To simplify coordination across multiple nodes, Atlas wish to be able to specify a start time, some small window into the future. This then relies upon the clock drift between nodes being small, which is assumed to be the case.

Additionally, to minimise the risk of customer cluster impact, we do not want recording to continue without end in the event that a component in Atlas (e.g.,) crashes or restarts. To minimise the state Atlas needs to track in such events, providing an end time is also useful. In the worst case, the recording continues for the expected amount of time, or halts early if disk usage grows too high (if Atlas is not shipping recording files away).