rs

db/repl/consensus.cpp

@@ -64,7 +64,7 @@ namespace mongo {
 
     static const int VETO = -10000;
 
-    bool ReplSet::Consensus::electSelf() {
+    void ReplSet::Consensus::_electSelf() {
         ReplSet::Member& me = *rs._self;
         electCmd = BSON(
                "replSetElect" << 1 <<
@@ -96,11 +96,26 @@ namespace mongo {
             if( time(0) - start > 30 ) {
                 // defensive; should never happen as we have timeouts on connection and operation for our conn
                 log() << "replSet too much time passed during election, ignoring result" << rsLog;
-                return false;
             }
-            return true;
+            /* succeeded. */
+            rs._myState = PRIMARY;
+            log() << "replSet elected self as primary" << rsLog;
+            return;
         } 
-        return false;
+    }
+
+    void ReplSet::Consensus::electSelf() {
+        static mutex m;
+        {
+            scoped_lock lk(m);
+            if( inprog ) return;
+            inprog = true;
+        }
+        try { _electSelf(); } catch(...) { }
+        {
+            scoped_lock lk(m);
+            inprog = false;
+        }
     }
 
 }

db/repl/manager.cpp

@@ -53,51 +53,51 @@ namespace mongo {
     /** called as the health threads get new results */
     void ReplSet::Manager::checkNewState() {
         static mutex m;
-        scoped_lock lk(m);
+        {
+            scoped_lock lk(m);
 
-        //log() << "replSet checkNewState " << rsLog;
+            const Member *p = _rs->currentPrimary();
-
+            const Member *p2 = findOtherPrimary();
-        const Member *p = _rs->currentPrimary();
+            try { p2 = findOtherPrimary(); }
-        const Member *p2 = findOtherPrimary();
+            catch(string s) { 
-        try { p2 = findOtherPrimary(); }
+                /* two other nodes think they are primary (asynchronously polled) -- wait for things to settle down. */
-        catch(string s) { 
+                log() << "replSet warning DIAG TODO " << s << rsLog;
-            log() << "replSet warning DIAG TODO " << s << rsLog;
-            return;
-        }
-
-        if( p == p2 && p ) return;
-
-        if( p2 ) { 
-            /* someone else thinks they are primary. */
-            if( p == p2 ) 
                 return;
-            if( p == 0 )
+            }
-                noteARemoteIsPrimary(p2); return;
-            if( p != _rs->_self )
-                noteARemoteIsPrimary(p2); return;
-            /* we thought we were primary, yet now someone else thinks they are. */
-            if( !_rs->elect.aMajoritySeemsToBeUp() )
-                noteARemoteIsPrimary(p2); return;
-            /* ignore for now, keep thinking we are master */
-            return;
-        }
 
-        if( p ) { 
+            if( p == p2 && p ) return;
-            /* we are already primary, and nothing significant out there has changed. */\
-            /* todo: if !aMajoritySeemsToBeUp, relinquish */
-            assert( p == _rs->_self );
-            return;
-        }
 
-        /* no one seems to be primary.  shall we try to elect ourself? */
+            if( p2 ) { 
-        if( !_rs->elect.aMajoritySeemsToBeUp() ) { 
+                /* someone else thinks they are primary. */
-            _rs->_self->_lastHeartbeatErrMsg.set("can't see a majority, won't elect self");
+                if( p == p2 ) // already match
-            return;
+                    return;
-        }
+                if( p == 0 )
+                    noteARemoteIsPrimary(p2); return;
+                if( p != _rs->_self )
+                    noteARemoteIsPrimary(p2); return;
+                /* we thought we were primary, yet now someone else thinks they are. */
+                if( !_rs->elect.aMajoritySeemsToBeUp() )
+                    noteARemoteIsPrimary(p2); return;
+                /* ignore for now, keep thinking we are master */
+                return;
+            }
 
-        log() << "replSet todo elect self as primary primary" << rsLog;
+            if( p ) { 
-        _rs->_self->_lastHeartbeatErrMsg.set("todo code #2");
+                /* we are already primary, and nothing significant out there has changed. */
+                /* todo: if !aMajoritySeemsToBeUp, relinquish */
+                assert( p == _rs->_self );
+                return;
+            }
+
+            /* no one seems to be primary.  shall we try to elect ourself? */
+            if( !_rs->elect.aMajoritySeemsToBeUp() ) { 
+                _rs->_self->_lastHeartbeatErrMsg.set("can't see a majority, won't elect self");
+                return;
+            }
+
+            _rs->_self->_lastHeartbeatErrMsg.set("");
+        }
+        _rs->elect.electSelf();
     }
 
-
 }

db/repl/replset.h

@@ -54,7 +54,6 @@ namespace mongo {
         State _myState;
 
     public:
-
         void fatal();
         bool isMaster(const char *client);
         void fillIsMaster(BSONObjBuilder&);
@@ -66,8 +65,7 @@ namespace mongo {
            replsetname/host1,host2:port,...
            where :port is optional.
 
-           throws exception if a problem initializing.
+           throws exception if a problem initializing. */
-        */
         ReplSet(string cfgString);
 
         /* call after constructing to start - returns fairly quickly after launching its threads */
@@ -90,12 +88,15 @@ namespace mongo {
         void loadConfig();
         void initFromConfig(ReplSetConfig& c);//, bool save);
 
-        struct Consensus {
+        class Consensus {
             ReplSet &rs;
-            Consensus(ReplSet *t) : rs(*t) { }
+            bool inprog;
+            void _electSelf();
+        public:
+            Consensus(ReplSet *t) : rs(*t),inprog(false) { }
             int totalVotes() const;
             bool aMajoritySeemsToBeUp() const;
-            bool electSelf();
+            void electSelf();
         } elect;
 
     public: