Guide
How to monitor and debug an automated workflow
A failed run is a record, not a retry queue. Open it, name the step, and decide whether the fix is a connection, an instruction, or a search that was too wide.
Start with the version
The run is pinned to the version that was in production when it started. If you have since deployed a fix, this run will still show the old steps. That is what you want when you are explaining yesterday’s mail.
Name the step
A waiting approval is not an outage. The send, the invite, or the delete is paused. A failed step is different: the tool returned an error, often a disconnected account or a bad id. The product does not replay that step on its own.
Fix, test, then deploy
Reconnect the account or edit the draft. Use a test run on the latest version. When the test matches what you wanted, deploy that version. New runs use it. The old failed run stays as history.
Questions
Where are the retries?
There isn’t an automatic retry loop in the executor. Debugging means reading the run and running again after the fix. An article that promised retries would be describing a feature the product does not have.