See #453 for a real problem of metrics that was down for at least two weeks.
Problem is that we don't receive a signal when metrics is down and as a result, can not automatically link it to other chain of events, such as library upgrade, new version deployment etc.
