mirror of
https://github.com/fawney19/Aether.git
synced 2026-09-01 17:00:21 +08:00
Unify pool score display as account health
This commit is contained in:
@@ -49,7 +49,7 @@ pub(crate) use self::planner::{
|
||||
build_standard_sync_plan_from_decision, candidate_auth_channel_skip_reason,
|
||||
extract_pool_sticky_session_token, maybe_build_stream_decision_payload,
|
||||
maybe_build_stream_plan_payload, maybe_build_sync_decision_payload,
|
||||
maybe_build_sync_plan_payload, planner_is_matching_stream_request,
|
||||
maybe_build_sync_plan_payload, planner_is_matching_stream_request, provider_key_pool_score_id,
|
||||
provider_key_pool_score_scope, read_candidate_transport_snapshot,
|
||||
record_local_runtime_candidate_skip_reason,
|
||||
set_local_openai_chat_execution_exhausted_diagnostic,
|
||||
|
||||
@@ -41,8 +41,9 @@ pub(crate) use self::plan_builders::{
|
||||
build_standard_stream_plan_from_decision, build_standard_sync_plan_from_decision,
|
||||
AiStreamAttempt, AiSyncAttempt,
|
||||
};
|
||||
pub(crate) use self::pool_scores::build_provider_key_pool_score_upsert;
|
||||
pub(crate) use self::pool_scores::provider_key_pool_score_scope;
|
||||
pub(crate) use self::pool_scores::{
|
||||
build_provider_key_pool_score_upsert, provider_key_pool_score_id, provider_key_pool_score_scope,
|
||||
};
|
||||
pub(crate) use self::route::is_matching_stream_request as planner_is_matching_stream_request;
|
||||
pub(crate) use self::runtime_miss::{
|
||||
apply_local_runtime_candidate_terminal_reason, record_local_runtime_candidate_skip_reason,
|
||||
|
||||
@@ -3,7 +3,7 @@ use aether_ai_serving::{
|
||||
};
|
||||
use aether_data_contracts::repository::pool_scores::{
|
||||
PoolMemberIdentity, PoolMemberProbeStatus, PoolScoreScope, UpsertPoolMemberScore,
|
||||
POOL_SCORE_SCOPE_KIND_MODEL,
|
||||
POOL_SCORE_CAPABILITY_ACCOUNT, POOL_SCORE_SCOPE_KIND_ACCOUNT,
|
||||
};
|
||||
use aether_data_contracts::repository::provider_catalog::StoredProviderCatalogKey;
|
||||
use serde_json::Value;
|
||||
@@ -13,14 +13,12 @@ use crate::handlers::shared::{provider_key_health_summary, provider_key_status_s
|
||||
pub(crate) fn build_provider_key_pool_score_upsert(
|
||||
key: &StoredProviderCatalogKey,
|
||||
provider_type: &str,
|
||||
api_format: &str,
|
||||
model_id: Option<&str>,
|
||||
existing: Option<&aether_data_contracts::repository::pool_scores::StoredPoolMemberScore>,
|
||||
now_unix_secs: u64,
|
||||
score_rules: PoolMemberScoreRules,
|
||||
) -> UpsertPoolMemberScore {
|
||||
let identity = PoolMemberIdentity::provider_api_key(key.provider_id.clone(), key.id.clone());
|
||||
let scope = provider_key_pool_score_scope(api_format, model_id);
|
||||
let scope = provider_key_pool_score_scope();
|
||||
let input = provider_key_score_input(
|
||||
key,
|
||||
provider_type,
|
||||
@@ -54,17 +52,11 @@ pub(crate) fn build_provider_key_pool_score_upsert(
|
||||
}
|
||||
}
|
||||
|
||||
pub(crate) fn provider_key_pool_score_scope(
|
||||
api_format: &str,
|
||||
model_id: Option<&str>,
|
||||
) -> PoolScoreScope {
|
||||
pub(crate) fn provider_key_pool_score_scope() -> PoolScoreScope {
|
||||
PoolScoreScope {
|
||||
capability: api_format.trim().to_ascii_lowercase(),
|
||||
scope_kind: POOL_SCORE_SCOPE_KIND_MODEL.to_string(),
|
||||
scope_id: model_id
|
||||
.map(str::trim)
|
||||
.filter(|value| !value.is_empty())
|
||||
.map(ToOwned::to_owned),
|
||||
capability: POOL_SCORE_CAPABILITY_ACCOUNT.to_string(),
|
||||
scope_kind: POOL_SCORE_SCOPE_KIND_ACCOUNT.to_string(),
|
||||
scope_id: None,
|
||||
}
|
||||
}
|
||||
|
||||
@@ -133,8 +125,7 @@ fn provider_key_score_input(
|
||||
.and_then(Value::as_bool)
|
||||
.unwrap_or(false),
|
||||
oauth_invalid_reason: key.oauth_invalid_reason.clone(),
|
||||
circuit_open: any_circuit_open
|
||||
|| circuit_open_for_scope(key.circuit_breaker_by_format.as_ref(), &scope),
|
||||
circuit_open: any_circuit_open,
|
||||
success_count: key.success_count.unwrap_or(0).into(),
|
||||
error_count: key.error_count.unwrap_or(0).into(),
|
||||
total_response_time_ms: key.total_response_time_ms.unwrap_or(0).into(),
|
||||
@@ -150,21 +141,6 @@ fn provider_key_score_input(
|
||||
}
|
||||
}
|
||||
|
||||
fn circuit_open_for_scope(circuit_by_format: Option<&Value>, scope: &PoolScoreScope) -> bool {
|
||||
let Some(formats) = circuit_by_format.and_then(Value::as_object) else {
|
||||
return false;
|
||||
};
|
||||
let keys = api_format_lookup_keys(&scope.capability);
|
||||
keys.iter().any(|key| {
|
||||
formats
|
||||
.get(key)
|
||||
.and_then(Value::as_object)
|
||||
.and_then(|value| value.get("open"))
|
||||
.and_then(Value::as_bool)
|
||||
.unwrap_or(false)
|
||||
})
|
||||
}
|
||||
|
||||
fn json_f64(value: &Value) -> Option<f64> {
|
||||
value.as_f64().or_else(|| {
|
||||
value
|
||||
@@ -175,17 +151,6 @@ fn json_f64(value: &Value) -> Option<f64> {
|
||||
})
|
||||
}
|
||||
|
||||
fn api_format_lookup_keys(api_format: &str) -> Vec<String> {
|
||||
let normalized = crate::ai_serving::normalize_api_format_alias(api_format);
|
||||
let mut keys = crate::ai_serving::api_format_storage_aliases(&normalized);
|
||||
if !keys.iter().any(|value| value == &normalized) {
|
||||
keys.push(normalized);
|
||||
}
|
||||
keys.sort();
|
||||
keys.dedup();
|
||||
keys
|
||||
}
|
||||
|
||||
fn stable_hash(bytes: &[u8]) -> u64 {
|
||||
let mut hash = 0xcbf29ce484222325u64;
|
||||
for byte in bytes {
|
||||
|
||||
@@ -387,10 +387,7 @@ impl<'a> PoolKeyCursor<'a> {
|
||||
}
|
||||
|
||||
async fn next_score_candidates(&mut self) -> Option<Vec<EligibleLocalExecutionCandidate>> {
|
||||
let scope = provider_key_pool_score_scope(
|
||||
self.group.candidate.endpoint_api_format.as_str(),
|
||||
Some(self.group.candidate.model_id.as_str()),
|
||||
);
|
||||
let scope = provider_key_pool_score_scope();
|
||||
let query = ListRankedPoolMembersQuery {
|
||||
pool_kind: POOL_KIND_PROVIDER_KEY_POOL.to_string(),
|
||||
pool_id: self.group.candidate.provider_id.clone(),
|
||||
|
||||
@@ -6,6 +6,7 @@ use crate::handlers::admin::shared::{provider_key_status_snapshot_payload, unix_
|
||||
use crate::provider_key_auth::{provider_key_auth_semantics, provider_key_effective_api_formats};
|
||||
use aether_admin::provider::pool as admin_provider_pool_pure;
|
||||
use aether_admin::provider::quota as admin_provider_quota_pure;
|
||||
use aether_data_contracts::repository::pool_scores::StoredPoolMemberScore;
|
||||
use aether_data_contracts::repository::provider_catalog::{
|
||||
StoredProviderCatalogEndpoint, StoredProviderCatalogKey,
|
||||
};
|
||||
@@ -906,6 +907,7 @@ pub(super) fn build_admin_pool_key_payload(
|
||||
key: &StoredProviderCatalogKey,
|
||||
runtime: &AdminProviderPoolRuntimeState,
|
||||
pool_config: Option<AdminProviderPoolConfig>,
|
||||
pool_score: Option<&StoredPoolMemberScore>,
|
||||
codex_cycle_usage_by_code: Option<&BTreeMap<String, StoredProviderApiKeyWindowUsageSummary>>,
|
||||
now_unix_secs: u64,
|
||||
) -> serde_json::Value {
|
||||
@@ -1087,6 +1089,34 @@ pub(super) fn build_admin_pool_key_payload(
|
||||
payload.insert("status_snapshot".to_string(), status_snapshot);
|
||||
payload.insert("quota_updated_at".to_string(), json!(quota_updated_at));
|
||||
payload.insert("health_score".to_string(), json!(health_score));
|
||||
payload.insert(
|
||||
"pool_score".to_string(),
|
||||
pool_score
|
||||
.map(|score| {
|
||||
json!({
|
||||
"id": score.id.clone(),
|
||||
"capability": score.capability.clone(),
|
||||
"scope_kind": score.scope_kind.clone(),
|
||||
"scope_id": score.scope_id.clone(),
|
||||
"score": score.score,
|
||||
"hard_state": score.hard_state.as_database(),
|
||||
"score_version": score.score_version,
|
||||
"score_reason": score.score_reason.clone(),
|
||||
"last_ranked_at": score.last_ranked_at,
|
||||
"last_scheduled_at": score.last_scheduled_at,
|
||||
"last_success_at": score.last_success_at,
|
||||
"last_failure_at": score.last_failure_at,
|
||||
"failure_count": score.failure_count,
|
||||
"last_probe_attempt_at": score.last_probe_attempt_at,
|
||||
"last_probe_success_at": score.last_probe_success_at,
|
||||
"last_probe_failure_at": score.last_probe_failure_at,
|
||||
"probe_failure_count": score.probe_failure_count,
|
||||
"probe_status": score.probe_status.as_database(),
|
||||
"updated_at": score.updated_at,
|
||||
})
|
||||
})
|
||||
.unwrap_or(serde_json::Value::Null),
|
||||
);
|
||||
payload.insert(
|
||||
"circuit_breaker_open".to_string(),
|
||||
json!(circuit_breaker_open),
|
||||
|
||||
@@ -7,9 +7,13 @@ use super::{
|
||||
AdminPoolKeySortField, AdminProviderPoolRuntimeState, ProviderCatalogKeyListOrder,
|
||||
ProviderCatalogKeyListQuery, ADMIN_POOL_PROVIDER_CATALOG_READER_UNAVAILABLE_DETAIL,
|
||||
};
|
||||
use crate::ai_serving::{provider_key_pool_score_id, provider_key_pool_score_scope};
|
||||
use crate::handlers::admin::request::{AdminAppState, AdminRequestContext};
|
||||
use crate::GatewayError;
|
||||
use aether_admin::provider::pool as admin_provider_pool_pure;
|
||||
use aether_data_contracts::repository::pool_scores::{
|
||||
GetPoolMemberScoresByIdsQuery, PoolMemberIdentity, StoredPoolMemberScore,
|
||||
};
|
||||
use aether_data_contracts::repository::provider_catalog::StoredProviderCatalogKey;
|
||||
use aether_data_contracts::repository::usage::{
|
||||
ProviderApiKeyWindowUsageRequest, StoredProviderApiKeyWindowUsageSummary,
|
||||
@@ -56,6 +60,36 @@ fn admin_pool_current_unix_secs() -> u64 {
|
||||
.unwrap_or(0)
|
||||
}
|
||||
|
||||
async fn read_admin_pool_scores_by_key_id(
|
||||
state: &AdminAppState<'_>,
|
||||
provider_id: &str,
|
||||
key_ids: &[String],
|
||||
) -> Result<BTreeMap<String, StoredPoolMemberScore>, GatewayError> {
|
||||
if key_ids.is_empty() {
|
||||
return Ok(BTreeMap::new());
|
||||
}
|
||||
|
||||
let score_scope = provider_key_pool_score_scope();
|
||||
let score_ids = key_ids
|
||||
.iter()
|
||||
.map(|key_id| {
|
||||
let identity =
|
||||
PoolMemberIdentity::provider_api_key(provider_id.to_string(), key_id.clone());
|
||||
provider_key_pool_score_id(&identity, &score_scope)
|
||||
})
|
||||
.collect::<Vec<_>>();
|
||||
let scores = state
|
||||
.app()
|
||||
.data
|
||||
.get_pool_member_scores_by_ids(&GetPoolMemberScoresByIdsQuery { ids: score_ids })
|
||||
.await
|
||||
.map_err(|err| GatewayError::Internal(format!("{err:?}")))?;
|
||||
Ok(scores
|
||||
.into_iter()
|
||||
.map(|score| (score.member_id.clone(), score))
|
||||
.collect::<BTreeMap<_, _>>())
|
||||
}
|
||||
|
||||
fn admin_pool_codex_cycle_usage_request(
|
||||
key: &StoredProviderCatalogKey,
|
||||
window: &serde_json::Map<String, serde_json::Value>,
|
||||
@@ -379,6 +413,9 @@ pub(super) async fn build_admin_pool_list_keys_response(
|
||||
};
|
||||
|
||||
let key_ids = keys.iter().map(|key| key.id.clone()).collect::<Vec<_>>();
|
||||
let pool_scores_by_key_id = read_admin_pool_scores_by_key_id(state, &provider.id, &key_ids)
|
||||
.await
|
||||
.unwrap_or_default();
|
||||
let endpoints = state
|
||||
.list_provider_catalog_endpoints_by_provider_ids(std::slice::from_ref(&provider.id))
|
||||
.await?;
|
||||
@@ -414,6 +451,7 @@ pub(super) async fn build_admin_pool_list_keys_response(
|
||||
&key,
|
||||
&runtime,
|
||||
pool_config.clone(),
|
||||
pool_scores_by_key_id.get(&key.id),
|
||||
codex_cycle_usage_by_key.get(&key.id),
|
||||
now_unix_secs,
|
||||
)
|
||||
|
||||
@@ -7,7 +7,7 @@ use crate::handlers::shared::query_param_value;
|
||||
use crate::GatewayError;
|
||||
use aether_data_contracts::repository::pool_scores::{
|
||||
ListPoolMemberScoresQuery, PoolMemberHardState, PoolMemberProbeStatus,
|
||||
POOL_KIND_PROVIDER_KEY_POOL, POOL_SCORE_SCOPE_KIND_MODEL,
|
||||
POOL_KIND_PROVIDER_KEY_POOL, POOL_SCORE_CAPABILITY_ACCOUNT, POOL_SCORE_SCOPE_KIND_ACCOUNT,
|
||||
};
|
||||
use axum::{
|
||||
body::Body,
|
||||
@@ -49,9 +49,6 @@ pub(super) async fn build_admin_pool_scores_response(
|
||||
}
|
||||
};
|
||||
let offset = page.saturating_sub(1).saturating_mul(page_size);
|
||||
let api_format = query_param_value(query, "api_format")
|
||||
.map(|value| crate::ai_serving::normalize_api_format_alias(value.as_str()));
|
||||
let model_id = query_param_value(query, "model_id");
|
||||
let hard_states = match parse_hard_state_filter(query) {
|
||||
Ok(value) => value,
|
||||
Err(message) => {
|
||||
@@ -77,9 +74,9 @@ pub(super) async fn build_admin_pool_scores_response(
|
||||
.list_pool_member_scores(&ListPoolMemberScoresQuery {
|
||||
pool_kind: POOL_KIND_PROVIDER_KEY_POOL.to_string(),
|
||||
pool_id: provider_id.clone(),
|
||||
capability: api_format.clone(),
|
||||
scope_kind: Some(POOL_SCORE_SCOPE_KIND_MODEL.to_string()),
|
||||
scope_id: model_id.clone(),
|
||||
capability: Some(POOL_SCORE_CAPABILITY_ACCOUNT.to_string()),
|
||||
scope_kind: Some(POOL_SCORE_SCOPE_KIND_ACCOUNT.to_string()),
|
||||
scope_id: None,
|
||||
hard_states,
|
||||
probe_statuses,
|
||||
offset,
|
||||
@@ -146,8 +143,8 @@ pub(super) async fn build_admin_pool_scores_response(
|
||||
"page": page,
|
||||
"page_size": page_size,
|
||||
"filters": {
|
||||
"api_format": api_format,
|
||||
"model_id": model_id,
|
||||
"api_format": serde_json::Value::Null,
|
||||
"model_id": serde_json::Value::Null,
|
||||
"hard_state": query_param_value(query, "hard_state"),
|
||||
"probe_status": query_param_value(query, "probe_status")
|
||||
},
|
||||
|
||||
@@ -1,7 +1,6 @@
|
||||
use std::collections::BTreeMap;
|
||||
use std::time::{Duration, SystemTime, UNIX_EPOCH};
|
||||
|
||||
use aether_data_contracts::repository::global_models::AdminProviderModelListQuery;
|
||||
use aether_data_contracts::repository::pool_scores::GetPoolMemberScoresByIdsQuery;
|
||||
use aether_data_contracts::repository::provider_catalog::{
|
||||
StoredProviderCatalogEndpoint, StoredProviderCatalogKey, StoredProviderCatalogProvider,
|
||||
@@ -113,32 +112,11 @@ fn provider_offset_cursor_key(provider_id: &str) -> String {
|
||||
format!("{POOL_SCORE_REBUILD_PROVIDER_OFFSET_PREFIX}:{provider_id}")
|
||||
}
|
||||
|
||||
fn score_combo_indices(
|
||||
flat_index: usize,
|
||||
model_count: usize,
|
||||
key_count: usize,
|
||||
) -> (usize, usize, usize) {
|
||||
let endpoint_stride = model_count.saturating_mul(key_count).max(1);
|
||||
let endpoint_index = flat_index / endpoint_stride;
|
||||
let remainder = flat_index % endpoint_stride;
|
||||
let model_index = remainder / key_count.max(1);
|
||||
let key_index = remainder % key_count.max(1);
|
||||
(endpoint_index, model_index, key_index)
|
||||
}
|
||||
|
||||
#[derive(Debug, Clone)]
|
||||
struct ProviderScoreBuildItem {
|
||||
endpoint_index: usize,
|
||||
model_index: usize,
|
||||
key_index: usize,
|
||||
score_id: String,
|
||||
}
|
||||
|
||||
pub(crate) async fn ensure_provider_key_pool_scores_for_keys(
|
||||
state: &AppState,
|
||||
provider: &StoredProviderCatalogProvider,
|
||||
pool_config: &AdminProviderPoolConfig,
|
||||
endpoints: &[StoredProviderCatalogEndpoint],
|
||||
_endpoints: &[StoredProviderCatalogEndpoint],
|
||||
keys: &[StoredProviderCatalogKey],
|
||||
now_unix_secs: u64,
|
||||
max_upserts: usize,
|
||||
@@ -151,63 +129,28 @@ pub(crate) async fn ensure_provider_key_pool_scores_for_keys(
|
||||
return Ok(0);
|
||||
}
|
||||
|
||||
let endpoints = endpoints
|
||||
.iter()
|
||||
.filter(|endpoint| endpoint.is_active && !endpoint.api_format.trim().is_empty())
|
||||
.collect::<Vec<_>>();
|
||||
let keys = keys
|
||||
.iter()
|
||||
.filter(|key| key.is_active && key.provider_id == provider.id)
|
||||
.collect::<Vec<_>>();
|
||||
if endpoints.is_empty() || keys.is_empty() {
|
||||
if keys.is_empty() {
|
||||
return Ok(0);
|
||||
}
|
||||
|
||||
let models = state
|
||||
.list_admin_provider_models(&AdminProviderModelListQuery {
|
||||
provider_id: provider.id.clone(),
|
||||
is_active: Some(true),
|
||||
offset: 0,
|
||||
limit: 10_000,
|
||||
})
|
||||
.await?
|
||||
let build_items = keys
|
||||
.into_iter()
|
||||
.filter(|model| model.is_available)
|
||||
.take(max_upserts)
|
||||
.map(|key| {
|
||||
let draft = build_provider_key_pool_score_upsert(
|
||||
key,
|
||||
provider.provider_type.as_str(),
|
||||
None,
|
||||
now_unix_secs,
|
||||
pool_config.score_rules,
|
||||
);
|
||||
(key, draft.id)
|
||||
})
|
||||
.collect::<Vec<_>>();
|
||||
if models.is_empty() {
|
||||
return Ok(0);
|
||||
}
|
||||
|
||||
let max_items = endpoints
|
||||
.len()
|
||||
.saturating_mul(models.len())
|
||||
.saturating_mul(keys.len())
|
||||
.min(max_upserts);
|
||||
let mut build_items = Vec::with_capacity(max_items);
|
||||
'outer: for (endpoint_index, endpoint) in endpoints.iter().enumerate() {
|
||||
for (model_index, model) in models.iter().enumerate() {
|
||||
for (key_index, key) in keys.iter().enumerate() {
|
||||
let draft = build_provider_key_pool_score_upsert(
|
||||
key,
|
||||
provider.provider_type.as_str(),
|
||||
endpoint.api_format.trim(),
|
||||
Some(model.id.as_str()),
|
||||
None,
|
||||
now_unix_secs,
|
||||
pool_config.score_rules,
|
||||
);
|
||||
build_items.push(ProviderScoreBuildItem {
|
||||
endpoint_index,
|
||||
model_index,
|
||||
key_index,
|
||||
score_id: draft.id,
|
||||
});
|
||||
if build_items.len() >= max_upserts {
|
||||
break 'outer;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
if build_items.is_empty() {
|
||||
return Ok(0);
|
||||
}
|
||||
@@ -217,7 +160,7 @@ pub(crate) async fn ensure_provider_key_pool_scores_for_keys(
|
||||
.get_pool_member_scores_by_ids(&GetPoolMemberScoresByIdsQuery {
|
||||
ids: build_items
|
||||
.iter()
|
||||
.map(|item| item.score_id.clone())
|
||||
.map(|(_, score_id)| score_id.clone())
|
||||
.collect(),
|
||||
})
|
||||
.await
|
||||
@@ -234,18 +177,13 @@ pub(crate) async fn ensure_provider_key_pool_scores_for_keys(
|
||||
.collect::<std::collections::BTreeSet<_>>();
|
||||
|
||||
let mut upserted = 0usize;
|
||||
for item in &build_items {
|
||||
if existing_score_ids.contains(&item.score_id) {
|
||||
for (key, score_id) in &build_items {
|
||||
if existing_score_ids.contains(score_id) {
|
||||
continue;
|
||||
}
|
||||
let endpoint = endpoints[item.endpoint_index];
|
||||
let model = &models[item.model_index];
|
||||
let key = keys[item.key_index];
|
||||
let upsert = build_provider_key_pool_score_upsert(
|
||||
key,
|
||||
provider.provider_type.as_str(),
|
||||
endpoint.api_format.trim(),
|
||||
Some(model.id.as_str()),
|
||||
None,
|
||||
now_unix_secs,
|
||||
pool_config.score_rules,
|
||||
@@ -292,19 +230,6 @@ pub(crate) async fn perform_pool_score_rebuild_once_with_config(
|
||||
.iter()
|
||||
.map(|(provider, _)| provider.id.clone())
|
||||
.collect::<Vec<_>>();
|
||||
let mut endpoints_by_provider = BTreeMap::new();
|
||||
for endpoint in state
|
||||
.list_provider_catalog_endpoints_by_provider_ids(&provider_ids)
|
||||
.await?
|
||||
.into_iter()
|
||||
.filter(|endpoint| endpoint.is_active)
|
||||
{
|
||||
endpoints_by_provider
|
||||
.entry(endpoint.provider_id.clone())
|
||||
.or_insert_with(Vec::new)
|
||||
.push(endpoint);
|
||||
}
|
||||
|
||||
let mut keys_by_provider = BTreeMap::new();
|
||||
for key in state
|
||||
.list_provider_catalog_keys_by_provider_ids(&provider_ids)
|
||||
@@ -315,6 +240,9 @@ pub(crate) async fn perform_pool_score_rebuild_once_with_config(
|
||||
.or_insert_with(Vec::new)
|
||||
.push(key);
|
||||
}
|
||||
for keys in keys_by_provider.values_mut() {
|
||||
keys.sort_by(|left, right| left.id.cmp(&right.id));
|
||||
}
|
||||
|
||||
let now = now_unix_secs();
|
||||
let mut summary = PoolScoreRebuildRunSummary {
|
||||
@@ -333,75 +261,39 @@ pub(crate) async fn perform_pool_score_rebuild_once_with_config(
|
||||
}
|
||||
last_provider_index = Some(provider_index);
|
||||
let (provider, pool_config) = providers[provider_index].clone();
|
||||
let endpoints = endpoints_by_provider
|
||||
.remove(&provider.id)
|
||||
.unwrap_or_default();
|
||||
let keys = keys_by_provider.remove(&provider.id).unwrap_or_default();
|
||||
if endpoints.is_empty() || keys.is_empty() {
|
||||
continue;
|
||||
}
|
||||
let models = state
|
||||
.list_admin_provider_models(&AdminProviderModelListQuery {
|
||||
provider_id: provider.id.clone(),
|
||||
is_active: Some(true),
|
||||
offset: 0,
|
||||
limit: 10_000,
|
||||
})
|
||||
.await?
|
||||
let keys = keys
|
||||
.into_iter()
|
||||
.filter(|model| model.is_available)
|
||||
.filter(|key| key.is_active)
|
||||
.collect::<Vec<_>>();
|
||||
if models.is_empty() {
|
||||
continue;
|
||||
}
|
||||
|
||||
let total_combinations = endpoints
|
||||
.len()
|
||||
.saturating_mul(models.len())
|
||||
.saturating_mul(keys.len());
|
||||
if total_combinations == 0 {
|
||||
if keys.is_empty() {
|
||||
continue;
|
||||
}
|
||||
let total_keys = keys.len();
|
||||
let provider_cursor_key = provider_offset_cursor_key(&provider.id);
|
||||
let provider_cursor =
|
||||
load_runtime_usize(state, &provider_cursor_key).await % total_combinations.max(1);
|
||||
let provider_cursor = load_runtime_usize(state, &provider_cursor_key).await % total_keys;
|
||||
let remaining_budget = config
|
||||
.max_upserts_per_tick
|
||||
.saturating_sub(summary.scores_upserted);
|
||||
let provider_budget = remaining_budget.min(total_combinations);
|
||||
let provider_budget = remaining_budget.min(total_keys);
|
||||
let mut build_items = Vec::with_capacity(provider_budget);
|
||||
for offset in 0..provider_budget {
|
||||
let flat_index = (provider_cursor + offset) % total_combinations;
|
||||
let (endpoint_index, model_index, key_index) =
|
||||
score_combo_indices(flat_index, models.len(), keys.len());
|
||||
let endpoint = &endpoints[endpoint_index];
|
||||
let api_format = endpoint.api_format.trim();
|
||||
if api_format.is_empty() {
|
||||
continue;
|
||||
}
|
||||
let model = &models[model_index];
|
||||
let key_index = (provider_cursor + offset) % total_keys;
|
||||
let key = &keys[key_index];
|
||||
let draft = build_provider_key_pool_score_upsert(
|
||||
key,
|
||||
provider.provider_type.as_str(),
|
||||
api_format,
|
||||
Some(model.id.as_str()),
|
||||
None,
|
||||
now,
|
||||
pool_config.score_rules,
|
||||
);
|
||||
build_items.push(ProviderScoreBuildItem {
|
||||
endpoint_index,
|
||||
model_index,
|
||||
key_index,
|
||||
score_id: draft.id,
|
||||
});
|
||||
build_items.push((key_index, draft.id));
|
||||
}
|
||||
if build_items.is_empty() {
|
||||
store_runtime_usize(
|
||||
state,
|
||||
&provider_cursor_key,
|
||||
(provider_cursor + provider_budget) % total_combinations,
|
||||
(provider_cursor + provider_budget) % total_keys,
|
||||
)
|
||||
.await;
|
||||
continue;
|
||||
@@ -411,7 +303,7 @@ pub(crate) async fn perform_pool_score_rebuild_once_with_config(
|
||||
.get_pool_member_scores_by_ids(&GetPoolMemberScoresByIdsQuery {
|
||||
ids: build_items
|
||||
.iter()
|
||||
.map(|item| item.score_id.clone())
|
||||
.map(|(_, score_id)| score_id.clone())
|
||||
.collect(),
|
||||
})
|
||||
.await
|
||||
@@ -428,19 +320,15 @@ pub(crate) async fn perform_pool_score_rebuild_once_with_config(
|
||||
.collect::<BTreeMap<_, _>>();
|
||||
let mut provider_upserts = 0usize;
|
||||
summary.keys_seen = summary.keys_seen.saturating_add(keys.len());
|
||||
for item in &build_items {
|
||||
for (key_index, score_id) in &build_items {
|
||||
if summary.scores_upserted >= config.max_upserts_per_tick {
|
||||
break;
|
||||
}
|
||||
let endpoint = &endpoints[item.endpoint_index];
|
||||
let model = &models[item.model_index];
|
||||
let key = &keys[item.key_index];
|
||||
let existing = existing_scores.get(&item.score_id);
|
||||
let key = &keys[*key_index];
|
||||
let existing = existing_scores.get(score_id);
|
||||
let upsert = build_provider_key_pool_score_upsert(
|
||||
key,
|
||||
provider.provider_type.as_str(),
|
||||
endpoint.api_format.trim(),
|
||||
Some(model.id.as_str()),
|
||||
existing,
|
||||
now,
|
||||
pool_config.score_rules,
|
||||
@@ -459,7 +347,7 @@ pub(crate) async fn perform_pool_score_rebuild_once_with_config(
|
||||
store_runtime_usize(
|
||||
state,
|
||||
&provider_cursor_key,
|
||||
(provider_cursor + provider_budget) % total_combinations,
|
||||
(provider_cursor + provider_budget) % total_keys,
|
||||
)
|
||||
.await;
|
||||
if provider_upserts > 0 {
|
||||
@@ -528,16 +416,3 @@ pub(crate) fn spawn_pool_score_rebuild_worker(
|
||||
}
|
||||
}))
|
||||
}
|
||||
|
||||
#[cfg(test)]
|
||||
mod tests {
|
||||
use super::score_combo_indices;
|
||||
|
||||
#[test]
|
||||
fn score_combo_indices_walks_endpoint_model_key_order() {
|
||||
assert_eq!(score_combo_indices(0, 2, 3), (0, 0, 0));
|
||||
assert_eq!(score_combo_indices(2, 2, 3), (0, 0, 2));
|
||||
assert_eq!(score_combo_indices(3, 2, 3), (0, 1, 0));
|
||||
assert_eq!(score_combo_indices(6, 2, 3), (1, 0, 0));
|
||||
}
|
||||
}
|
||||
|
||||
@@ -5,8 +5,7 @@ use aether_data::repository::pool_scores::InMemoryPoolMemberScoreRepository;
|
||||
use aether_data::repository::provider_catalog::InMemoryProviderCatalogReadRepository;
|
||||
use aether_data::repository::usage::InMemoryUsageReadRepository;
|
||||
use aether_data_contracts::repository::pool_scores::{
|
||||
PoolMemberHardState, PoolMemberProbeStatus, StoredPoolMemberScore, POOL_KIND_PROVIDER_KEY_POOL,
|
||||
POOL_MEMBER_KIND_PROVIDER_API_KEY, POOL_SCORE_SCOPE_KIND_MODEL,
|
||||
PoolMemberHardState, PoolMemberIdentity, PoolMemberProbeStatus, StoredPoolMemberScore,
|
||||
};
|
||||
use aether_data_contracts::repository::provider_catalog::ProviderCatalogReadRepository;
|
||||
use aether_data_contracts::repository::usage::StoredRequestUsageAudit;
|
||||
@@ -20,6 +19,7 @@ use super::super::{
|
||||
build_router_with_state, sample_endpoint, sample_key, sample_provider, start_server, AppState,
|
||||
};
|
||||
use crate::admin_api::{maybe_build_local_admin_pool_response, AdminAppState, AdminRequestContext};
|
||||
use crate::ai_serving::{provider_key_pool_score_id, provider_key_pool_score_scope};
|
||||
use crate::audit::AdminAuditEvent;
|
||||
use crate::constants::{
|
||||
GATEWAY_HEADER, TRUSTED_ADMIN_MANAGEMENT_TOKEN_ID_HEADER, TRUSTED_ADMIN_SESSION_ID_HEADER,
|
||||
@@ -372,16 +372,19 @@ async fn gateway_handles_admin_pool_scores_locally_with_trusted_admin_principal(
|
||||
Vec::new(),
|
||||
vec![key.clone()],
|
||||
));
|
||||
let score_scope = provider_key_pool_score_scope();
|
||||
let score_identity = PoolMemberIdentity::provider_api_key("provider-openai", "key-openai-a");
|
||||
let score_id = provider_key_pool_score_id(&score_identity, &score_scope);
|
||||
let pool_score_repository = Arc::new(InMemoryPoolMemberScoreRepository::seed(vec![
|
||||
StoredPoolMemberScore {
|
||||
id: "pms-provider-openai-key-openai-a-openai-chat-model-1".to_string(),
|
||||
pool_kind: POOL_KIND_PROVIDER_KEY_POOL.to_string(),
|
||||
pool_id: "provider-openai".to_string(),
|
||||
member_kind: POOL_MEMBER_KIND_PROVIDER_API_KEY.to_string(),
|
||||
member_id: "key-openai-a".to_string(),
|
||||
capability: "openai:chat".to_string(),
|
||||
scope_kind: POOL_SCORE_SCOPE_KIND_MODEL.to_string(),
|
||||
scope_id: Some("model-1".to_string()),
|
||||
id: score_id,
|
||||
pool_kind: score_identity.pool_kind.clone(),
|
||||
pool_id: score_identity.pool_id.clone(),
|
||||
member_kind: score_identity.member_kind.clone(),
|
||||
member_id: score_identity.member_id.clone(),
|
||||
capability: score_scope.capability.clone(),
|
||||
scope_kind: score_scope.scope_kind.clone(),
|
||||
scope_id: score_scope.scope_id.clone(),
|
||||
score: 0.875,
|
||||
hard_state: PoolMemberHardState::Available,
|
||||
score_version: 1,
|
||||
@@ -414,7 +417,7 @@ async fn gateway_handles_admin_pool_scores_locally_with_trusted_admin_principal(
|
||||
|
||||
let response = reqwest::Client::new()
|
||||
.get(format!(
|
||||
"{gateway_url}/api/admin/pool/provider-openai/scores?api_format=openai:chat&model_id=model-1"
|
||||
"{gateway_url}/api/admin/pool/provider-openai/scores"
|
||||
))
|
||||
.header(crate::constants::GATEWAY_HEADER, "rust-phase3b")
|
||||
.header(TRUSTED_ADMIN_USER_ID_HEADER, "admin-user-123")
|
||||
@@ -430,6 +433,9 @@ async fn gateway_handles_admin_pool_scores_locally_with_trusted_admin_principal(
|
||||
Some(1)
|
||||
);
|
||||
assert_eq!(payload["items"][0]["member_id"], json!("key-openai-a"));
|
||||
assert_eq!(payload["items"][0]["capability"], json!("account"));
|
||||
assert_eq!(payload["items"][0]["scope_kind"], json!("account"));
|
||||
assert_eq!(payload["items"][0]["scope_id"], serde_json::Value::Null);
|
||||
assert_eq!(payload["items"][0]["key"]["name"], json!("score key"));
|
||||
assert_eq!(payload["items"][0]["probe_status"], json!("ok"));
|
||||
|
||||
@@ -734,14 +740,46 @@ async fn gateway_handles_admin_pool_list_keys_locally_with_trusted_admin_princip
|
||||
Vec::new(),
|
||||
vec![primary_key, cooldown_key, inactive_key],
|
||||
));
|
||||
let score_scope = provider_key_pool_score_scope();
|
||||
let score_identity = PoolMemberIdentity::provider_api_key("provider-openai", "key-openai-a");
|
||||
let pool_score_repository = Arc::new(InMemoryPoolMemberScoreRepository::seed(vec![
|
||||
StoredPoolMemberScore {
|
||||
id: provider_key_pool_score_id(&score_identity, &score_scope),
|
||||
pool_kind: score_identity.pool_kind.clone(),
|
||||
pool_id: score_identity.pool_id.clone(),
|
||||
member_kind: score_identity.member_kind.clone(),
|
||||
member_id: score_identity.member_id.clone(),
|
||||
capability: score_scope.capability.clone(),
|
||||
scope_kind: score_scope.scope_kind.clone(),
|
||||
scope_id: score_scope.scope_id.clone(),
|
||||
score: 0.875,
|
||||
hard_state: PoolMemberHardState::Available,
|
||||
score_version: 1,
|
||||
score_reason: json!({ "weights": { "manual_priority": 0.3 } }),
|
||||
last_ranked_at: Some(1_700_000_000),
|
||||
last_scheduled_at: Some(1_700_000_010),
|
||||
last_success_at: Some(1_700_000_020),
|
||||
last_failure_at: None,
|
||||
failure_count: 0,
|
||||
last_probe_attempt_at: Some(1_700_000_030),
|
||||
last_probe_success_at: Some(1_700_000_040),
|
||||
last_probe_failure_at: None,
|
||||
probe_failure_count: 0,
|
||||
probe_status: PoolMemberProbeStatus::Ok,
|
||||
updated_at: 1_700_000_050,
|
||||
},
|
||||
]));
|
||||
|
||||
let (upstream_url, upstream_handle) = start_server(upstream).await;
|
||||
let gateway = build_router_with_state(
|
||||
AppState::new()
|
||||
.expect("gateway should build")
|
||||
.with_data_state_for_tests(GatewayDataState::with_provider_catalog_reader_for_tests(
|
||||
provider_catalog_repository,
|
||||
)),
|
||||
.with_data_state_for_tests(
|
||||
GatewayDataState::with_provider_catalog_reader_for_tests(
|
||||
provider_catalog_repository,
|
||||
)
|
||||
.with_pool_score_repository_for_tests(pool_score_repository),
|
||||
),
|
||||
);
|
||||
let (gateway_url, gateway_handle) = start_server(gateway).await;
|
||||
|
||||
@@ -765,6 +803,9 @@ async fn gateway_handles_admin_pool_list_keys_locally_with_trusted_admin_princip
|
||||
let keys = payload["keys"].as_array().expect("keys should be array");
|
||||
assert_eq!(keys.len(), 2);
|
||||
assert_eq!(keys[0]["key_name"], json!("alpha"));
|
||||
assert_eq!(keys[0]["pool_score"]["score"], json!(0.875));
|
||||
assert_eq!(keys[0]["pool_score"]["scope_kind"], json!("account"));
|
||||
assert_eq!(keys[0]["pool_score"]["scope_id"], serde_json::Value::Null);
|
||||
assert_eq!(keys[0]["scheduling_status"], json!("available"));
|
||||
assert_eq!(keys[1]["key_name"], json!("beta"));
|
||||
assert_eq!(keys[1]["scheduling_reason"], json!("available"));
|
||||
|
||||
Reference in New Issue
Block a user