SERVER-16237 Don't check shard version if the replica set monitor knows the primary is down

4 files changed, 34 insertions, 1 deletions

diff --git a/src/mongo/client/dbclient_rs.cpp b/src/mongo/client/dbclient_rs.cpp
index 29942ea4bf0..9311bf3d76b 100644
--- a/src/mongo/client/dbclient_rs.cpp
+++ b/src/mongo/client/dbclient_rs.cpp
@@ -178,6 +178,13 @@ namespace {
         return rsm->getServerAddress();
     }
 
+    HostAndPort DBClientReplicaSet::getSuspectedPrimaryHostAndPort() const {
+        if (!_master) {
+            return HostAndPort();
+        }
+        return _master->getServerHostAndPort();
+    }
+
     void DBClientReplicaSet::setRunCommandHook(DBClientWithCommands::RunCommandHookFunc func) {
         // Set the hooks in both our sub-connections and in ourselves.
         if (_master) {
diff --git a/src/mongo/client/dbclient_rs.h b/src/mongo/client/dbclient_rs.h
index cbd1ff01438..f77dcd8fc28 100644
--- a/src/mongo/client/dbclient_rs.h
+++ b/src/mongo/client/dbclient_rs.h
@@ -133,6 +133,17 @@ namespace mongo {
 
         // ----- informational ----
 
+        /**
+         * Gets the replica set name of the set we are connected to.
+         */
+        const std::string& getSetName() const { return _setName; }
+
+        /**
+         * Returns the HostAndPort of the server this connection believes belongs to the primary,
+         * or returns an empty HostAndPort if it doesn't know about a current primary.
+         */
+        HostAndPort getSuspectedPrimaryHostAndPort() const;
+
         double getSoTimeout() const { return _so_timeout; }
 
         string toString() const { return getServerAddress(); }
diff --git a/src/mongo/client/dbclientinterface.h b/src/mongo/client/dbclientinterface.h
index 8ac36ea15b4..6f05fce5637 100644
--- a/src/mongo/client/dbclientinterface.h
+++ b/src/mongo/client/dbclientinterface.h
@@ -1309,6 +1309,7 @@ namespace mongo {
         }
 
         string getServerAddress() const { return _serverString; }
+        const HostAndPort& getServerHostAndPort() const { return _server; }
 
         virtual void killCursor( long long cursorID );
         virtual bool callRead( Message& toSend , Message& response ) { return call( toSend , response ); }
diff --git a/src/mongo/client/parallel.cpp b/src/mongo/client/parallel.cpp
index 4316c0ef29a..8ef983abc53 100644
--- a/src/mongo/client/parallel.cpp
+++ b/src/mongo/client/parallel.cpp
@@ -22,6 +22,8 @@
 
 #include "mongo/client/connpool.h"
 #include "mongo/client/dbclientcursor.h"
+#include "mongo/client/dbclient_rs.h"
+#include "mongo/client/replica_set_monitor.h"
 #include "mongo/db/dbmessage.h"
 #include "mongo/db/query/lite_parsed_query.h"
 #include "mongo/s/chunk.h"
@@ -571,8 +573,20 @@ namespace mongo {
         bool allowShardVersionFailure =
             rawConn->type() == ConnectionString::SET &&
             DBClientReplicaSet::isSecondaryQuery( _qSpec.ns(), _qSpec.query(), _qSpec.options() );
+        bool connIsDown = rawConn->isFailed();
+        if (allowShardVersionFailure && !connIsDown) {
+            // If the replica set connection believes that it has a valid primary that is up,
+            // confirm that the replica set monitor agrees that the suspected primary is indeed up.
+            const DBClientReplicaSet* replConn = dynamic_cast<const DBClientReplicaSet*>(rawConn);
+            ReplicaSetMonitorPtr rsMonitor = ReplicaSetMonitor::get(replConn->getSetName());
+            if (!rsMonitor->isHostUp(replConn->getSuspectedPrimaryHostAndPort())) {
+                connIsDown = true;
+            }
+        }
 
-        if ( allowShardVersionFailure && rawConn->isFailed() ) {
+        if (allowShardVersionFailure && connIsDown) {
+            // If we're doing a secondary-allowed query and the primary is down, don't attempt to
+            // set the shard version.
 
             state->conn->donotCheckVersion();