PBRS API Errors, Execution States and Limits
Interpret HTTP responses, asynchronous execution states, ResultJson, retries and operational limits consistently.
Review status: Draft for Development and Support review. Routes and models labelled validation required must be tested against the supported PBRS build before publication.
Error taxonomy
| Class | Examples | Retry guidance |
|---|---|---|
| Connection | DNS, refused/reset connection, TLS failure | Retry only after connectivity is restored |
| Authentication | Token failure, expiry, incorrect scheme | Refresh or correct credentials; bounded retry |
| Authorization | Endpoint or PBRS role denied | Do not retry unchanged |
| Validation | Content type, casing, nesting, enum or model error | Correct the request; do not retry unchanged |
| Capacity | No free threads, waiting work, timeout | Back off, inspect queue and reconcile state |
| Business execution | Source, rendering, artifact or destination failure | Fix the failed layer before rerun |
Asynchronous execution
Record the execution identifier and poll with bounded exponential backoff. Treat completed, failed, cancelled and unknown or stale states separately. Parse ResultJson, then verify artifacts and destinations; a terminal API status alone does not prove delivery.
Limits
Queue retention may expose only a recent operating window. Publish exact retention, pagination, batch size, payload, output size, rate, concurrency and timeout limits only after Development tests them against a named supported build.
Safe errors
Public responses and examples must not expose SQL statements, .NET stack traces, local paths, credentials or customer data. Sanitize evidence before sharing it.
Operational health and alerting
| Signal | Alert condition to define | Response |
|---|---|---|
| Synthetic ping | Consecutive local or remote failures | Check service, registered path, network and TLS |
| Protected read | Ping succeeds but authenticated read fails | Check token endpoint, scheme, role and build |
| Queue depth | Waiting work exceeds the environment baseline | Check scheduler, free threads, rendering and source systems |
| Execution age | Executing longer than the validated workload limit | Correlate, cancel/terminate under runbook and reconcile |
| Delivery | Terminal execution without expected artifact/destination result | Investigate source, rendering and destination separately |
Development and Operations must set numeric thresholds from load tests; do not publish arbitrary universal values.
Scheduler maintenance semantics requiring validation
- Whether stop is graceful or immediate.
- Disposition of executing and queued jobs.
- Whether schedules missed during downtime run on restart.
- How
NextRunchanges after manual, asynchronous and scheduled-time execution. - How duplicate or unsafe catch-up work is prevented.
Documentation standard: Record the PBRS build used for verification, test in a non-production environment, redact credentials and customer data, and confirm the saved PBRS state after every write.