Debug A Flow Run
Read the node-by-node trace for a request that went through a flow, and tell flow behavior apart from plugin or provider behavior.
Debug A Flow Run
Use this when a request behaved unexpectedly and an organisation or key flow could have affected it.
Find the request in Usage Records and open its record detail page.
Confirm the served model, status, and latency.
Open the flow trace on the record.
Each flow run records node id, node type, outcome, and duration, grouped by scope and phase. It appears beside the plugin execution summary.
Compare the trace to the graph in Flow Studio.
Open the flow that ran and match trace node ids to the canvas. Skipped usually means a branch went another way. Errored means a runtime problem, not a deliberate abort.
Check whether the outcome was an abort or a node error.
| Trace shows | Likely cause |
|---|---|
| An Abort node, or an HTTP Request with strict-fail behavior, in the trace | The flow deliberately blocked the request — check the condition that led there. |
| A node marked errored, request continued | A node failed but on_error is set to continue — expected for best-effort steps like optional notifications. |
| A node marked errored, request stopped with a 500 | on_error is set to block for this flow — the failing node was treated as load-bearing. |
| No flow trace at all, even though a flow is activated | The flow's canvas for that phase was empty (only start/end), the flow was disabled, or an earlier sequential-lane step aborted the request first. |
If the issue looks like ordering rather than a single node, re-check Execution Order.
A step that did not run is often in a later layer blocked by an earlier sequential abort.
Cross-Checking With Traces And Logs
If the observability stack is enabled, the same request id links the flow trace to:
- Traffic Analytics: Plugins for aggregate plugin and flow time across many requests.
- LGTM Stack: Investigate one request to follow the same request into traces and logs if you need lower-level detail than the trace summary provides.
Before Saving: Catch It Earlier
Most graph mistakes, such as missing edges, impossible conditions, or undefined variables, are caught before production. See Testing & Verification.