Any time a sprint commitment must be broken the decision rests on the team's product manager who is best (or _should_ be best) suited to understanding the tradeoff between sprint completion and a potential emergency bug situation. If the PM decides it's not worth it, then that's on them. It's worth noting, you may want to include a wide enough distribution so it's known that you noticed but PM didn't want you to work on it in order to cover yourself.

As for how to account for that work in your velocity, I don't believe it is realistic to size a story for fixing a potential issue (not to mention the difficulties in sizing bugs anyway [1]). However, it is possible (or at least a bit easier) to size a story for research that answers a specific question as its acceptance criteria -- in this case, the acceptance criteria for this article's situation would've been something like 1) "will latency continue to rise?" and 2) if so and it is unacceptable, what is the cause and add an implementation story to the backlog"

Some may say "well, how do you know how what is causing the issue", and if no one really can figure it out the answer to the research story question and the bug has manifested itself as an emergency, then sure your velocity will tank as you break the sprint(s), but it's up to the team/business to understand how to remove outliers for an accurate velocity.

[1]: http://www.agileforall.com/2010/05/agile-antipattern-sizing-...