From a database perspective, each TigerBeetle request is a single transaction: an ordered group of micro-operations which execute atomically. Each TigerBeetle node has a single data file, which is divided into zones at predictable offsets. The session documentation reaffirmed this stance: a TigerBeetle client “will never time out” and “does not surface network errors”. Because TigerBeetle clients handle all failures through a silent internal retry mechanism, they unnecessarily convert definite errors into indefinite ones. Long-lasting unavailability could cause TigerBeetle clients to consume unbounded memory as they attempt to buffer and retry an ever-growing set of requests.