SERVER-8710 check balancer is stopped before upgrading

5 files changed, 47 insertions, 0 deletions

diff --git a/jstests/multiVersion/upgrade_cluster_v3_to_v4.js b/jstests/multiVersion/upgrade_cluster_v3_to_v4.js
index f922bc1906d..3431a8e735e 100644
--- a/jstests/multiVersion/upgrade_cluster_v3_to_v4.js
+++ b/jstests/multiVersion/upgrade_cluster_v3_to_v4.js
@@ -32,6 +32,9 @@ var options = {
 
 var st = new ShardingTest({ shards : 2, mongos : 2, other : options });
 
+// Just stop balancer, to simulate race conds
+st.setBalancer(false);
+
 var shards = st.s0.getDB("config").shards.find().toArray();
 var configConnStr = st._configDB;
 
diff --git a/jstests/multiVersion/upgrade_cluster_v3_to_v4_db.js b/jstests/multiVersion/upgrade_cluster_v3_to_v4_db.js
index 000484deb50..46124a8227a 100644
--- a/jstests/multiVersion/upgrade_cluster_v3_to_v4_db.js
+++ b/jstests/multiVersion/upgrade_cluster_v3_to_v4_db.js
@@ -19,6 +19,9 @@ var options = {
 
 var st = new ShardingTest({ shards : 1, mongos : 1, other : options });
 
+// Just set balancer to false, don't wait for it
+st.setBalancer(false);
+
 var mongos = st.s0
 var config = mongos.getDB("config")
 var admin = mongos.getDB("admin")
diff --git a/jstests/multiVersion/upgrade_cluster_v3_to_v4_wait_for_mongos.js b/jstests/multiVersion/upgrade_cluster_v3_to_v4_wait_for_mongos.js
index 64da139e609..ba859edd7a7 100644
--- a/jstests/multiVersion/upgrade_cluster_v3_to_v4_wait_for_mongos.js
+++ b/jstests/multiVersion/upgrade_cluster_v3_to_v4_wait_for_mongos.js
@@ -19,6 +19,9 @@ var options = {
 
 var st = new ShardingTest({ shards : 1, mongos : 1, other : options });
 
+// Turn balancer off, don't wait
+st.setBalancer(false);
+
 var mongos = st.s0;
 
 jsTest.log( "Starting v2.0 mongos..." );
diff --git a/jstests/multiVersion/upgrade_cluster_v3_to_v4_with_parallel_ops.js b/jstests/multiVersion/upgrade_cluster_v3_to_v4_with_parallel_ops.js
index b0ad24d55b8..77579de5615 100644
--- a/jstests/multiVersion/upgrade_cluster_v3_to_v4_with_parallel_ops.js
+++ b/jstests/multiVersion/upgrade_cluster_v3_to_v4_with_parallel_ops.js
@@ -176,6 +176,10 @@ printShardingStatus(config, true);
 
 jsTest.log("Upgrading config db from v3 to v4...");
 
+// Just stop the balancer, but don't wait for it to stop, to simulate race conds
+st.setBalancer(false);
+printjson(config.settings.find().toArray());
+
 var startTime = new Date();
 
 // Make sure up
diff --git a/src/mongo/s/config_upgrade.cpp b/src/mongo/s/config_upgrade.cpp
index 62a550151e5..8a8b91b4d84 100644
--- a/src/mongo/s/config_upgrade.cpp
+++ b/src/mongo/s/config_upgrade.cpp
@@ -25,6 +25,7 @@
 #include "mongo/s/mongo_version_range.h"
 #include "mongo/s/type_config_version.h"
 #include "mongo/s/type_database.h"
+#include "mongo/s/type_settings.h"
 #include "mongo/s/type_shard.h"
 #include "mongo/util/assert_util.h"
 #include "mongo/util/version.h"
@@ -277,6 +278,30 @@ namespace mongo {
         return VersionStatus_NeedUpgrade;
     }
 
+    // Returns true if we can confirm the balancer is stopped
+    bool _isBalancerStopped(const ConnectionString& configLoc, string* errMsg) {
+        
+        // Get the balancer information
+        scoped_ptr<ScopedDbConnection> connPtr;
+        
+        BSONObj balancerDoc;
+        try {
+            connPtr.reset(ScopedDbConnection::getInternalScopedDbConnection(configLoc, 30));
+            ScopedDbConnection& conn = *connPtr;
+            
+            balancerDoc = conn->findOne(SettingsType::ConfigNS,
+                                        BSON(SettingsType::key("balancer")));
+        }
+        catch (const DBException& e) {
+            *errMsg = e.toString();
+            return false;
+        }
+
+        connPtr->done();
+        
+        return balancerDoc[SettingsType::balancerStopped()].trueValue();
+    }
+
     // Checks that all config servers are online
     bool _checkConfigServersAlive(const ConnectionString& configLoc, string* errMsg) {
         
@@ -450,6 +475,15 @@ namespace mongo {
             return false;
         }
 
+        // Check whether or not the balancer is online, if it is online we will not upgrade
+        if (!_isBalancerStopped(configLoc, errMsg)) {
+            
+            *errMsg = stream() << "balancer must be stopped for config upgrade"
+                               << causedBy(errMsg);
+            
+            return false;
+        }
+
         //
         // Acquire a lock for the upgrade process.
         //