Error Handling & Retries

Configure automatic retries and fallback error paths in a Step’s Test & Debug tab. Both controls are off by default, so an unconfigured Step stops the run on its first failure. To inspect a past failure, open Activity and check Step-Level Executions.

At a glanceVerified in app · Oct 30, 2026

Use it to: Retry transient failures automatically, branch on Step errors, or trigger a workspace alert workflow
Find it: Step drawer → Test & Debug → Retry on fail / If this step fails, or Tools → Start → Error Trigger
Cost: Free to configure · Chargeable failed attempts and retries use Step credits
Limits: Default retry: maxRetries 3, retryDelay 5 s (up to 10 retries and 60 s delay)

Retry on Fail

Enable Retry on fail in Test & Debug for operations that are safe to repeat.

Check the destination before retrying a write. A timeout, 429, or 5xx status does not prove that a write failed. Enable retries on write actions or HTTP Request Steps only when repeating the call is safe.

Setting	Default	Range
Retry on fail	Off	On / Off
Max retries	3	`0` to `10` retries (`maxRetries: 3` allows up to 4 total attempts)
Wait between tries	5 s	`0` to `60` seconds base delay (`retryDelay: 5`), plus jitter and `Retry-After` waits

Initial runWait 5s + jitterRetry 1Retry 2Retry 3Error Path

Each retry appears in Executions as Attempt 2, Attempt 3, and so on. Each attempt records its own timestamp and error message.

When automatic retries are skipped

Glow retries only transient HTTP statuses, connection drops, or explicitly retryable errors. Isolated Step tests skip automation retry policies (Testing & Debugging). Run for each item Steps use their own per-item retry budget.

Failure Type	Why It Skips Retries & Action to Take
Unavailable connection	Reconnect the account in Managing Connections (distinct from a raw HTTP `401` / `403`).
Stop and Error	Fails deliberately. Adjust upstream routing if reached unexpectedly.
Interrupted external call	Stops to prevent duplicate writes when an earlier attempt may have dispatched. Inspect the destination first.
Non-retryable AI or limit error	AI Agent caps (`AGENT_LIMIT_REACHED`, `retry.decision: not_retryable`), token limits (`GLOW_AI_TOKEN_LIMIT`), and schema errors require config changes.
HTTP policy or size failure	Blocked destination or oversized response. Request a smaller payload or permitted host.
Invalid input (`400`, `404`, `422`)	Client and unrecognized errors never retry. Fix the field mapping or permissions.

If This Step Fails: Stop or Continue

Once retries are skipped or exhausted, If this step fails decides what happens next.

Triggers always stop on failure. Triggers have no error path. An invalid webhook signature or malformed payload stops the run immediately.

Default behaviorStop Workflow

Halts the branch immediately after retries are exhausted.
Marks the overall automation execution as Failed in Activity.
Downstream Steps on that branch are skipped at 0 credits.

Custom recoveryContinue & Select Error Path

Reveals Select Error Path to wire a dedicated failure branch on the canvas.
Passes the error message forward as output to the error path.
If no Error Path is selected, execution continues to the next normal Step.

Building fallback logic

Fallback Branches (Switch)

Always wire the Fallback branch on a Switch Step. Unmatched data routes to Fallback so you can alert Slack or queue the record for review.

Knowing when something failed

Failed runs appear in Activity and Step-Level Executions. You can alert on individual Steps or across your whole workspace.

Per-Step Error Path Alert

Set the step to Continue

Set If this step fails to Continue and select an Error Path target Step.

Send the alert

Add a Slack, email, or HTTP Request Step on the error path. Format upstream values with Data Transformation if needed:


🚨 _Step Failure Detected_
• _Time (UTC):_ {{ $now }}
• _Automation:_ {{ $workflow_url }}
• _Created by:_ {{ $user.name }} ({{ $user.email }})
• _Error:_ Check the error path Step's input for the failing Step's output and error details.

End with Stop and Error

Connect Stop and Error after the alert Step. This records the run as Failed instead of Passed.

Verify it works

Run: Open Test & Debug on an action Step, enable Retry on fail, or set If this step fails to Continue and select an Error Path.
You should see: The canvas draws the error path from the Step, and transient failures retry up to your configured Max retries before routing there.

What’s Next?

👉 Activity →: find the failed run and identify the affected Step.
Step-Level Executions: inspect each attempt and the Step’s recorded output.
Troubleshooting & Common Errors: diagnose a symptom or exact error message.