Health
Hear about a quiet workflow before your client does.
A step that fails throws an error. A workflow that has stopped being started throws nothing, and the client is usually the first to notice.
Thrum learns how often each workflow normally runs, and emails you when one goes quiet, keeps failing or slows down.
Thrum opens soon. Sign up now and you’ll be among the first in.
The rhythm
It learns each workflow’s rhythm, then watches for the gap.
A lead form that fires a dozen times a day and a weekly report have very different ideas of normal. Health learns each one’s usual gap between runs, and calls it Stopped when the silence runs to three times that.
A floor and a ceiling keep it sensible: a busy workflow doesn’t page you over a quiet lunchtime, and a slow one is still checked.
Four words, and they mean the same thing on every screen.
- Thrumming
- Running at its normal rhythm. Nothing to do.
- Degraded
- Still running, but failing or slowing. Worth a look today.
- Stopped
- Not running. Your client may already have noticed.
- Idle
- Nothing has come in yet, or you paused it. Not a problem.
An example: Lead Rescue for a plumbing client usually runs every 4 hours. It stops running and nothing fails. It stays Thrumming until 12 hours of silence, then turns Stopped, and the operator gets an email. The client is not told.
The alert
One email when it breaks, one a day until it’s fixed, one when it’s back.
The alert names the client and the workflow, says how long it has been quiet and why that counts, and links straight to it.
While it stays broken you get one reminder a day, not one per check. When it runs again you hear that too, but only if the bad news went out first. Health emails go to you, never to your client.
Failing and slow
Running, but not well, is caught too.
A workflow is Degraded when its last few runs failed in a row, or a share of its recent runs failed. Failures count from the very first run, so a workflow that is broken from day one is still caught.
It is also Degraded when it runs far slower than usual. Slowness is never urgent, so it goes into one daily digest instead of an alert.
Keys, limits and listening
The quiet failures, named, with the fix on the alert.
A client’s key refused by a service, a service saying “too many requests” through every retry, or an app that stopped sending the events that start the workflow. Each has its own plain sentence.
Workflows started by an app’s webhook are checked daily. When the webhook is gone or switched off, Reconnect sets it up again in one press. App triggers are in beta.
Clients
The client in trouble is always at the top.
A client is as healthy as its worst workflow, so anything Stopped floats to the top, with runs this month and a 30-day chart beside it.
One level down, filter every workflow by client, by state or by recent failures, and bookmark the view.
Step numbers
See what every step did, added up.
Runs, failures, retries that worked, median and slowest times, the way each branch went, how long each wait really waited, and the errors seen most.
Over a week, a month, a quarter or a year, from a daily count that keeps no lead’s data.
Specs
Health thresholds
Fixed, and the same for every workflow.
- Runs before it knows the rhythmFailures are checked from the first run
- 10
- Window it learns fromOr the latest 20 runs, when the window holds fewer
- 30 days
- Stopped after
- 3 × the usual gap
- Shortest threshold
- 6 hours
- Longest threshold
- 7 days
- Degraded: failures in a row
- 3
- Degraded: share of recent runs failedOnce at least 5 have finished
- 20% of the last 20
- SlowOnce at least 10 runs are in; daily digest only
- 3 × the usual duration
- Reminder while it stays broken
- Once a day
- Attempts at a failing stepWaits double from 5 seconds, capped at 10 minutes
- 3
- Step numbers
- 7, 30, 90 or 365 days
Common questions
Does my client get told when something breaks?
No. Your client sees their leads and their month. If a text or email to a lead could not be sent, that lead says so in plain words.
What about a new workflow that has not run 10 times yet?
Failures are caught from the first run. Only the silence check waits for 10 runs, and you can start it straight away by saying how often the workflow runs at least.
Can I pause a workflow without getting an alert?
Yes. A paused workflow is Idle, never alerts, and does not count against its client. Switching it back on starts the silence clock again, so a paused fortnight is not a Stopped workflow.
What happens to a step that fails?
It is tried 3 times, waiting 5 seconds and then 10, and only when trying again can help: a server error, a rate limit, a timeout. Fix the cause, then retry the run from the step that failed, or retry every failed run of a client at once from the Overview.
Thrum is cooking.
Get early access and you’ll be among the first in. Early members get a launch perk.