repl: logKeepAlive can block readers SERVER-1202

2 files changed, 64 insertions, 12 deletions

diff --git a/db/concurrency.h b/db/concurrency.h
index de8f242d61f..5c5c11c9c9a 100644
--- a/db/concurrency.h
+++ b/db/concurrency.h
@@ -113,16 +113,26 @@ namespace mongo {
         bool atLeastReadLocked() { return _state.get() != 0; }
         void assertAtLeastReadLocked() { assert(atLeastReadLocked()); }
 
-        void lock() { 
+        bool _checkWriteLockAlready(){
             //DEV cout << "LOCK" << endl;
             DEV assert( haveClient() );
                 
             int s = _state.get();
             if( s > 0 ) {
                 _state.set(s+1);
-                return;
+                return true;
             }
+
             massert( 10293 , (string)"internal error: locks are not upgradeable: " + sayClientState() , s == 0 );
+
+            return false;
+        }
+
+        void lock() { 
+            
+            if ( _checkWriteLockAlready() )
+                return;
+            
             _state.set(1);
 
             curopWaitingForLock( 1 );
@@ -131,6 +141,23 @@ namespace mongo {
 
             _minfo.entered();
         }
+
+        bool lock_try() {
+            if ( _checkWriteLockAlready() )
+                return true;            
+            
+            curopWaitingForLock( 1 );
+            bool got = _m.try_lock(); 
+            curopGotLock();
+            
+            if ( got ){
+                _minfo.entered();
+                _state.set(1);
+            }                
+            
+            return got;
+        }
+        
         void unlock() { 
             //DEV cout << "UNLOCK" << endl;
             int s = _state.get();
@@ -227,12 +254,21 @@ namespace mongo {
         void lock() { 
 #ifdef HAVE_READLOCK
             m.lock();
+#error this should be impossible?
 #else
             boost::detail::thread::lock_ops<boost::recursive_mutex>::lock(m);
 #endif
             _minfo.entered();
         }
 
+        bool lock_try(){
+            bool got = boost::detail::thread::lock_ops<boost::recursive_mutex>::trylock(m);
+            if ( got ){
+                _minfo.entered();
+            }
+            return got;
+        }
+
         void releaseEarly() {
             assertWriteLocked(); // aso must not be recursive, although we don't verify that in the old boost version
             assert( !_releasedEarly.get() );
@@ -326,6 +362,23 @@ namespace mongo {
         }
         bool _got;
     };
+
+    struct writelocktry {
+        writelocktry( const string&ns ){
+            _got = dbMutex.lock_try();
+        }
+        ~writelocktry() {
+            if ( _got ){
+                dbunlocking_write();
+                dbMutex.unlock();
+            }
+        }
+        bool got(){
+            return _got;
+        }
+        bool _got;
+    };
+
     
     struct atleastreadlock {
         atleastreadlock( const string& ns ){
diff --git a/db/repl.cpp b/db/repl.cpp
index a0ac16e9bc1..efb078b3ac3 100644
--- a/db/repl.cpp
+++ b/db/repl.cpp
@@ -1720,22 +1720,21 @@ namespace mongo {
         sleepsecs(4);
         Client::initThread("replmaster");
         while( 1 ) {
-            {
-                dblock lk;
-                cc().getAuthenticationInfo()->authorize("admin");   
-            }
             sleepsecs(10);
             /* write a keep-alive like entry to the log.  this will make things like 
                printReplicationStatus() and printSlaveReplicationStatus() stay up-to-date
                even when things are idle.
             */
             {
-                writelock lk("");
-                try { 
-                    logKeepalive();
-                }
-                catch(...) { 
-                    log() << "caught exception in replMasterThread()" << endl;
+                writelocktry lk("");
+                if ( lk.got() ){
+                    cc().getAuthenticationInfo()->authorize("admin");   
+                    try { 
+                        logKeepalive();
+                    }
+                    catch(...) { 
+                        log() << "caught exception in replMasterThread()" << endl;
+                    }
                 }
             }
         }