Nightly check that opens a job
The situation
Something needs checking every night: the website should be up, yesterday's takings should reconcile, stock should match the system. Most nights it is fine and you would rather not hear about it. On the night it is not, you want the work started before you wake and an account of it at breakfast.
Start simple
If a message is all you want, no flow is needed. Ask for a watcher.
"Check our website every night at two, London time, and message me on Telegram if it is down or slower than five seconds."
That is one agent running one instruction on a clock, set up in minutes, and before it arms it sends a message on your channel to prove the alert path works. See Scheduled tasks and watchers and Schedule something. Go further only when finding the problem should start the fixing.
What you need
A flow, set up with your assistant. Flows are early access, so give this one a few supervised nights first. Write your first flow walks it through.
A connected channel, so the report reaches your phone.
Your timezone, said out loud: your installation keeps its own clock, and two in the morning without one can land hours away.
The conversation
Describe the routine as you would to a new member of staff.
"Every night at two, London time, check that our website answers in under five seconds, that yesterday's card takings match what the bank settled, and that no stock has gone negative. If all three are clean, tell me nothing. If any is wrong, open a job: the team works out why and fixes what can safely be fixed overnight, then messages me on Telegram with what was wrong and what still needs me."
Your assistant asks what a new starter would ask. Does an error page count as down? Is a penny out a problem, or a pound? What may the team fix on its own? It writes the routine back in your words, then compiles it in the background.
Test it on a night you know is clean and a case you know is broken. Every correction is kept beside the process, so the next build honours it. The flow then sits on Automations > Flows (/manage/flows) with its nightly trigger on its row.
What happens at night
At two the trigger fires. On a clean night the run checks the three things, finds nothing, and finishes without telling you.
On a bad night it reaches the point where your description says a problem is a problem, and hands the heavy work over. It opens a job: your brief, a coordinator and a team. The coordinator splits it into tasks and picks a specialist for each, one to work the two sets of figures against each other, one to write up what it found. The flow waits, suspended rather than finished.
When the job is done the flow picks up from the step it stopped at and sends your message.
Nightly check, Tuesday, 02:04. Card takings did not reconcile: 4,182.60 taken against 4,061.10 settled. Three Sunday refunds had never been posted, and are posted now. Website answered in 0.9 seconds. Stock clean. One for you: the Redgate line still shows negative stock; nobody entered the delivery note.
Two places to look in the morning: the run on Automations > Flows (/manage/flows), drawn as a map with every step and decision, and the job on Automations > Jobs (/jobs), with its plan, board and files.
Make it yours
Ask before anything is changed. Say so when you describe the routine, and the run stops before the fix and puts the question on your channel. Answer in ordinary words and it carries on.
Make it weekly. A reconciliation that only matters before Monday's meeting does not need every night, and costs less.
Check more things in one run. Backups finished, certificates not near expiry, no order stuck unshipped: ten checks in one run cost less than ten runs.
If it did not run
The Runs tab on Automations > Flows (/manage/flows) is the history, with last night near the top. If there is no run at all, check the flow's row in Library: each trigger has a switch, and a Disable all button stops every trigger at once. You can fire one as a test instead of waiting. If it ran at the wrong hour, suspect the timezone.
If the run is still suspended long after the night, look at the job it opened. A job that seems frozen is sometimes only waiting for capacity from your AI provider, and picks itself up. One that could not sort itself out shows Needs attention until you answer in the chat it is talking to, and where you are reached is set on Admin > Alerts (/admin/alerts). A run that was in the middle of its checks when your installation restarted is not resumed, so you wait for tonight; one that was only waiting on the job carries on.
Related
- Flows and Jobs, the two parts this joins.
- Create your first job, for what a team does with a brief.
- Set up a trigger, for the other ways a flow starts.
- A morning briefing that talks to you, the version that only tells you.