On node startup we call recoverFromOplogUpTo that will try to apply all the oplog entries up to the recoverToOplogTimestamp on top of the last stable checkpoint. This is done by calling _applyOplogOperations in RecoveryMode::kStartupFromStableTimestamp . Only after this first recovery phase has been completed RepllcaSetAwareServiceRegistry::onStartup() will be executed.

All the recoverable critical sections stored on disk are re-installed in-memory as part of the onStartup procedure, so up until that moment we should skip any recoverable CS opObserver actions.