In NewOplogFetcher::_connect, if connect/auth fails, it depends on _oplogFetcherRestartDecision->shouldContinue to determine whether it should retry the connection. During initial sync, shouldContinue always return true for network errors. This means, the while loop in _connect() is effectively an infinite loop during initial sync. This is fine but it should honor the shutdown signal. And it would be better to backoff a bit on retries.