We found that a bug in a newly added feature caused certain interruptions, impacting other running requests and ultimately failing them.