Maintenance windows
A homelab has outages you arranged. The NAS is restarted for a firmware update, the snapshot runs every night at three, the disk gets swapped on a Saturday afternoon. Without anything else, each of those is an incident, an alert and a red number — for something you knew about in advance.
A maintenance window is a period in which a sensor is expected to be down. Inside it:
- no incident is opened, and no alert is sent;
- the checks that fail there do not count against the sensor's uptime;
- the checks keep running, and the results are still recorded.
That last point is the one that makes the first two trustworthy. A window is not a pause: the probe still goes out, and a box that comes back early shows up in the chart as having come back. What a window suppresses is the claim that somebody needs to do something about it — not the measurement.
What it looks like
Two of these cards are inside a window and two are not. The status pill is the same on all four — a box that answers is up, and a window does not change that — and the second pill is the one that says the difference:
The ring is what makes a grid readable without reading it: at a glance, the cards somebody arranged to be quiet are the ones in violet.
Making one
Settings → Monitoring → Maintenance windows, then New window. Anything the installation shows you is read-only if a configuration file owns it.
| Field | What it is |
|---|---|
| Applies to | one sensor, or a whole group — every sensor carrying that group name |
| Repeats | once, between two times; or every week, on chosen days at a chosen time of day |
| Note | what it is for. Optional, and the only place you will remember in six months |
| Active | the switch. A window that is off silences nothing and stays in the list |
The times of a weekly window are read in your time zone, which the form fills in from the browser when you create it. That is deliberate and it is the only place in ProbeGlass that a zone other than UTC appears: a container has no idea where its operator lives, and "Sunday at three" means three on your wall — twice a year and all.
What happens at the edges
Three cases are worth knowing about, because each of them has an answer that is not obvious.
A window begins while an outage is already open. The outage is closed, with a note saying why, and nothing is sent about it — the message that opened it already went out. Leaving it open would mean the dashboard showing an outage nobody is being told about, for as long as the window lasts and then some.
The window ends and the box is still away. None of what failed inside the window is evidence about now, so the sensor starts again from the first failure outside it: it has to use up its retries before it is called down, exactly as it would for any other outage. The new incident is dated from after the window, not from before it.
The sensor comes back inside the window. Nothing is sent. No "down" ever went out, so an "up" would be a message about a problem you were never told about.
What the dashboard shows
The sensor's status is whatever the probe found — if it is not answering, the card is red, because that is true. What the window changes is that the card says the downtime is expected rather than broken, and the uptime figure on it leaves the window's checks out.
A window that is running right now is marked in the list, which is the thing to check when a sensor goes quiet and you want to know whether you arranged it.
Two things it does not do
A window saves nothing for later. It starts applying the moment you save it. The checks already recorded are not rewritten, so a window declared for an outage that already happened leaves the numbers for that morning as they were. A window is a statement about the future, and it is honest about being one.
It is not a pause. Pausing a sensor stops the checks altogether and the sensor reads paused; a window lets them run and only silences what they would have said. If you want the probes to stop, pause the sensor instead.
In the configuration file
The windows are made in the dashboard. A maintenance: section in
probeglass.yaml is on the roadmap and not in 0.2.0 — an
installation driven entirely from a file has to add them through the interface for
now.
See also
- Incidents — what a window is not, and how acknowledging one differs
- History and uptime — what the figures mean once the window's checks have been taken out of them
- The dashboard — pausing a sensor, which is the other thing people reach for