Starting at 06:35 UTC on Tuesday, 4 October 2016, we were unable to receive build requests on travis-ci.com for approximately 50 minutes.
That morning, GitHub experienced a service outage. Once the issue at GitHub was rectified, our system was flooded with build requests at a much greater rate than usual. This caused our Redis instances to hit their maximum allocated memory usage. Our system recovered quickly, but sadly, around 4000 build requests were irretrievably dropped.
We are sorry for the difficulties this will have caused. We are already taking steps to avoid this in the future, including splitting our Redis usage across a greater number of dedicated instances and storing GitHub API events redundantly, in order to replay them in the event of an outage.