COPS ticket: https://jira.mesosphere.com/browse/COPS-4333
The metrics API occasionally (but consistently) returns 204s on the /containers endpoint, causing a "No metrics found" message on cli `dcos task metrics summary` commands.
It looks like the timestamp on a metric coming from telegraf is not actually set. This timestamp field is used in dcos-metrics to determine the age of the metric, and dcos-metrics is currently configured to expire the cache every 2m. The "janitor" cleanup process is run every minute - because the timestamp is 0, it registers as the 1970 0 timestamp, which comes out to >2m - thus every run of janitor causes the cache to refresh, causing 204s every time this happens.