backport of fix for SERVER-2481

3 files changed, 44 insertions, 1 deletions

diff --git a/client/dbclient_rs.cpp b/client/dbclient_rs.cpp
index ae01da3e5b3..3da036326aa 100644
--- a/client/dbclient_rs.cpp
+++ b/client/dbclient_rs.cpp
@@ -544,7 +544,7 @@ namespace mongo {
             // checkSlave will try a different slave automatically after a failure
             for ( int i=0; i<2; i++ ) {
                 try {
-                    return checkSlave()->query(ns,query,nToReturn,nToSkip,fieldsToReturn,queryOptions,batchSize);
+                    return checkSlaveQueryResult( checkSlave()->query(ns,query,nToReturn,nToSkip,fieldsToReturn,queryOptions,batchSize) );
                 }
                 catch ( DBException &e ) {
                     log() << "can't query replica set slave " << i << " : " << _slaveHost << e.what() << endl;
@@ -581,6 +581,38 @@ namespace mongo {
         assert(0);
     }
 
+    auto_ptr<DBClientCursor> DBClientReplicaSet::checkSlaveQueryResult( auto_ptr<DBClientCursor> result ){
+
+        bool isError = result->hasResultFlag( ResultFlag_ErrSet );
+        if( ! isError ) return result;
+
+        BSONObj error = result->peekOne();
+
+        BSONElement code = error["code"];
+        if( code.eoo() || ! code.isNumber() ){
+            warning() << "no code for error from secondary host " << _slaveHost << ", error was " << error << endl;
+            return result;
+        }
+
+        // We only check for "not master or secondary" errors here
+
+        // If the error code here ever changes, we need to change this code also
+        if( code.Int() == 13436 /* not master or secondary */ ){
+            isntSecondary();
+            throw DBException( str::stream() << "slave " << _slaveHost.toString() << " is no longer secondary", 14812 );
+        }
+
+        return result;
+    }
+
+    void DBClientReplicaSet::isntSecondary() {
+        log() << "slave no longer has secondary status: " << _slaveHost << endl;
+        // Failover to next slave
+        _monitor->notifySlaveFailure( _slaveHost );
+        _slave.reset();
+    }
+
+
     void DBClientReplicaSet::isntMaster() { 
         log() << "got not master for: " << _masterHost << endl;
         _monitor->notifyFailure( _masterHost );
diff --git a/client/dbclient_rs.h b/client/dbclient_rs.h
index e942d7b5b2a..548b46a2176 100644
--- a/client/dbclient_rs.h
+++ b/client/dbclient_rs.h
@@ -215,6 +215,9 @@ namespace mongo {
         /* this is the callback from our underlying connections to notify us that we got a "not master" error.
          */
         void isntMaster();
+        /* this is used to indicate we got a "not master or secondary" error from a secondary.
+         */
+        void isntSecondary();
 
         // ----- status ------
 
@@ -240,6 +243,9 @@ namespace mongo {
 
     private:
 
+        // Used to simplify slave-handling logic on errors
+        auto_ptr<DBClientCursor> checkSlaveQueryResult( auto_ptr<DBClientCursor> result );
+
         DBClientConnection * checkMaster();
         DBClientConnection * checkSlave();
 
diff --git a/client/dbclientcursor.h b/client/dbclientcursor.h
index 5d795f4d6e1..d176b892d33 100644
--- a/client/dbclientcursor.h
+++ b/client/dbclientcursor.h
@@ -86,6 +86,11 @@ namespace mongo {
             WARNING: no support for _putBack yet!
         */
         void peek(vector<BSONObj>&, int atMost);
+        BSONObj peekOne(){
+            vector<BSONObj> v;
+            peek( v, 1 );
+            return v.size() > 0 ? v[0] : BSONObj();
+        }
 
         /**
            iterate the rest of the cursor and return the number if items