RB010: Agent upload failed¶
Applicability: GCS primary; S3 optional through the shared object-storage contract. Operator: CMS/application owner. Verified: 2026-10-23.
Symptom¶
An agent upload cannot start or finalize, remains pending, fails validation, or the installed object cannot be found.
Investigation¶
- Record the upload ID and sanitized request ID. Do not record the signed URL, token, object key if tenant-sensitive, or object contents.
- Determine whether failure occurred at admission, signed-upload transfer,
object
HEAD, size/header validation, immutable finalization, installation, or cleanup reconciliation. - Check application logs, upload state, scoped quota reservation, object
metadata, bucket availability, IAM/service identity, CORS, and provider
status. Use the
ObjectStorageboundary; compatibility module names do not make a GCP upload an S3 workflow. - Confirm the object belongs to the same authorized scope and upload record before any cleanup.
Resolution¶
Retry admission or finalization through the CMS upload service after correcting the classified failure. Retire an invalid or superseded upload through the owning service and allow its cleanup reconciliation to delete the object. Never upload directly with operator credentials, move objects between scopes, change database byte counts, or delete by guessed prefix. Verify immutable installed identity, audit, quota release, and absence of abandoned pending reservations.
Prevention¶
Monitor admission/finalization failures, pending age, cleanup backlog, object storage availability, and quota headroom. Keep size and header validation on the server and test provider adapters with bounded synthetic objects.
Escalate ownership mismatch, repeated signed-upload failure, missing object custody, or any request to inspect uploaded content outside the authorized scope.