i/Aether
1
0
Fork 0
mirror of https://github.com/fawney19/Aether.git synced 2026-10-05 00:47:48 +08:00
Code Issues Packages Projects Releases Wiki Activity

feat: expose upstream response model in usage records

Browse Source
This commit is contained in:
AAEE86
2026-09-21 09:51:33 +08:00
parent ba7c9f8b27
commit f960bbd2c8
24 changed files with 466 additions and 83 deletions
Download Patch File Download Diff File Expand all files Collapse all files
crates/aether-usage/runtime/src/event_wire.rs
+10 -3
View File
@@ -15,9 +15,9 @@ use super::{BorrowedUsageEventEnvelope, UsageEvent, UsageEventData, USAGE_EVENT_
use crate::body_capture::mark_usage_event_capture_truncated;
use crate::request_metadata::{
attach_client_request_body_metadata, attach_provider_request_body_metadata,
attach_provider_response_body_metadata, clear_client_request_body_metadata,
clear_provider_request_body_metadata, request_body_derived_facts_action,
RequestBodyDerivedFactsAction,
attach_provider_response_body_metadata, attach_provider_response_model_metadata,
clear_client_request_body_metadata, clear_provider_request_body_metadata,
request_body_derived_facts_action, RequestBodyDerivedFactsAction,
};
const DIAGNOSTIC_FIELDS: [&str; 8] = [
@@ -238,6 +238,13 @@ impl WireOverrides {
RequestBodyDerivedFactsAction::Clear | RequestBodyDerivedFactsAction::Preserve => {}
}
metadata = attach_provider_response_body_metadata(metadata, data.response_body.as_ref());
metadata = attach_provider_response_model_metadata(
metadata,
request_body,
data.request_body_state,
data.response_body.as_ref(),
data.response_body_state,
);
// Billing reads raw-body TTL before metadata regardless of capture state.
// Preserve that precedence independently of reasoning and tier authority.
if let Some(cache_ttl) = body_cache_ttl {
crates/aether-usage/runtime/src/record.rs
+11 -3
View File
@@ -5,9 +5,9 @@ use aether_data_contracts::DataLayerError;
use crate::request_metadata::{
attach_client_request_body_metadata, attach_provider_request_body_metadata,
clear_client_request_body_metadata, clear_provider_request_body_metadata,
request_body_derived_facts_action, sanitize_usage_request_metadata,
RequestBodyDerivedFactsAction,
attach_provider_response_model_metadata, clear_client_request_body_metadata,
clear_provider_request_body_metadata, request_body_derived_facts_action,
sanitize_usage_request_metadata, RequestBodyDerivedFactsAction,
};
use crate::{UsageEvent, UsageEventType};
@@ -85,6 +85,14 @@ pub fn build_upsert_usage_record_from_event(
}
RequestBodyDerivedFactsAction::Preserve => {}
}
// 响应模型必须在 body 被裁剪前从客户端请求体和上游响应体共同派生；缺少权威 body 时保留已派生事实。
data.request_metadata = attach_provider_response_model_metadata(
data.request_metadata,
data.request_body.as_ref(),
data.request_body_state,
data.response_body.as_ref(),
data.response_body_state,
);
let now_unix_secs = event.timestamp_ms / 1_000;
Ok(UpsertUsageRecord {
crates/aether-usage/runtime/src/request_metadata.rs
+110 -6
View File
@@ -1,13 +1,15 @@
use aether_contracts::ExecutionPlan;
use aether_data_contracts::repository::usage::{
extract_provider_actual_service_tier_from_response,
extract_provider_reasoning_effort_from_body, extract_provider_service_tier_from_body,
normalize_provider_service_tier, resolve_provider_cache_ttl_minutes,
extract_provider_reasoning_effort_from_body, extract_provider_response_model_from_bodies,
extract_provider_service_tier_from_body, normalize_provider_service_tier,
resolve_provider_cache_ttl_minutes,
sanitize_usage_request_metadata as project_usage_request_metadata,
sanitize_usage_request_metadata_object as project_usage_request_metadata_object,
sanitize_usage_request_metadata_ref as project_usage_request_metadata_ref,
UsageBodyCaptureState, PROVIDER_ACTUAL_SERVICE_TIER_METADATA_KEY,
PROVIDER_CACHE_TTL_MINUTES_METADATA_KEY, PROVIDER_REASONING_EFFORT_METADATA_KEY,
usage_body_capture_is_authoritative, UsageBodyCaptureState,
PROVIDER_ACTUAL_SERVICE_TIER_METADATA_KEY, PROVIDER_CACHE_TTL_MINUTES_METADATA_KEY,
PROVIDER_REASONING_EFFORT_METADATA_KEY, PROVIDER_RESPONSE_MODEL_METADATA_KEY,
PROVIDER_SERVICE_TIER_METADATA_KEY, REQUESTED_REASONING_EFFORT_METADATA_KEY,
};
use serde_json::{Map, Value};
@@ -249,6 +251,75 @@ pub(crate) fn attach_provider_response_body_metadata(
attach_provider_actual_service_tier_metadata(metadata, actual_service_tier.as_deref())
}
pub(crate) fn attach_provider_response_model_metadata(
metadata: Option<Value>,
request_body: Option<&Value>,
request_body_state: Option<UsageBodyCaptureState>,
response_body: Option<&Value>,
response_body_state: Option<UsageBodyCaptureState>,
) -> Option<Value> {
let both_bodies_are_authoritative =
usage_body_capture_is_authoritative(request_body, request_body_state)
&& usage_body_capture_is_authoritative(response_body, response_body_state);
let response_model = extract_provider_response_model_from_bodies(
request_body,
request_body_state,
response_body,
response_body_state,
);
if !both_bodies_are_authoritative && response_model.is_none() {
return metadata;
}
let mut object = match metadata {
Some(Value::Object(object)) => object,
_ => Map::new(),
};
// 完整终态 body 是最终候选的权威事实；相同、无效或缺失模型都要清除旧候选值。
if both_bodies_are_authoritative {
object.remove(PROVIDER_RESPONSE_MODEL_METADATA_KEY);
}
if let Some(response_model) = response_model {
object.insert(
PROVIDER_RESPONSE_MODEL_METADATA_KEY.to_string(),
Value::String(response_model),
);
}
(!object.is_empty()).then_some(Value::Object(object))
}
/// 终态候选无法完成比较时，显式清除旧响应模型，避免重试/故障转移残留。
pub(crate) fn refresh_provider_response_model_metadata(
metadata: Option<Value>,
request_body: Option<&Value>,
request_body_state: Option<UsageBodyCaptureState>,
response_body: Option<&Value>,
response_body_state: Option<UsageBodyCaptureState>,
) -> Option<Value> {
let mut object = match metadata {
Some(Value::Object(object)) => object,
_ => Map::new(),
};
object.remove(PROVIDER_RESPONSE_MODEL_METADATA_KEY);
if usage_body_capture_is_authoritative(request_body, request_body_state)
&& usage_body_capture_is_authoritative(response_body, response_body_state)
{
if let Some(response_model) = extract_provider_response_model_from_bodies(
request_body,
request_body_state,
response_body,
response_body_state,
) {
object.insert(
PROVIDER_RESPONSE_MODEL_METADATA_KEY.to_string(),
Value::String(response_model),
);
}
}
(!object.is_empty()).then_some(Value::Object(object))
}
/// Refreshes the response-derived tier for a terminal snapshot. Complete response objects are
/// authoritative even when they contain no tier (which clears a stale candidate value). Capture
/// placeholders/absent bodies are not authoritative, so a terminal summary already present in
@@ -323,8 +394,9 @@ mod tests {
use super::{
attach_client_request_body_metadata, attach_provider_actual_service_tier_metadata,
attach_provider_request_body_metadata, attach_provider_response_body_metadata,
build_usage_request_metadata_seed, merge_usage_request_metadata,
merge_usage_request_metadata_owned, refresh_provider_response_body_metadata,
attach_provider_response_model_metadata, build_usage_request_metadata_seed,
merge_usage_request_metadata, merge_usage_request_metadata_owned,
refresh_provider_response_body_metadata, refresh_provider_response_model_metadata,
retain_first_byte_request_metadata, sanitize_usage_request_metadata,
sanitize_usage_request_metadata_ref,
};
@@ -802,6 +874,38 @@ mod tests {
);
}
#[test]
fn response_model_metadata_is_independent_from_mapping_and_clears_stale_values() {
let metadata = attach_provider_response_model_metadata(
Some(json!({"provider_response_model": "old-model", "trace_id": "trace-1"})),
Some(&json!({"model": "gpt-5"})),
Some(UsageBodyCaptureState::Inline),
Some(&json!({"model": "gpt-5.1"})),
Some(UsageBodyCaptureState::Inline),
)
.expect("response model should be attached");
assert_eq!(metadata["provider_response_model"], "gpt-5.1");
assert_eq!(metadata["trace_id"], "trace-1");
let metadata = refresh_provider_response_model_metadata(
Some(json!({"provider_response_model": "gpt-5.1"})),
Some(&json!({"model": "gpt-5"})),
Some(UsageBodyCaptureState::Inline),
Some(&json!({"model": "gpt-5"})),
Some(UsageBodyCaptureState::Inline),
);
assert!(metadata.is_none());
let metadata = refresh_provider_response_model_metadata(
Some(json!({"provider_response_model": "gpt-5.1"})),
None,
Some(UsageBodyCaptureState::Disabled),
Some(&json!({"model": "gpt-5.2"})),
Some(UsageBodyCaptureState::Inline),
);
assert!(metadata.is_none());
}
#[test]
fn terminal_response_refresh_replaces_stale_actual_tier() {
let metadata = refresh_provider_response_body_metadata(
crates/aether-usage/runtime/src/runtime.rs
+12 -4
View File
@@ -21,9 +21,10 @@ use crate::executor::spawn_on_usage_background_runtime;
use crate::queue::is_permanent_enqueue_error;
use crate::request_metadata::{
attach_client_request_body_metadata, attach_provider_request_body_metadata,
attach_provider_response_body_metadata, clear_client_request_body_metadata,
clear_provider_request_body_metadata, request_body_derived_facts_action,
retain_first_byte_request_metadata, RequestBodyDerivedFactsAction,
attach_provider_response_body_metadata, attach_provider_response_model_metadata,
clear_client_request_body_metadata, clear_provider_request_body_metadata,
request_body_derived_facts_action, retain_first_byte_request_metadata,
RequestBodyDerivedFactsAction,
};
use crate::settlement::{
reconcile_usage_policy_cost_for_event_with_result, settle_usage_with_reconciled_cost,
@@ -5297,8 +5298,15 @@ fn preserve_request_facts_with_legacy_missing(
fn preserve_provider_response_facts(event: &mut UsageEvent) {
let metadata = event.data.request_metadata.take();
event.data.request_metadata =
let metadata =
attach_provider_response_body_metadata(metadata, event.data.response_body.as_ref());
event.data.request_metadata = attach_provider_response_model_metadata(
metadata,
event.data.request_body.as_ref(),
event.data.request_body_state,
event.data.response_body.as_ref(),
event.data.response_body_state,
);
}
impl UsageQueueHealthSnapshot {
crates/aether-usage/runtime/src/write.rs
+25 -3
View File
@@ -17,9 +17,10 @@ use crate::body_capture::{
};
use crate::request_metadata::{
attach_client_request_body_metadata, attach_provider_actual_service_tier_metadata,
attach_provider_request_body_metadata, build_usage_request_metadata_seed,
merge_usage_request_metadata, merge_usage_request_metadata_owned,
refresh_provider_response_body_metadata, sanitize_usage_request_metadata,
attach_provider_request_body_metadata, attach_provider_response_model_metadata,
build_usage_request_metadata_seed, merge_usage_request_metadata,
merge_usage_request_metadata_owned, refresh_provider_response_body_metadata,
refresh_provider_response_model_metadata, sanitize_usage_request_metadata,
sanitize_usage_request_metadata_ref,
};
use crate::{
@@ -724,6 +725,13 @@ fn build_terminal_usage_event_from_seed_impl(
Some(model.as_str()),
provider_request.as_ref(),
);
let request_metadata = attach_provider_response_model_metadata(
request_metadata,
request_body.as_ref(),
body_states.request_body_state,
provider_response.as_ref(),
body_states.response_body_state,
);
let mut data = UsageEventData {
user_id,
@@ -1047,6 +1055,13 @@ pub fn build_sync_terminal_usage_seed(
context_seed.request_metadata,
provider_response_full.as_ref(),
);
let request_metadata = refresh_provider_response_model_metadata(
request_metadata,
context_seed.request_body.as_ref(),
context_seed.body_states.request_body_state,
provider_response_full.as_ref(),
provider_response_body_state,
);
TerminalUsageSeed {
terminal_state,
@@ -1225,6 +1240,13 @@ pub fn build_stream_terminal_usage_seed(
context_seed.request_metadata,
provider_response_full.as_ref(),
);
let request_metadata = refresh_provider_response_model_metadata(
request_metadata,
context_seed.request_body.as_ref(),
context_seed.body_states.request_body_state,
provider_response_full.as_ref(),
provider_response_body_state,
);
// The parser's terminal summary is authoritative when a response body is truncated or the
// body and summary disagree; attach it after the body refresh so it wins.
let request_metadata = attach_provider_actual_service_tier_metadata(