SERVER-37929 reload ShardRegistry on config servers on rollback

(cherry picked from commit fc1b17cfdeffa13b5326050893658e7ba4df57a1)

3 files changed, 32 insertions, 0 deletions

diff --git a/src/mongo/db/SConscript b/src/mongo/db/SConscript
index d915f3d1a4f..1315ef36baa 100644
--- a/src/mongo/db/SConscript
+++ b/src/mongo/db/SConscript
@@ -764,6 +764,8 @@ env.Library(
         'views/views_mongod',
         '$BUILD_DIR/mongo/base',
         '$BUILD_DIR/mongo/db/catalog/uuid_catalog',
+        '$BUILD_DIR/mongo/s/coreshard',
+        "$BUILD_DIR/mongo/s/grid",
     ],
     LIBDEPS_PRIVATE=[
         '$BUILD_DIR/mongo/db/commands/mongod_fcv',
diff --git a/src/mongo/db/op_observer_impl.cpp b/src/mongo/db/op_observer_impl.cpp
index 23e0f7427a4..7b106277c61 100644
--- a/src/mongo/db/op_observer_impl.cpp
+++ b/src/mongo/db/op_observer_impl.cpp
@@ -52,6 +52,8 @@
 #include "mongo/db/server_options.h"
 #include "mongo/db/session_catalog.h"
 #include "mongo/db/views/durable_view_catalog.h"
+#include "mongo/s/client/shard_registry.h"
+#include "mongo/s/grid.h"
 #include "mongo/scripting/engine.h"
 #include "mongo/util/assert_util.h"
 #include "mongo/util/fail_point_service.h"
@@ -982,6 +984,19 @@ void OpObserverImpl::onReplicationRollback(OperationContext* opCtx,
     if (rbInfo.shardIdentityRolledBack) {
         fassertFailedNoTrace(50712);
     }
+
+    // The code below will force the config server to update its shard registry.
+    // Otherwise it may have the stale data that has been just rolled back.
+    if (serverGlobalParams.clusterRole == ClusterRole::ConfigServer) {
+        if (auto shardRegistry = Grid::get(opCtx)->shardRegistry()) {
+            auto& readConcernArgs = repl::ReadConcernArgs::get(opCtx);
+            ON_BLOCK_EXIT([ argsCopy = readConcernArgs, &readConcernArgs ] {
+                readConcernArgs = std::move(argsCopy);
+            });
+            readConcernArgs = repl::ReadConcernArgs(repl::ReadConcernLevel::kLocalReadConcern);
+            shardRegistry->reload(opCtx);
+        }
+    }
 }
 
 }  // namespace mongo
diff --git a/src/mongo/db/repl/rs_rollback.cpp b/src/mongo/db/repl/rs_rollback.cpp
index f1a9c2d3493..65ba8fff86b 100644
--- a/src/mongo/db/repl/rs_rollback.cpp
+++ b/src/mongo/db/repl/rs_rollback.cpp
@@ -73,6 +73,8 @@
 #include "mongo/db/repl/rslog.h"
 #include "mongo/db/s/shard_identity_rollback_notifier.h"
 #include "mongo/db/session_catalog.h"
+#include "mongo/s/client/shard_registry.h"
+#include "mongo/s/grid.h"
 #include "mongo/util/exit.h"
 #include "mongo/util/fail_point_service.h"
 #include "mongo/util/log.h"
@@ -1465,6 +1467,19 @@ void rollback_internal::syncFixUp(OperationContext* opCtx,
         validator->resetKeyManagerCache();
     }
 
+    // The code below will force the config server to update its shard registry.
+    // Otherwise it may have the stale data that has been just rolled back.
+    if (serverGlobalParams.clusterRole == ClusterRole::ConfigServer) {
+        if (auto shardRegistry = Grid::get(opCtx)->shardRegistry()) {
+            auto& readConcernArgs = repl::ReadConcernArgs::get(opCtx);
+            ON_BLOCK_EXIT([ argsCopy = readConcernArgs, &readConcernArgs ] {
+                readConcernArgs = std::move(argsCopy);
+            });
+            readConcernArgs = repl::ReadConcernArgs(repl::ReadConcernLevel::kLocalReadConcern);
+            shardRegistry->reload(opCtx);
+        }
+    }
+
     // Reload the lastAppliedOpTime and lastDurableOpTime value in the replcoord and the
     // lastAppliedHash value in bgsync to reflect our new last op. The rollback common point does
     // not necessarily represent a consistent database state. For example, on a secondary, we may