Executions
Follow workflow runs step by step, stop them, and troubleshoot failures.
Every time a workflow runs, SPARK records an execution. The Executions tab of the workflow section is where you watch progress, confirm success, stop a run, and trace a failure to the step that caused it. The same runs appear in the editor's Executions side-panel tab, filtered to that workflow, and a run an agent starts for you shows its progress beside the thread.
The executions list
Open Ecosystem › Workflows, then the Executions tab. Each row is one run:
- Workflow shows the workflow's name, with the values the run was given under it. A run started by a schedule carries the scheduler's icon before the name; hover it to see which schedule started the run and when it runs.
- An unlabelled column names the step the run is on while it is live.
- Duration, Status, Created and Updated follow. A failed run's Status also gives how far it got, for example Failed 50%.
Use Filter to narrow by Workflow name or Status, the refresh control to read the list again, and the archive toggle to include runs of archived workflows. While a run on the page is live, the list checks for changes every 10 seconds on its own (Checking for changes); click the control to turn that off (Auto-refresh off).
The Status tells you where a run stands:
| Status | Meaning |
|---|---|
| Not started | Created, has not begun |
| In progress | Currently running |
| Succeeded | Finished, every step passed or was skipped |
| Failed | Stopped on an error |
| Paused | Held |
| Awaiting review | Waiting on a person's decision |
| Stopped | Someone stopped it before it finished |
A run that shows Awaiting review is paused on a review step. See Human-in-the-loop for how to clear it.
Inspect a single run
- Click a row, or choose View execution from its ⋯. A run that hasn't started yet has nothing to view.
- The run opens on its own page: the workflow's name and the values it was given in the toolbar, the run's graph on the canvas, and a side panel with the run's summary.
- Progress lists every step in run order as one line, Name · summary. While the run is live it shows Running at the right and the page checks for changes by itself; once the run ends, the run's status icon takes its place. The page opens on the latest result, the last step that ended.
- Click a step's line, or its card on the canvas, to open it. Its card has three tabs: Output, what it produced; Setup, how it was configured for this run; and Meta, its Status, a Reason when it did not succeed, and its Duration.
- Meta, folded at the bottom, holds the run's Last Updated, Duration, Execution ID, Run ID and Workflow ID, each id with a copy button.
A step's Output starts with a line such as 10 of 52 records, with Export and View all at its right. View all lists every row a page at a time; Export saves a CSV of up to the first 10,000 rows, and warns when the output holds more. Large outputs are read a page at a time, so even a step of many thousands of rows opens. Steps that produce nothing say why: Not started yet, Skipped, so it has no output, or Stopped before it finished, so it has no output. For an Ask Agent step, the output also shows the agent's activity and token usage; see Ask an agent.
The toolbar of a run's page has Stop (while the run is going), Rerun, and Edit workflow, which opens the workflow it ran. A step that failed opens on This action failed with the reason, for example Analytics compute request was rejected (503): Service Unavailable.
A run keeps the workflow it ran
When a run starts, it keeps its own copy of the workflow: its name, steps, connections and context variables. Editing the workflow afterwards changes only the runs started after you save. A run already in progress finishes with the steps it started with, Rerun runs again what that run ran, and a run stays readable even if its workflow is later deleted. Runs from before v0.33 still read the live workflow.
Read step statuses
Within a run, each step carries its own status: Pending, Running, Succeeded, Skipped, Omitted, Failed, Error or Stopped. A Failed or Error step is the one that broke, and its output opens on This action failed with the reason.
A Skipped step was not run, for example a disabled action or a step after one; on a succeeded run, skipped steps count as completed.
Stop a run
To end a run that is going, choose Stop from its row's ⋯ in Executions, click Stop on its page, or use the stop control in the header of the run beside a thread. SPARK asks first: Stop execution?, Actions still running end now and show as stopped. Actions that already finished keep their results.
The run then reads Stopped, as do the steps the stop ended. A stopped run is not a failure, and nothing turns it back into one.
Recover from a failure
For a run that is Failed or Stopped, the row's ⋯ (View execution, Open in new tab, Copy ID) adds two options:
- Retry failed actions (on a stopped run, Continue run) carries on in place, in the same execution: from the failed steps, or from the step where the run stopped.
- Rerun creates a new execution and runs every action again from the start, after a Rerun workflow? confirmation. A run of an archived workflow can't be rerun until the workflow is restored.
A run that's Paused or Awaiting review hasn't failed, so it isn't rerun: you move it forward by resolving its review in the Interventions tab. See Human-in-the-loop.
Copy ID copies the execution ID, which is useful when raising a support request. To point someone to a run, paste the address of its page into a thread: it reads as a chip named after the run, and a click opens it beside the thread.
If a scheduled or file-based workflow never appears in Executions, it may not have triggered yet. Check the Trigger States tab: Pending means SPARK is still waiting on required inputs, while Ready, Triggered and Expired show how far the trigger got. For a schedule, check the Schedulers tab too: a schedule whose runs keep failing pauses itself. See Scheduling.
Where to go next
- Clear a paused run in Human-in-the-loop.
- Set up recurring runs in Scheduling.
- Revisit the building blocks in Build a workflow.