Service status

Each state above is read by ewpire from its own records when the page is generated: whether the database answers, whether background jobs start on time, whether a model of each kind is switched on, and how many of the requests to models in the last 15 minutes failed on the model's side.

The reading of background jobs covers the last 24 hours: they are shown as starting late when, in that time, a job came due and has waited more than 120 seconds to start, or a run stopped and its job has waited that long to be picked up again. A job waiting to start while another job of the same workspace is running is not counted.

Models of a kind are shown as degraded only when at least 5 requests were made to them in the last 15 minutes, more than 20% of those requests failed on the model's side, and the failures were in more than one workspace. So failures in a single workspace, or among fewer than 5 requests, do not show here, and Working on a row does not mean that every request succeeded.

The page calls no model to test it. A model that has stopped answering shows here only after real requests to it begin to fail.

The page is generated again at most once every 60 seconds, and it is served by the same web app it reports on. If the time at the top is more than a few minutes old, or the page does not open, the web app itself may be the part that is not working.

There is no uptime percentage and no history of incidents here, because the page has no record to take them from. It is information about the service, not a service level commitment: what applies to availability is in the Terms of Service.