This broke when changes were made to allow processing data more recent than
the last top-of-the-minute (~8 months ago). The only "updates" that worked
since then were for 1-second data. It's a little bit lazy because it relies
on the output_factory to return traces with proper starttimes and endtimes
(that is, they fall on the tops of minutes). Probably we should be more
explicit in correcting the starttimes and endtimes to line up on expected
timestamps. Pushing this fix now because it has operational impacts.