SERVER-20748 Handle epoch mismatch explicitly in cluster find command

5 files changed, 34 insertions, 5 deletions

diff --git a/src/mongo/base/error_codes.err b/src/mongo/base/error_codes.err
index 7044584ab00..d900cb1119e 100644
--- a/src/mongo/base/error_codes.err
+++ b/src/mongo/base/error_codes.err
@@ -149,6 +149,7 @@ error_code("ExceededMemoryLimit", 146)
 error_code("ZLibError", 147)
 error_code("ReadConcernMajorityNotEnabled", 148)
 error_code("NoConfigMaster", 149)
+error_code("StaleEpoch", 150)
 
 # Non-sequential error codes (for compatibility only)
 error_code("RecvStaleConfig", 9996)
@@ -171,3 +172,5 @@ error_code("PrepareConfigsFailed", 13104);
 error_class("NetworkError", ["HostUnreachable", "HostNotFound", "NetworkTimeout"])
 error_class("Interruption", ["Interrupted", "InterruptedAtShutdown", "ExceededTimeLimit"])
 error_class("NotMasterError", ["NotMaster", "NotMasterNoSlaveOk"])
+error_class("StaleShardingError",
+            ["RecvStaleConfig", "SendStaleConfig", "StaleShardVersion", "StaleEpoch"])
diff --git a/src/mongo/db/commands.cpp b/src/mongo/db/commands.cpp
index ec130328bef..bd3d95e7295 100644
--- a/src/mongo/db/commands.cpp
+++ b/src/mongo/db/commands.cpp
@@ -424,8 +424,9 @@ void _generateErrorResponse(OperationContext* txn,
             static_cast<const SendStaleConfigException&>(exception);
         replyBuilder->setCommandReply(scex.toStatus(),
                                       BSON("ns" << scex.getns() << "vReceived"
-                                                << scex.getVersionReceived().toBSON() << "vWanted"
-                                                << scex.getVersionWanted().toBSON()));
+                                                << BSONArray(scex.getVersionReceived().toBSON())
+                                                << "vWanted"
+                                                << BSONArray(scex.getVersionWanted().toBSON())));
     } else {
         replyBuilder->setCommandReply(exception.toStatus());
     }
diff --git a/src/mongo/db/query/cursor_response.cpp b/src/mongo/db/query/cursor_response.cpp
index 3c86439f622..2954da683c9 100644
--- a/src/mongo/db/query/cursor_response.cpp
+++ b/src/mongo/db/query/cursor_response.cpp
@@ -34,6 +34,7 @@
 
 #include "mongo/bson/bsontypes.h"
 #include "mongo/rpc/get_status_from_command_result.h"
+#include "mongo/s/chunk_version.h"
 
 namespace mongo {
 
@@ -97,6 +98,13 @@ CursorResponse& CursorResponse::operator=(CursorResponse&& other) {
 StatusWith<CursorResponse> CursorResponse::parseFromBSON(const BSONObj& cmdResponse) {
     Status cmdStatus = getStatusFromCommandResult(cmdResponse);
     if (!cmdStatus.isOK()) {
+        if (ErrorCodes::isStaleShardingError(cmdStatus.code())) {
+            auto vWanted = ChunkVersion::fromBSON(cmdResponse, "vWanted");
+            auto vReceived = ChunkVersion::fromBSON(cmdResponse, "vReceived");
+            if (!vWanted.hasEqualEpoch(vReceived)) {
+                return Status(ErrorCodes::StaleEpoch, cmdStatus.reason());
+            }
+        }
         return cmdStatus;
     }
 
diff --git a/src/mongo/s/config.h b/src/mongo/s/config.h
index 39b56f529c1..68a2ca18ae8 100644
--- a/src/mongo/s/config.h
+++ b/src/mongo/s/config.h
@@ -166,6 +166,10 @@ public:
 
     void setPrimary(OperationContext* txn, const std::string& s);
 
+    /**
+     * Returns true if it is successful at loading the DBConfig, false if the database is not found,
+     * and throws on all other errors.
+     */
     bool load(OperationContext* txn);
     bool reload(OperationContext* txn);
 
@@ -182,6 +186,10 @@ protected:
                                  std::set<ShardId>& shardIds,
                                  std::string& errmsg);
 
+    /**
+     * Returns true if it is successful at loading the DBConfig, false if the database is not found,
+     * and throws on all other errors.
+     */
     bool _load(OperationContext* txn);
 
     void _save(OperationContext* txn, bool db = true, bool coll = true);
diff --git a/src/mongo/s/query/cluster_find.cpp b/src/mongo/s/query/cluster_find.cpp
index 75b1582d085..8b197b81eb5 100644
--- a/src/mongo/s/query/cluster_find.cpp
+++ b/src/mongo/s/query/cluster_find.cpp
@@ -369,8 +369,8 @@ StatusWith<CursorId> ClusterFind::runQuery(OperationContext* txn,
         }
         auto status = std::move(cursorId.getStatus());
 
-        if (status != ErrorCodes::SendStaleConfig && status != ErrorCodes::RecvStaleConfig) {
-            // Errors other than receiving a stale config message from MongoD are fatal to the
+        if (!ErrorCodes::isStaleShardingError(status.code())) {
+            // Errors other than receiving a stale metadata message from MongoD are fatal to the
             // operation. Network errors and replication retries happen at the level of the
             // AsyncResultsMerger.
             return status;
@@ -379,7 +379,16 @@ StatusWith<CursorId> ClusterFind::runQuery(OperationContext* txn,
         LOG(1) << "Received error status for query " << query.toStringShort() << " on attempt "
                << retries << " of " << kMaxStaleConfigRetries << ": " << status;
 
-        chunkManager = dbConfig.getValue()->getChunkManagerIfExists(txn, query.nss().ns(), true);
+        const bool staleEpoch = (status == ErrorCodes::StaleEpoch);
+        if (staleEpoch) {
+            if (!dbConfig.getValue()->reload(txn)) {
+                // If the reload failed that means the database wasn't found, so successfully return
+                // an empty result set without creating a cursor.
+                return CursorId(0);
+            }
+        }
+        chunkManager =
+            dbConfig.getValue()->getChunkManagerIfExists(txn, query.nss().ns(), true, staleEpoch);
         if (!chunkManager) {
             dbConfig.getValue()->getChunkManagerOrPrimary(
                 txn, query.nss().ns(), chunkManager, primary);