Update: after several configuration improvements over the weekend, the error rates have stabilized close to zero, so we're closing the incident.
Some of our users experienced increased error rates, sometimes with the error message upstream connect error or disconnect/reset before headers. reset reason: connection termination, on Thursday May 31st from 5pm to 11pm (UTC). Our engineers applied an emergency fix and the problem went away. However, the root causes of the issue and a final fix for it are still being investigated.
We're sorry for the inconveniences caused. We will keep posting updates as the investigation progresses.