The status page on Atheron Progress answers one question: are our public services up. It answers it for today, and it answers it for the last ninety days. This article explains how to read the strip of coloured days under each service, so you know what you are looking at before you draw a conclusion from it.
One row per service, one square per day
Each public service gets its own row. Under the name you see ninety squares in a line. The square on the right is today. The square on the left is ninety days ago. When a new day starts, the oldest day drops off and a new one appears.
Ninety days is a deliberate choice. It is long enough that a single bad week cannot hide, and short enough that the strip still describes the service you are using now rather than the one we ran last year.
A day is coloured by its worst event
This is the rule that matters most, so we will state it plainly. Each day takes the colour of the worst thing that happened on that day. Not the average. Not the most common state. The worst.
If a service was fine for twenty three hours and down for one, that day shows as a down day. If a service was fine all day except for a short stretch of slowness, that day shows as a slow day. A square only shows the clean colour when nothing worse than normal operation was recorded at all.
We chose the worst event rather than an average because an average hides the thing you actually care about. You do not want to know that a service was 96 percent fine on Tuesday. You want to know whether Tuesday had a problem, and what kind.
The kinds of event you will see
The status page distinguishes between the kinds of event that can set a day's colour. Here is what each one means for you.
| Kind of day | What was recorded | What it means for you |
|---|---|---|
| Clean | No event worse than normal operation | The service did what it should, all day |
| Slow | The service answered, but slower than it should | You could use it, but you may have waited |
| Maintenance | We took the service down or limited it on purpose | Planned work, not a failure |
| Down | The service did not answer | You could not use it for some part of the day |
Slow and maintenance are worth a closer look, because they are the two that people most often misread.
Slowness is its own colour
A slow service is not an up service, and it is not a down service either. It is a service that answered, but took longer than it should have. We record slowness separately so that it does not disappear into a clean day and does not get counted as an outage it was not.
If you see a run of slow days on a service, that tells you something different from a run of down days. The service was working. It was not working well. Both are worth knowing, and they are not the same thing.
Maintenance is not a failure
Sometimes we take a service down ourselves, or limit what it can do, to deploy a change or do work that cannot be done safely while it runs. Those days are marked as maintenance, not as outages.
The reason for the separate mark is simple. An outage is something that happened to us. Maintenance is something we did. If we let the two share a colour, the strip would tell you that a service is unreliable when the truth is that we were working on it. If we hid maintenance entirely, the strip would tell you the service was fine when you could not reach it. Neither is honest, so maintenance gets its own mark.
The worst event rule still applies. If a day had planned maintenance and also an unplanned outage, that day shows as down, because down is worse than maintenance.
How to read a row quickly
Here is the way we read it ourselves.
- Look at the right hand end first. That is now. If today is clean, the service is working as you read this.
- Scan left for clusters. Several bad squares in a row mean something was wrong for a stretch. A lone bad square means one event on one day.
- Check the kind. A cluster of maintenance days is planned work. A cluster of slow days is a performance problem. A cluster of down days is an availability problem.
- Cross check with the changelog. If a square is bad, the changelog on the same site is the record of what we did about it.
What the strip does not tell you
The strip tells you what kind of event was the worst on a given day. It does not tell you how long that event lasted, and it does not tell you how many people it affected. A one minute outage and a six hour outage both colour a day as down. If you need the duration, the colour alone is not enough.
It also only covers public services. Work that is still building, or marked next or later on the project tracker, has no status row because there is nothing public to be up or down yet. The status page and the project tracker are two different views, and we keep them separate on purpose.
The colour of a day is a flag, not a verdict. It tells you where to look. The changelog tells you what you will find.
We would rather you read the strip and ask us a hard question than glance at it and assume. That is why it is public, and why the rule for colouring it is written down here.
