Automation troubleshooting

Start with Problems and Run history. Problems explains why the draft cannot run or publish; Run history shows what a real execution received, resolved, returned, retried, or failed.

The workflow opens away from its content

Newly opened workflows should centre their content at 100% and keep the trigger visible. Choose 100% to reset zoom. Choose Overview to locate a distant branch, then return to 100% for configuration. The minimap reveals nodes placed far from the main graph.

A change does not appear saved

Draft changes autosave. Check the save-state indicator:

• Saving means the draft is still being written.
• Saved means the current edit is stored.
• A retry/error state means the write failed; keep the page open, restore connectivity, and retry.

Before leaving after an error, confirm the indicator returns to Saved. Save & Run also flushes the current draft before starting a test.

Save & Run or Publish is disabled

Open Problems and resolve issues from the trigger forward. Common blockers are a missing trigger resource, required action field, recipient, same-service item, managed credential, document input binding, or unconnected required branch.

Some actions are shown only when their owning module is enabled for the service.

The trigger does not start a run

Check that:

1. the workflow is published;
2. the event belongs to the same service;
3. selected project, form, calendar, resource, item, lifecycle moment, and filters match;
4. the schedule timezone and next-run preview are correct;
5. the webhook signature or HTTP credential is valid;
6. the source actually emits the documented payload.

Use synthetic test data to prove the downstream graph separately from the live trigger.

A schedule runs at the wrong time

Verify Timezone, frequency or cron expression, and the next-run preview. Cron fields run in the selected timezone, not automatically in the viewer’s current location. Review daylight-saving boundaries where the selected region observes them.

A webhook is rejected

Confirm HMAC is enabled, the expected algorithm and signature header match the sender, and both sides sign the same raw request bytes. Re-select the managed credential after rotation. Do not compensate by disabling verification in production.

HTTP Poll never detects a change

Use Poll now and inspect the returned fragment. Confirm URL, managed credential, headers, interval, and JSONPath. Select a small stable value that changes when work is required; a volatile timestamp can cause constant runs, while an unchanged container can cause none.

An output is missing from the expression picker

The source action must be upstream and connected. Test it successfully so the designer can learn its output shape. Give the action a unique label, reopen the downstream configuration, and select the value rather than typing the reference.

For dynamic objects, you can still enter a known path, but verify it against Run history before publishing.

An action tests successfully but changes nothing

The default test is a dry run. It reports what a side-effecting action would do without sending, creating, updating, deleting, or calling for real. Enable Run for real, provide trigger test data if requested, read the confirmation, and use a non-production destination.

A workflow is stuck in waiting

Wait Until and Human Tasks intentionally pause a run. Check the target time, timezone, assignee, task status, SLA, timeout policy, and escalation target. Unpublishing prevents new trigger runs but does not automatically cancel an already waiting run.

A branch does not run

Inspect the Conditional or Switch result in Run history. Confirm the source expression resolves to the expected type and value, and that the named output port is connected. Connect a default/fallback path for values not known when the workflow was authored.

Parallel Join never continues

Every branch required by its wait rule must reach the join or an explicit handled terminal state. Confirm branch names, incoming connections, action timeouts, error paths, and the join timeout. Do not make one parallel branch depend on a sibling result.

A loop or retry creates too much work

Unpublish the workflow to stop new trigger-driven runs. Set Max iterations, a real break condition, bounded retry attempts, and increasing delay. Check whether a downstream write re-triggers the same workflow. Add idempotency keys or origin filters before republishing.

Email or notifications do not arrive

Confirm sender configuration, recipients, team membership, expression results, template variables, and attachment outputs. Inspect the action result in Run history. Test with a controlled recipient and check spam/quarantine or in-app notification settings.

A DataHub write affects the wrong rows

Stop live testing. Review source, table, and every filter. Update and Delete should normally use a primary key or equally selective identifier. Test with a non-production table and confirm the resolved filter in Run history before Run for real.

Imported workflows show missing resources

Exports preserve workflow structure but environment-specific IDs may not exist in the destination. Re-select forms, DataHub sources and tables, projects, boards, calendars, workspaces, documents, email configurations, AI registrations, recipients, and managed credentials. Then rebuild Test data and run the draft again.

Run history contains sensitive data

Unpublish the workflow if new runs could repeat the exposure. Remove secrets and unnecessary personal data from Test data, Workflow Log messages, prompts, templates, and error text. Replace inline credentials with managed credentials. Follow the organisation’s data-handling and incident process for existing records.

Before asking for support

Collect the workflow ID, run ID, failing node label, time and timezone, trigger type, terminal status, and redacted error. Export the draft only if it contains no secret or customer-sensitive values. Never send managed credential values, authorization headers, full customer payloads, or private form answers.