SERVER-9417 assert if we stepDown while waiting to satisfy GLE

2 files changed, 48 insertions, 2 deletions

diff --git a/jstests/replsets/stepdown3.js b/jstests/replsets/stepdown3.js
new file mode 100644
index 00000000000..ec3261b6637
--- /dev/null
+++ b/jstests/replsets/stepdown3.js
@@ -0,0 +1,39 @@
+// Test that GLE asserts when the primary steps down while we're waiting for w:
+
+var replTest = new ReplSetTest({ name: 'testSet', nodes: 2 });
+var nodes = replTest.startSet();
+replTest.initiate();
+var master = replTest.getMaster();
+
+// do a write to allow stepping down of the primary;
+// otherwise, the primary will refuse to step down
+print("\ndo a write");
+master.getDB("test").foo.insert({x:1});
+replTest.awaitReplication();
+
+// lock secondary, to pause replication
+print("\nlock secondary");
+var locked = replTest.liveNodes.slaves[0];
+printjson( locked.getDB("admin").runCommand({fsync : 1, lock : 1}) );
+
+// do a write
+print("\ndo a write");
+master.getDB("test").foo.insert({x:2});
+
+// step down the primary asyncronously
+print("stepdown");
+var command = "sleep(4000); tojson(rs.stepDown());"
+var mongoprogram = startMongoProgram( "mongo", "--quiet", "--port", "" + master.port, 
+                                      "--eval", command );
+
+print("getlasterror; should assert");
+var gleFunction = function() { 
+    master.getDB("test").runCommand({getLastError : 1, w: 2 , wtimeout :30000 }) 
+};
+var result = assert.throws(gleFunction);
+print("result of gle:");
+printjson(result);
+
+// unlock and shut down
+printjson(locked.getDB("admin").$cmd.sys.unlock.findOne());
+replTest.stopSet();
diff --git a/src/mongo/db/repl_block.cpp b/src/mongo/db/repl_block.cpp
index 37d9321ea16..2bd59d530c3 100644
--- a/src/mongo/db/repl_block.cpp
+++ b/src/mongo/db/repl_block.cpp
@@ -179,7 +179,9 @@ namespace mongo {
         }
 
         bool waitForReplication(OpTime& op, int w, int maxSecondsToWait) {
-            massert( 16806, "waitForReplication called but not master anymore", _isMaster() );
+            static const int noLongerMasterAssertCode = 16806;
+            massert(noLongerMasterAssertCode, 
+                    "waitForReplication called but not master anymore", _isMaster() );
 
             if ( w <= 1 )
                 return true;
@@ -192,8 +194,13 @@ namespace mongo {
             
             scoped_lock mylk(_mutex);
             while ( ! _replicatedToNum_slaves_locked( op, w ) ) {
-                if ( ! _threadsWaitingForReplication.timed_wait( mylk.boost() , xt ) )
+                if ( ! _threadsWaitingForReplication.timed_wait( mylk.boost() , xt ) ) {
+                    massert(noLongerMasterAssertCode,
+                            "waitForReplication called but not master anymore", _isMaster());
                     return false;
+                }
+                massert(noLongerMasterAssertCode, 
+                        "waitForReplication called but not master anymore", _isMaster());
             }
             return true;
         }