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

mongodb · Apr 28, 2011 · eb964f1 · eb964f1
2 parents 87ab46c + ef042e4
commit eb964f1
Show file tree

Hide file tree

Showing 38 changed files with 225 additions and 81 deletions.
diff --git a/buildscripts/smoke.py b/buildscripts/smoke.py
@@ -145,7 +145,7 @@ def start(self):
         utils.ensureDir(dir_name)
         argv = [mongod_executable, "--port", str(self.port), "--dbpath", dir_name]
         if self.kwargs.get('small_oplog'):
-            argv += ["--master", "--oplogSize", "128"]
+            argv += ["--master", "--oplogSize", "256"]
         if self.slave:
             argv += ['--slave', '--source', 'localhost:' + str(srcport)]
         print "running " + " ".join(argv)

diff --git 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
@@ -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;
     }
 
     /**
@@ -309,7 +309,7 @@ namespace mongo {
             BSONObj o;
             c->isMaster(isMaster, &o);
 
-            log( ! verbose ) << "ReplicaSetMonitor::_checkConnection: " << c->toString() << ' ' << o << '\n';
+            log( ! verbose ) << "ReplicaSetMonitor::_checkConnection: " << c->toString() << ' ' << o << endl;
 
             // add other nodes
             string maybePrimary;
@@ -530,12 +530,12 @@ namespace mongo {
             // we're ok sending to a slave
             // we'll try 2 slaves before just using master
             // checkSlave will try a different slave automatically after a failure
-            for ( int i=0; i<2; i++ ) {
+            for ( int i=0; i<3; i++ ) {
                 try {
                     return checkSlave()->query(ns,query,nToReturn,nToSkip,fieldsToReturn,queryOptions,batchSize);
                 }
-                catch ( DBException & ) {
-                    LOG(1) << "can't query replica set slave: " << _slaveHost << endl;
+                catch ( DBException &e ) {
+                    LOG(1) << "can't query replica set slave " << i << " : " << _slaveHost << causedBy( e ) << endl;
                 }
             }
         }
@@ -548,12 +548,12 @@ namespace mongo {
             // we're ok sending to a slave
             // we'll try 2 slaves before just using master
             // checkSlave will try a different slave automatically after a failure
-            for ( int i=0; i<2; i++ ) {
+            for ( int i=0; i<3; i++ ) {
                 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;
                 }
             }
         }
@@ -581,12 +581,12 @@ namespace mongo {
             DbMessage dm( toSend );
             QueryMessage qm( dm );
             if ( qm.queryOptions & QueryOption_SlaveOk ) {
-                for ( int i=0; i<2; i++ ) {
+                for ( int i=0; i<3; i++ ) {
                     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;
                     }
                 }
             }
@@ -601,15 +601,15 @@ namespace mongo {
             DbMessage dm( toSend );
             QueryMessage qm( dm );
             if ( qm.queryOptions & QueryOption_SlaveOk ) {
-                for ( int i=0; i<2; i++ ) {
+                for ( int i=0; i<3; i++ ) {
                     try {
                         DBClientConnection* s = checkSlave();
                         if ( actualServer )
                             *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
@@ -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
@@ -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.vcxproj b/db/db.vcxproj
@@ -493,6 +493,10 @@
     <ClCompile Include="mongommf.cpp" />
     <ClCompile Include="oplog.cpp" />
     <ClCompile Include="projection.cpp" />
+    <ClCompile Include="querypattern.cpp">
+      <PrecompiledHeader Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">NotUsing</PrecompiledHeader>
+      <PrecompiledHeader Condition="'$(Configuration)|$(Platform)'=='Release|x64'">NotUsing</PrecompiledHeader>
+    </ClCompile>
     <ClCompile Include="repl.cpp" />
     <ClCompile Include="repl\consensus.cpp" />
     <ClCompile Include="repl\heartbeat.cpp" />
@@ -658,6 +662,7 @@
     <ClInclude Include="namespace-inl.h" />
     <ClInclude Include="oplogreader.h" />
     <ClInclude Include="projection.h" />
+    <ClInclude Include="queryutil.h" />
     <ClInclude Include="repl.h" />
     <ClInclude Include="replpair.h" />
     <ClInclude Include="repl\connections.h" />

diff --git a/db/db.vcxproj.filters b/db/db.vcxproj.filters
@@ -155,6 +155,7 @@
     <ClCompile Include="repl\rs_config.cpp" />
     <ClCompile Include="security_key.cpp" />
     <ClCompile Include="..\util\file_allocator.cpp" />
+    <ClCompile Include="querypattern.cpp" />
   </ItemGroup>
   <ItemGroup>
     <ClInclude Include="..\client\dbclientcursor.h" />
@@ -293,6 +294,7 @@
     <ClInclude Include="dur_journalimpl.h" />
     <ClInclude Include="..\util\concurrency\race.h" />
     <ClInclude Include="..\util\alignedbuilder.h" />
+    <ClInclude Include="queryutil.h" />
   </ItemGroup>
   <ItemGroup>
     <ResourceCompile Include="db.rc" />

diff --git a/db/dur_journal.cpp b/db/dur_journal.cpp
@@ -457,7 +457,7 @@ namespace mongo {
             if something highly surprising, throws to abort
         */
         unsigned long long LSNFile::get() {
-            uassert(13614, "unexpected version number of lsn file in journal/ directory", ver == 0);
+            uassert(13614, str::stream() << "unexpected version number of lsn file in journal/ directory got: " << ver , ver == 0);
             if( ~lsn != checkbytes ) {
                 log() << "lsnfile not valid. recovery will be from log start. lsn: " << hex << lsn << " checkbytes: " << hex << checkbytes << endl;
                 return 0;
@@ -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/queryutil.h b/db/queryutil.h
@@ -278,7 +278,7 @@ namespace mongo {
         FieldRangeSet _singleKey;
         FieldRangeSet _multiKey;
         friend class FieldRangeOrSet;
-        friend class QueryUtilIndexed;
+        friend struct QueryUtilIndexed;
     };
 
     class IndexSpec;
@@ -407,7 +407,7 @@ namespace mongo {
         // ensure memory is owned
         list<FieldRangeSetPair> _oldOrSets;
         bool _orFound;
-        friend class QueryUtilIndexed;
+        friend struct QueryUtilIndexed;
     };
 
     /** returns a string that when used as a matcher, would match a super set of regex()

diff --git 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.h b/db/repl/rs.h
@@ -379,7 +379,7 @@ namespace mongo {
         void loadConfig();
 
         list<HostAndPort> memberHostnames() const;
-        const ReplSetConfig::MemberCfg& myConfig() const { return _self->config(); }
+        const ReplSetConfig::MemberCfg& myConfig() const { assert( _self ); return _self->config(); }
         bool iAmArbiterOnly() const { return myConfig().arbiterOnly; }
         bool iAmPotentiallyHot() const { return myConfig().potentiallyHot() && elect.steppedDown <= time(0); }
     protected:
@@ -505,6 +505,7 @@ namespace mongo {
 
     inline Member::Member(HostAndPort h, unsigned ord, const ReplSetConfig::MemberCfg *c, bool self) :
         _config(*c), _h(h), _hbinfo(ord) {
+        assert(c);
         if( self )
             _hbinfo.health = 1.0;
     }

diff --git a/db/repl/rs_config.cpp b/db/repl/rs_config.cpp
@@ -68,7 +68,7 @@ namespace mongo {
 
             cx.db()->flushFiles(true);
         }
-        DEV log() << "replSet saveConfigLocally done" << rsLog;
+        log() << "replSet saveConfigLocally done" << rsLog;
     }
 
     bo ReplSetConfig::MemberCfg::asBson() const {

diff --git 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();
@@ -488,7 +493,11 @@ namespace mongo {
                are no heartbeat threads, so we do it here to be sure.  this is relevant if the singleton
                member has done a stepDown() and needs to come back up.
                */
-            OCCASIONALLY mgr->send( boost::bind(&Manager::msgCheckNewState, theReplSet->mgr) );
+            OCCASIONALLY {
+            	log() << "replSet default heartbeat starting..." << rsLog;
+            	mgr->send( boost::bind(&Manager::msgCheckNewState, theReplSet->mgr) );
+            	log() << "replSet heartbeat finished" << rsLog;
+            }
         }
     }
 

diff --git a/dbtests/test.vcxproj b/dbtests/test.vcxproj
@@ -285,6 +285,10 @@
     <ClCompile Include="..\db\geo\haystack.cpp" />
     <ClCompile Include="..\db\mongommf.cpp" />
     <ClCompile Include="..\db\projection.cpp" />
+    <ClCompile Include="..\db\querypattern.cpp">
+      <PrecompiledHeader Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">NotUsing</PrecompiledHeader>
+      <PrecompiledHeader Condition="'$(Configuration)|$(Platform)'=='Release|x64'">NotUsing</PrecompiledHeader>
+    </ClCompile>
     <ClCompile Include="..\db\repl\consensus.cpp" />
     <ClCompile Include="..\db\repl\heartbeat.cpp" />
     <ClCompile Include="..\db\repl\manager.cpp" />

diff --git a/dbtests/test.vcxproj.filters b/dbtests/test.vcxproj.filters
@@ -773,6 +773,9 @@
     <ClCompile Include="..\db\dbcommands_admin.cpp">
       <Filter>db\cpp</Filter>
     </ClCompile>
+    <ClCompile Include="..\db\querypattern.cpp">
+      <Filter>db</Filter>
+    </ClCompile>
   </ItemGroup>
   <ItemGroup>
     <None Include="..\SConstruct">

diff --git a/jstests/check_shard_index.js b/jstests/check_shard_index.js
@@ -42,4 +42,9 @@ assert.eq( 3 , f.count() , "2. count after initial insert should be 3" );
 res = db.runCommand( { checkShardingIndex: "test.jstests_shardingindex" , keyPattern: {x:1, y:1} , force: true });
 assert.eq( false , res.ok , "2b " + tojson(res) );
 
+// 
+res = db.runCommand( { checkShardingIndex: "test.jstests_shardingindex" , keyPattern: {_id:1} , force: true });
+assert.eq( true , res.ok , "3a " + tojson(res) );
+assert( res.idskip , "3b " + tojson(res) )
+
 print("PASSED");
diff --git a/jstests/delx.js b/jstests/delx.js
@@ -23,6 +23,7 @@ x.next();
 y.next();
 
 a.foo.remove( { _id : { $gt : 50 } } );
+db.getLastError();
 
 assert.eq( 51 , a.foo.find().itcount() , "B1" )
 assert.eq( 100 , b.foo.find().itcount() , "B2" )