assert( cursor.get() );
                 while ( cursor->more() ){
                     BSONObj o = cursor->next();
                     {
diff --git a/s/d_state.cpp b/s/d_state.cpp
index 26e44a1fcdd..dd2fecef45d 100644
--- a/s/d_state.cpp
+++ b/s/d_state.cpp
@@ -184,6 +184,7 @@ namespace mongo {
         }
 
         auto_ptr<DBClientCursor> cursor = conn->query( "config.chunks" , Query(q).sort( "min" ) );
+        assert( cursor.get() );
         if ( ! cursor->more() ){
             if ( scoped.get() )
                 scoped->done();
diff --git a/s/shard.cpp b/s/shard.cpp
index 7879cc031dc..4ef68c0103c 100644
--- a/s/shard.cpp
+++ b/s/shard.cpp
@@ -33,6 +33,7 @@ namespace mongo {
             {
                 ScopedDbConnection conn( configServer.getPrimary() );
                 auto_ptr<DBClientCursor> c = conn->query( ShardNS::shard , Query() );
+                assert( c.get() );
                 while ( c->more() ){
                     all.push_back( c->next().getOwned() );
                 }
diff --git a/scripting/engine.cpp b/scripting/engine.cpp
index 7b47a0fb2d1..9e20a3a031c 100644
--- a/scripting/engine.cpp
+++ b/scripting/engine.cpp
@@ -170,6 +170,7 @@ namespace mongo {
         
         static DBClientBase * db = createDirectClient();
         auto_ptr<DBClientCursor> c = db->query( coll , Query() );
+        assert( c.get() );
         
         set<string> thisTime;
         
diff --git a/shell/servers.js b/shell/servers.js
index a4651353f92..08e8bfd6126 100644
--- a/shell/servers.js
+++ b/shell/servers.js
@@ -1199,6 +1199,7 @@ ReplSetTest.prototype.awaitReplication = function() {
    this.getMaster();
 
    latest = this.liveNodes.master.getDB("local")['oplog.rs'].find({}).sort({'$natural': -1}).limit(1).next()['ts']['t']
+   print(latest);
 
    this.attempt({context: this, timeout: 30000, desc: "awaiting replication"},
        function() {
@@ -1214,9 +1215,14 @@ ReplSetTest.prototype.awaitReplication = function() {
              }
 
              slave.getDB("admin").getMongo().setSlaveOk();
-             var log = slave.getDB("local")['replset.minvalid'];
-             if(log.find().hasNext()) {
-               synced == synced && log.find().next()['ts']['t'];
+             var log = slave.getDB("local")['oplog.rs'];
+             if(log.find({}).sort({'$natural': -1}).limit(1).hasNext()) {
+               var entry = log.find({}).sort({'$natural': -1}).limit(1).next();
+               printjson( entry );
+               var ts = entry['ts']['t'];
+               print("TS for " + slave + " is " + ts + " and latest is " + latest);
+               print("Oplog size for " + slave + " is " + log.count());
+               synced = (synced && (latest == ts));
              }
              else {
                synced = false;
@@ -1245,8 +1251,11 @@ ReplSetTest.prototype.start = function( n , options , restart ){
 
     print("Starting....");
     print( o );
-    if ( restart )
-        return startMongoProgram.apply( null , o );
+    if ( restart ) {
+        this.nodes[n] = startMongoProgram.apply( null , o );
+        printjson(this.nodes);
+        return this.nodes[n];
+    }
     else {
         return startMongod.apply( null , o );
     }
diff --git a/util/concurrency/task.cpp b/util/concurrency/task.cpp
index 7e410c81174..6102666654e 100644
--- a/util/concurrency/task.cpp
+++ b/util/concurrency/task.cpp
@@ -120,13 +120,16 @@ namespace mongo {
             rq = false;
             while( 1 ) { 
                 lam f;
-                {
+                try {
                     boost::mutex::scoped_lock lk(m);
                     while( d.empty() )
                         c.wait(lk);
                     f = d.front();
                     d.pop_front();
                 }
+                catch(...) { 
+                    log() << "ERROR exception in Server:doWork?" << endl;
+                }
                 try {
                     f();
                     if( rq ) {
@@ -135,6 +138,8 @@ namespace mongo {
                     }
                 } catch(std::exception& e) { 
                     log() << "Server::doWork() exception " << e.what() << endl;
+                } catch(...) {
+                    log() << "Server::doWork() unknown exception!" << endl;
                 }
             }
         }