self-hosted error monitor Open the panel

Dashboards

the ones you build, and the ones that come ready open the screen

What it is for

Panels built over the same occurrence data. They exist for the question that comes back every week and that nobody wants to rebuild in the filter: "is the new release worse?", "where does the code hurt?", "are we getting better?".

The ones that come ready

Thirteen, and not all of them appear in the menu: Insights, Applications, On call, Code, Is this customer suffering?, Is the new release worse?, Regressions, Production versus staging, When it breaks, Where the code hurts, Are we getting better?, Machines and users and Where the error comes from. The ones outside the menu open from the dashboard list.

How to build one

A panel is the combination of four choices:

  1. The shape: columns, horizontal bars, lines, stacked areas, a big number, or a table.
  2. The measure: occurrences, distinct errors, distinct customers, or distinct versions.
  3. Occurrences per 1,000 usage events: the rate instead of the count, for applications that send usage events. Ten errors at a customer that used the system all day are not ten errors at one that opened a single screen. By day, by hour, by customer, user, version or machine, each bucket is divided by its own usage; by error class or code, by the usage of the whole period.
  4. The dimension: class, application, customer, version, origin in the code, environment, platform, machine, user, or hour of day.
  5. The slice: any filter, written as an expression. A production only panel is a panel with ambiente:producao saved inside it.
What meulog decides for you

Each panel's result is cached for five minutes, and the key carries the highest occurrence id: the cache invalidates itself when something comes in, not when the clock says so. A dashboard over half a million occurrences opens in a fraction of what it used to, without showing stale numbers.

What you should see

The dashboard's panels, in the order they were placed. A dashboard with no panels shows the invitation to create the first one. A panel with no data in the period says so, instead of drawing an empty chart.

Sending a dashboard on a schedule

In the dashboard's settings, "Send a summary" sends it as text every day at 8am (covering yesterday) or every Monday at 8am (covering the last seven days). Each number goes with its change against the previous period, each category chart with its top three, and a link to the screen. It goes out as a summary, the same event as the daily summary, so it reaches whatever Google Chat space, e-mail or channel your rules already send summaries to.

Lines on the daily charts

Daily line and area charts carry two kinds of dashed line: blue for each deploy in the period, with the version on top, and red for each outage of a watched URL, with its name. An outage only counts when it reached the number of failed checks that sends a warning; a single blip is not drawn.