| timestamp |
string (RFC 3339, UTC) |
Request receipt time at the Spark edge, to millisecond precision. |
| request_id |
string (UUIDv4) |
Unique identifier for the request. Returned to the caller in the X-Request-Id response header for correlation. |
| tenant_id |
string |
The customer tenant making the call. Opaque identifier scoped to the customer’s Spark account; not a user identifier. |
| route |
string |
The Spark endpoint invoked (e.g. /v1/chat, /v1/embed). Path only — no query string, no body. |
| method |
string |
HTTP method: GET, POST, PUT, DELETE. |
| status |
integer |
HTTP response status code returned to the caller. |
| payload_size_bytes |
integer |
Length of the request body in bytes. The body itself is not stored. |
| response_size_bytes |
integer |
Length of the response body in bytes. The body itself is not stored. |
| latency_ms |
integer |
End-to-end processing time at the Spark edge, in milliseconds. |
| region |
string |
The data-residency region that handled the request (e.g. us-east, eu-west). |
| tier |
string |
The Spark tier on which the request was served: cloud, connect, or connect-byok. |
| model_provider |
string · nullable |
The upstream model provider (e.g. anthropic, bedrock, vertex, azure-openai). On Connect BYOK this points to the customer’s own endpoint provider. |
| model_id |
string · nullable |
The model identifier invoked. Identifier only — no prompt, no completion. |
| tokens_in |
integer · nullable |
Input token count reported by the model provider. Count only — no token contents. |
| tokens_out |
integer · nullable |
Output token count reported by the model provider. Count only — no token contents. |
| client_ip_hash |
string (SHA-256, salted) |
One-way salted hash of the caller IP. Used for abuse detection and rate-limit accounting; the raw IP is not retained. |
| error_code |
string · nullable |
Spark-internal error code when status >= 400. Code only — no stack trace, no message body. |