Runs, monitoring and recovery
Follow a run step by step, monitor the workspace, retry failed records from the dead-letter queue and cancel runs.
Updated 07/10/2026
Run lifecycle
A run moves through queued, initialising, reading, processing and writing (or waiting_for_rate_limit when a provider asks Nexra to slow down) before it finishes. The run page shows:
each step, with timing and status;
each record, with its source and destination IDs, the mapped payload and any error;
the recovery options available for the run.
Open Dashboard → Connect → Runs to filter runs by flow and status. The 50 most recent appear first.
When records fail
Nexra separates failures that are safe to retry from those that need a person:
Rate limits and temporary provider failures are retried automatically with backoff, honouring Retry-After.
A write whose outcome is uncertain is never retried automatically when the operation has no duplicate protection. The record goes to the dead-letter queue instead.
Validation errors (for example a missing required field) go to the dead-letter queue with the provider's safe error message.
Dead-letter queue
Open Connect → Dead letter to review failed records.
| Action | Where | Effect |
|---|---|---|
| Retry | Console and API | Re-runs the record with a new idempotency key. Only open and retry_failed records can be retried. |
| Retry failed (on a run) | Console and API | Retries up to 100 eligible records from that run |
| Bulk retry | API: POST /api/console/connect/dead-letter/retry | Retries up to 100 records by ID |
| Ignore | API: POST …/dead-letter/:id/ignore | Closes the record without writing it, with an optional note |
| Resolve | API: POST …/dead-letter/:id/resolve | Marks the record as fixed elsewhere, with an optional note |
Before retrying a record whose write had no duplicate protection, check the destination system so that you do not create a duplicate.
Cancel a run
With POST /api/console/connect/executions/:executionId/cancel, a run can be cancelled while it is queued, initialising, reading, processing, waiting for a rate limit or writing. Cancellation stops the run at the next safe point; records already written stay written.
Monitoring
Connect → Monitoring summarises the last 24 hours:
connection health;
runs by status;
open dead-letter records;
pending outbox events;
enabled schedules.
Agents can read the same information through the MCP tools list_connect_flows, get_connect_run and get_connect_integration_event.