Retry Failed Raw Event

POST /v1/projects/:project_id/raw-events/:id/process

Authentication and path

Send Authorization: Bearer <api-key> with rules:test permission. project_id identifies the owning project; id is the original raw_event_id. There are no query parameters.

Request body

Field	Required	Meaning
processing_log_id	Yes	Failure log id from List Processing Failures. Only retried_at = 0 can be processed.
extractor_rule_set_id	No	Override the extractor; must be supplied with extractor_version. Takes precedence over the logged or replacement event’s extractor ID.
extractor_version	No	Positive version number of the specified extractor. Must be published and accessible to this tenant/project; no fallback to the latest version.
raw_event	No	Complete replacement business input. Omit to use the stored input. This is an object, not a JSON string.
raw_event.extractor_rule_set_id	Yes when replacing without a root override	Extractor rule set to use.
raw_event.idempotency_key	Yes when replacing	Original idempotency key; changing event identity is rejected.
raw_event.occurred_at	Yes when replacing	Original occurrence time in Unix seconds; changing event identity is rejected.
raw_event.raw_json	Yes when replacing	Corrected original usage JSON.
raw_event.source	No	Event source.
raw_event.external_event_id	No	External identifier.
raw_event.subject_type, raw_event.subject_id	No	Explicit subject; provide both or neither.
raw_event.x-billing	No	Reported dynamic charge items and pricing/discount hints; same contract as ingest.
raw_event.x-usage-control	No	Usage-control extension; same contract as ingest.
raw_event.metadata	No	Business string map. Internal processing metadata is preserved by the service.

Copy raw_event from the failure list to edit business data. An explicit replacement is complete, not a partial patch. Tenant/project ownership, revision and original ingested_at are managed by the service. The ID is recalculated from tenant, project, idempotency key, occurred_at and the stored revision, and must match id.

The extractor fields use the same names as reprice. Omit both to retain the existing selection behavior. Provide both to pin an exact published version, even without replacing raw_event. Invalid versions are rejected before claiming the log. If processing fails, the new failure snapshot retains the selected extractor and version for subsequent retries.

{
  "processing_log_id": "rpl_attempt_1",
  "extractor_rule_set_id": "ers_gateway",
  "extractor_version": 2,
  "raw_event": {
    "extractor_rule_set_id": "ers_gateway",
    "source": "gateway",
    "idempotency_key": "request_123",
    "occurred_at": 1791602563,
    "raw_json": {"usage": {"tokens": 12}}
  }
}

Response

A successful synchronous retry returns HTTP 200:

{
  "raw_event_id": "raw_original",
  "subject_type": "user",
  "subject_id": "user_123",
  "dimensions": {"model": "example"},
  "charge_items": [],
  "unmatched_items": []
}

raw_event_id, subject_type, subject_id are strings; dimensions is a string map; charge_items and unmatched_items are arrays of the rated and unmatched charge-item schema. Normal wallet debit, limits and analytics publication apply on success.

Validation failures (including changed identity) return HTTP 400 and leave the log eligible. A processing failure returns HTTP 400, sets the old log’s retried_at and creates a new failure with corrected input and retried_at = 0. The old log disappears from the failure list and cannot be processed again. Concurrent requests can claim a log once; raw-event conflicts never debit again. This endpoint does not repair already successful events or analytics projections.

The process API claims the failure log after input/identity validation and before rating. Shared batch persistence only inserts raw events and financial facts; it does not claim failure logs. Request cancellation still attempts to persist a new failure using a bounded recovery context.