SERVER-48518 Fix rollback via refetch anomaly

5 files changed, 16 insertions, 8 deletions

diff --git a/src/mongo/db/repl/replication_coordinator_impl.cpp b/src/mongo/db/repl/replication_coordinator_impl.cpp
index 0140ba94d06..83fe6464888 100644
--- a/src/mongo/db/repl/replication_coordinator_impl.cpp
+++ b/src/mongo/db/repl/replication_coordinator_impl.cpp
@@ -4991,6 +4991,22 @@ void ReplicationCoordinatorImpl::finishRecoveryIfEligible(OperationContext* opCt
         return;
     }
 
+    // Rolling back with eMRC false, we set initialDataTimestamp to max(local oplog top, source's
+    // oplog top), then rollback via refetch. Data is inconsistent until lastApplied >=
+    // initialDataTimestamp.
+    auto initialTs = opCtx->getServiceContext()->getStorageEngine()->getInitialDataTimestamp();
+    if (lastApplied.getTimestamp() < initialTs) {
+        invariant(!serverGlobalParams.enableMajorityReadConcern);
+        LOGV2_DEBUG(4851800,
+                    2,
+                    "We cannot transition to SECONDARY state because our 'lastApplied' optime is "
+                    "less than the initial data timestamp and enableMajorityReadConcern = false",
+                    "minValid"_attr = minValid,
+                    "lastApplied"_attr = lastApplied,
+                    "initialDataTimestamp"_attr = initialTs);
+        return;
+    }
+
     // Execute the transition to SECONDARY.
     auto status = setFollowerMode(MemberState::RS_SECONDARY);
     if (!status.isOK()) {
diff --git a/src/mongo/db/storage/storage_engine_impl.cpp b/src/mongo/db/storage/storage_engine_impl.cpp
index 37264db0224..45bf7d03dd2 100644
--- a/src/mongo/db/storage/storage_engine_impl.cpp
+++ b/src/mongo/db/storage/storage_engine_impl.cpp
@@ -852,7 +852,6 @@ void StorageEngineImpl::setStableTimestamp(Timestamp stableTimestamp, bool force
 }
 
 void StorageEngineImpl::setInitialDataTimestamp(Timestamp initialDataTimestamp) {
-    _initialDataTimestamp = initialDataTimestamp;
     _engine->setInitialDataTimestamp(initialDataTimestamp);
 }
 
diff --git a/src/mongo/db/storage/storage_engine_impl.h b/src/mongo/db/storage/storage_engine_impl.h
index 316a02c3042..a00ff2b2369 100644
--- a/src/mongo/db/storage/storage_engine_impl.h
+++ b/src/mongo/db/storage/storage_engine_impl.h
@@ -404,7 +404,6 @@ private:
 
     const bool _supportsDocLocking;
     const bool _supportsCappedCollections;
-    Timestamp _initialDataTimestamp = Timestamp::kAllowUnstableCheckpointsSentinel;
 
     std::unique_ptr<RecordStore> _catalogRecordStore;
     std::unique_ptr<DurableCatalogImpl> _catalog;
diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine.cpp b/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine.cpp
index d59a0e8cd8b..476f1cd6ccb 100644
--- a/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine.cpp
+++ b/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine.cpp
@@ -2415,10 +2415,6 @@ Timestamp WiredTigerKVEngine::getCheckpointTimestamp() const {
     return Timestamp(_getCheckpointTimestamp());
 }
 
-Timestamp WiredTigerKVEngine::getInitialDataTimestamp() const {
-    return Timestamp(_initialDataTimestamp.load());
-}
-
 std::uint64_t WiredTigerKVEngine::_getCheckpointTimestamp() const {
     char buf[(2 * 8 /*bytes in hex*/) + 1 /*nul terminator*/];
     invariantWTOK(_conn->query_timestamp(_conn, buf, "get=last_checkpoint"));
diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine.h b/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine.h
index df0dccc4373..09152ec80fe 100644
--- a/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine.h
+++ b/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine.h
@@ -323,8 +323,6 @@ public:
     Timestamp getOldestTimestamp() const override;
     Timestamp getCheckpointTimestamp() const override;
 
-    Timestamp getInitialDataTimestamp() const;
-
     /**
      * Returns the data file path associated with an ident on disk. Returns boost::none if the data
      * file can not be found. This will attempt to locate a file even if the storage engine's own