Troubleshoot workflows

Diagnose missing runs, configuration failures, empty output, and interrupted processing.

Start with one specific workflow and run. Note the selected mode, first unexpected job, input and output counts, and visible error. A focused example is more useful than repeatedly starting a large process.

The workflow is missing or cannot be edited

Clear list filters and search, check the workspace, and confirm access. App-managed recipes can protect their graph intentionally. An app blocking deletion can belong to another member or remain in Trash; its absence from your own Apps list does not remove its reference.

Saving or validation fails

Finish incomplete jobs, correct malformed JSON, and check required fields. Inspect missing dependencies, cycles, named ports, and incompatible record kinds. A trigger restriction can require Manual. A revision conflict requires refreshing and comparing the current definition before retrying the edit.

If autosave says Unsaved, resolve the error and confirm a successful save. A local draft is not proof that the server has the change.

A run does not start

Check enabled state, trigger configuration, source scope, and extension availability. Read the run-trigger error before submitting again. Queued work can be waiting for capacity or budgets. Confirm whether a run already exists so retries do not create unintended parallel work.

For scheduled work, check the effective timezone. For events, verify an actual matching event and use a narrow source boundary rather than trusting a descriptive record-type label alone.

Output is empty or incomplete

Select the correct run, then clear job, type, and text filters. Trace counts from the source forward. Check file selection, JSONPaths, conditions, input limits, incremental baseline, and whether the destination persists records or writes elsewhere. A skipped condition sends empty output but does not automatically skip descendants.

A run fails, times out, or is cancelled

Fix permanent input and configuration errors before retrying. Inspect partial writes and committed chunks. An uncertain model outcome can mean the provider processed a call whose result was not recorded; do not blindly repeat it. For stale results, inspect caching separately from run mode.

Share the workflow, run, job identity, and safe error text with support, excluding credentials and unnecessary document content. See reliability, outputs, and validation.