I noticed this while doing some performance profiling - a histogram was causing a ton of Sprintfs via Error.Errorf. I dug in and found out that the Durations histogram that's part of TxnMetrics always gets updated with a preposterously large number, which gets capped at 10 seconds. I'm not sure why this happens, probably the txnStartNanos is set to 0 or something.

Here's a screenshot of the UI which agrees with my finding: