cubeship
Apps

Metrics

What each copy is using, sampled every 30 seconds and kept for a day. 100 is one core.

The app's Overview tab charts CPU and memory for the last hour, six hours or a day.

  • Sampled every 30 seconds, kept for a day. No downsampling behind that; what a day buys is the question people ask, which is what happened overnight.
  • 100% is one core. 250 is two and a half. Not rescaled to the machine, so a busy container looks busy whatever the host's size. (The machine's own charts on the Overview are the other way round: there 100% is the whole box, and each chart says which.)
  • Memory is usage minus reclaimable page cache, what docker stats shows. Without the subtraction every container looks about to run out.

An app with several copies is charted as the average across them, and the peak shown is the busiest copy's. Each copy records against the app from whichever machine it is on.

The reading is taken by the daemon that can reach the container's cgroup, so a copy on another machine is sampled there and reported on its poll — on the same scale, into the same table.

cubeship app get shop/api      # the newest reading is on the app

The instance_metrics and instance_containers tools give an agent the machine and the heaviest containers on it; an app's own series is GET /apps/{ref}/metrics.

Logs

The container's own output, per copy, coloured the way the program wrote it.

Placement

Which machines an app runs on and how many copies. One machine is the ordinary case; a second is a row, not a different kind of app.