This makes is pretty much useless when comparing telemetry events.

The ISO string returned from this utility is only accurate to the second.

And all start times should use microtime();

Actual Behavior

Start times are truncated at seconds.

Expected Behavior

All times should be accurate to at least milliseconds.

Workaround

We've had to implement a workaround right before flushing. We store the start_time in microseconds, then rewrite the start time right before flushing.