Using the workspace

Watch: keeping your live systems healthy

Watch checks your live apps and servers, opens one incident per problem, fixes what you have allowed, asks you about the rest, and confirms the recovery from outside.

Watch is the part of Duscode that looks after software once it's live. It checks your systems regularly, notices when something breaks, and either fixes it within the limits you set or asks you one plain question.

Warning

Watch runs inside Duscode on your computer. It checks your systems only while Duscode is running; while your computer is asleep or switched off, nothing is checked. Each such gap is recorded and listed in your daily brief, so you always know when nobody was watching.

Coming soon

Watching while your computer is switched off, and a Watch view in the workspace where you add systems, choose access levels, write standing orders and set maintenance windows. Today Watch looks after the apps Dus deployed, you hear from it through notifications and the daily brief, and the settings below are available to Duscode's local API.

What it watches

Apps Dus deployed are watched automatically, one system per deploy target. A system Duscode didn't build can be added by its address; what Dus can't find out, he lists as questions for you ("Where does it run?").

Each system has checks, run from outside the server:

• The address answers, quickly and without server errors.
• The certificate isn't about to expire (a warning 14 days before).
• The domain still points where it should.
• The last deploy didn't fail.
• The server has disk, memory and headroom (read-only, over the deploy key).
• The app's log shows no known error patterns in the last 10 minutes.

A check that can't be taken (your computer is offline, say) is reported as "Can't check right now", never as "down".

What happens when something breaks

1. Detected. Two bad readings in a row open an incident. One problem opens one incident, not a flood of alerts, and a check that keeps flipping waits until it settles.
2. Diagnosed. Dus looks at what changed (a recent release, the logs) and says what he thinks is wrong, in plain words.
3. Fixed, or asked. If you've allowed the fix in advance (a standing order such as "restart it if it stops answering"), Dus does it. Otherwise he asks you, with options and a recommendation.
4. Verified. Recovery counts only when the checks pass again from outside, two readings in a row.
5. Stops if a fix fails. A failed fix is never retried blindly: Dus stops, tells you what happened, and asks again without that option.

What Dus may do: access levels

You choose one level per system. New systems start at Observe.

Lowering the level takes effect at once: standing orders that need a higher level stop applying.

Maintenance windows

Planned work shouldn't page you. Set a maintenance window (up to 7 days) and Watch opens no new incidents for that system during it.

Related

• Duties
• Deploy
• Notifications
• When Dus needs you