SERVER-56307 fix the convergence criteria to end the catchup phase

(cherry picked from commit 49209fc34e19bbb15405b0927e38ff3e7d9e9dc5)

4 files changed, 31 insertions, 8 deletions

diff --git a/src/mongo/db/s/migration_chunk_cloner_source_legacy.cpp b/src/mongo/db/s/migration_chunk_cloner_source_legacy.cpp
index b1ac46654ff..81e26da2d8a 100644
--- a/src/mongo/db/s/migration_chunk_cloner_source_legacy.cpp
+++ b/src/mongo/db/s/migration_chunk_cloner_source_legacy.cpp
@@ -535,6 +535,7 @@ void MigrationChunkClonerSourceLegacy::_addToTransferModsQueue(
         case 'd': {
             stdx::lock_guard<Latch> sl(_mutex);
             _deleted.push_back(idObj);
+            ++_untransferredDeletesCounter;
             _memoryUsed += idObj.firstElement().size() + 5;
         } break;
 
@@ -542,6 +543,7 @@ void MigrationChunkClonerSourceLegacy::_addToTransferModsQueue(
         case 'u': {
             stdx::lock_guard<Latch> sl(_mutex);
             _reload.push_back(idObj);
+            ++_untransferredUpsertsCounter;
             _memoryUsed += idObj.firstElement().size() + 5;
         } break;
 
@@ -746,7 +748,9 @@ Status MigrationChunkClonerSourceLegacy::nextModsBatch(OperationContext* opCtx,
     // Put back remaining ids we didn't consume
     stdx::unique_lock<Latch> lk(_mutex);
     _deleted.splice(_deleted.cbegin(), deleteList);
+    _untransferredDeletesCounter = _deleted.size();
     _reload.splice(_reload.cbegin(), updateList);
+    _untransferredUpsertsCounter = _reload.size();
 
     return Status::OK();
 }
@@ -758,7 +762,9 @@ void MigrationChunkClonerSourceLegacy::_cleanup(OperationContext* opCtx) {
     _drainAllOutstandingOperationTrackRequests(lk);
 
     _reload.clear();
+    _untransferredUpsertsCounter = 0;
     _deleted.clear();
+    _untransferredDeletesCounter = 0;
 }
 
 StatusWith<BSONObj> MigrationChunkClonerSourceLegacy::_callRecipient(const BSONObj& cmdObj) {
@@ -1050,14 +1056,24 @@ Status MigrationChunkClonerSourceLegacy::_checkRecipientCloningStatus(OperationC
         }
 
         if (res["state"].String() == "catchup" && supportsCriticalSectionDuringCatchUp) {
-            int64_t estimatedUntransferredModsSize = _deleted.size() * _averageObjectIdSize +
-                _reload.size() * _averageObjectSizeForCloneLocs;
+            int64_t estimatedUntransferredModsSize =
+                _untransferredDeletesCounter * _averageObjectIdSize +
+                _untransferredUpsertsCounter * _averageObjectSizeForCloneLocs;
             auto estimatedUntransferredChunkPercentage =
                 (std::min(_args.getMaxChunkSizeBytes(), estimatedUntransferredModsSize) * 100) /
                 _args.getMaxChunkSizeBytes();
             if (estimatedUntransferredChunkPercentage < maxCatchUpPercentageBeforeBlockingWrites) {
                 // The recipient is sufficiently caught-up with the writes on the donor.
                 // Block writes, so that it can drain everything.
+                LOGV2_DEBUG(5630700,
+                            1,
+                            "moveChunk data transfer within threshold to allow write blocking",
+                            "_untransferredUpsertsCounter"_attr = _untransferredUpsertsCounter,
+                            "_untransferredDeletesCounter"_attr = _untransferredDeletesCounter,
+                            "_averageObjectSizeForCloneLocs"_attr = _averageObjectSizeForCloneLocs,
+                            "_averageObjectIdSize"_attr = _averageObjectIdSize,
+                            "maxChunksSizeBytes"_attr = _args.getMaxChunkSizeBytes(),
+                            "_sessionId"_attr = _sessionId.toString());
                 return Status::OK();
             }
         }
diff --git a/src/mongo/db/s/migration_chunk_cloner_source_legacy.h b/src/mongo/db/s/migration_chunk_cloner_source_legacy.h
index f41127afad1..7bc9b80df0b 100644
--- a/src/mongo/db/s/migration_chunk_cloner_source_legacy.h
+++ b/src/mongo/db/s/migration_chunk_cloner_source_legacy.h
@@ -373,10 +373,16 @@ private:
     // List of _id of documents that were modified that must be re-cloned (xfer mods)
     std::list<BSONObj> _reload;
 
+    // Amount of upsert xfer mods that have not yet reached the recipient.
+    size_t _untransferredUpsertsCounter{0};
+
     // List of _id of documents that were deleted during clone that should be deleted later (xfer
     // mods)
     std::list<BSONObj> _deleted;
 
+    // Amount of delete xfer mods that have not yet reached the recipient.
+    size_t _untransferredDeletesCounter{0};
+
     // Total bytes in _reload + _deleted (xfer mods)
     uint64_t _memoryUsed{0};
 
diff --git a/src/mongo/db/s/migration_destination_manager.cpp b/src/mongo/db/s/migration_destination_manager.cpp
index 877cc863702..9e788b4ec40 100644
--- a/src/mongo/db/s/migration_destination_manager.cpp
+++ b/src/mongo/db/s/migration_destination_manager.cpp
@@ -286,7 +286,7 @@ MigrationDestinationManager::State MigrationDestinationManager::getState() const
     return _state;
 }
 
-void MigrationDestinationManager::setState(State newState) {
+void MigrationDestinationManager::_setState(State newState) {
     stdx::lock_guard<Latch> sl(_mutex);
     _state = newState;
     _stateChangedCV.notify_all();
@@ -1065,7 +1065,7 @@ void MigrationDestinationManager::_migrateDriver(OperationContext* outerOpCtx) {
     repl::OpTime lastOpApplied;
     {
         // 3. Initial bulk clone
-        setState(CLONE);
+        _setState(CLONE);
 
         _sessionMigration->start(opCtx->getServiceContext());
 
@@ -1175,7 +1175,7 @@ void MigrationDestinationManager::_migrateDriver(OperationContext* outerOpCtx) {
 
     {
         // 4. Do bulk of mods
-        setState(CATCHUP);
+        _setState(CATCHUP);
 
         while (true) {
             auto res = uassertStatusOKWithContext(
@@ -1272,7 +1272,7 @@ void MigrationDestinationManager::_migrateDriver(OperationContext* outerOpCtx) {
 
     {
         // 5. Wait for commit
-        setState(STEADY);
+        _setState(STEADY);
 
         bool transferAfterCommit = false;
         while (getState() == STEADY || getState() == COMMIT_START) {
@@ -1341,7 +1341,7 @@ void MigrationDestinationManager::_migrateDriver(OperationContext* outerOpCtx) {
         return;
     }
 
-    setState(DONE);
+    _setState(DONE);
 
     timing.done(6);
     migrateThreadHangAtStep6.pauseWhileSet();
diff --git a/src/mongo/db/s/migration_destination_manager.h b/src/mongo/db/s/migration_destination_manager.h
index b90a56f52f1..5559b75ace9 100644
--- a/src/mongo/db/s/migration_destination_manager.h
+++ b/src/mongo/db/s/migration_destination_manager.h
@@ -85,7 +85,6 @@ public:
     static MigrationDestinationManager* get(OperationContext* opCtx);
 
     State getState() const;
-    void setState(State newState);
 
     /**
      * Checks whether the MigrationDestinationManager is currently handling a migration.
@@ -181,6 +180,8 @@ private:
     void _setStateFail(StringData msg);
     void _setStateFailWarn(StringData msg);
 
+    void _setState(State newState);
+
     /**
      * Thread which drives the migration apply process on the recipient side.
      */