mirror of
https://github.com/fawney19/Aether.git
synced 2026-09-02 17:30:23 +08:00
Improve pool score probing rules
This commit is contained in:
@@ -1,6 +1,8 @@
|
||||
use crate::handlers::admin::admin_provider_pool_config;
|
||||
use crate::handlers::admin::provider::shared::paths::admin_provider_id_for_keys;
|
||||
use crate::handlers::admin::provider::shared::payloads::AdminProviderKeyCreateRequest;
|
||||
use crate::handlers::admin::request::{AdminAppState, AdminRequestContext};
|
||||
use crate::maintenance::ensure_provider_key_pool_scores_for_keys;
|
||||
use crate::provider_key_auth::provider_key_effective_api_formats;
|
||||
use crate::{model_fetch::perform_model_fetch_for_key, GatewayError};
|
||||
use axum::{
|
||||
@@ -98,6 +100,27 @@ pub(super) async fn maybe_handle(
|
||||
let endpoints = state
|
||||
.list_provider_catalog_endpoints_by_provider_ids(std::slice::from_ref(&provider.id))
|
||||
.await?;
|
||||
if let Some(pool_config) = admin_provider_pool_config(&provider) {
|
||||
let score_ensure_budget = (pool_config.score_fallback_scan_limit as usize).clamp(1, 50_000);
|
||||
if let Err(err) = ensure_provider_key_pool_scores_for_keys(
|
||||
state.as_ref(),
|
||||
&provider,
|
||||
&pool_config,
|
||||
&endpoints,
|
||||
std::slice::from_ref(&created),
|
||||
now_unix_secs,
|
||||
score_ensure_budget,
|
||||
)
|
||||
.await
|
||||
{
|
||||
tracing::debug!(
|
||||
provider_id = %provider.id,
|
||||
key_id = %created.id,
|
||||
error = ?err,
|
||||
"gateway admin provider key create: failed to seed pool score rows"
|
||||
);
|
||||
}
|
||||
}
|
||||
let api_formats =
|
||||
provider_key_effective_api_formats(&created, &provider.provider_type, &endpoints);
|
||||
|
||||
|
||||
@@ -1,6 +1,8 @@
|
||||
use crate::handlers::admin::admin_provider_pool_config;
|
||||
use crate::handlers::admin::provider::shared::paths::admin_update_key_id;
|
||||
use crate::handlers::admin::provider::shared::payloads::AdminProviderKeyUpdatePatch;
|
||||
use crate::handlers::admin::request::{AdminAppState, AdminRequestContext};
|
||||
use crate::maintenance::ensure_provider_key_pool_scores_for_keys;
|
||||
use crate::provider_key_auth::provider_key_effective_api_formats;
|
||||
use crate::{model_fetch::perform_model_fetch_for_key, GatewayError};
|
||||
use axum::{
|
||||
@@ -121,6 +123,27 @@ pub(super) async fn maybe_handle(
|
||||
let endpoints = state
|
||||
.list_provider_catalog_endpoints_by_provider_ids(std::slice::from_ref(&provider.id))
|
||||
.await?;
|
||||
if let Some(pool_config) = admin_provider_pool_config(&provider) {
|
||||
let score_ensure_budget = (pool_config.score_fallback_scan_limit as usize).clamp(1, 50_000);
|
||||
if let Err(err) = ensure_provider_key_pool_scores_for_keys(
|
||||
state.as_ref(),
|
||||
&provider,
|
||||
&pool_config,
|
||||
&endpoints,
|
||||
std::slice::from_ref(&updated),
|
||||
now_unix_secs,
|
||||
score_ensure_budget,
|
||||
)
|
||||
.await
|
||||
{
|
||||
tracing::debug!(
|
||||
provider_id = %provider.id,
|
||||
key_id = %updated.id,
|
||||
error = ?err,
|
||||
"gateway admin provider key update: failed to seed pool score rows"
|
||||
);
|
||||
}
|
||||
}
|
||||
let api_formats =
|
||||
provider_key_effective_api_formats(&updated, &provider.provider_type, &endpoints);
|
||||
|
||||
|
||||
@@ -1,6 +1,7 @@
|
||||
use crate::handlers::admin::provider::shared::support::{
|
||||
AdminProviderPoolConfig, AdminProviderPoolSchedulingPreset, AdminProviderPoolUnschedulableRule,
|
||||
};
|
||||
use aether_ai_serving::{PoolMemberScoreRules, PoolMemberScoreWeights};
|
||||
use serde_json::{Map, Value};
|
||||
|
||||
const POOL_ALLOWED_SCHEDULING_PRESETS: &[&str] = &[
|
||||
@@ -26,6 +27,123 @@ fn json_u64(value: &Value) -> Option<u64> {
|
||||
.or_else(|| value.as_i64().and_then(|raw| u64::try_from(raw).ok()))
|
||||
}
|
||||
|
||||
fn json_f64(value: &Value) -> Option<f64> {
|
||||
value.as_f64().or_else(|| {
|
||||
value
|
||||
.as_str()
|
||||
.map(str::trim)
|
||||
.filter(|value| !value.is_empty())
|
||||
.and_then(|value| value.parse::<f64>().ok())
|
||||
})
|
||||
}
|
||||
|
||||
fn pool_score_weight(object: &Map<String, Value>, names: &[&str], current: f64) -> f64 {
|
||||
names
|
||||
.iter()
|
||||
.find_map(|name| {
|
||||
object
|
||||
.get(*name)
|
||||
.and_then(json_f64)
|
||||
.filter(|value| value.is_finite() && *value >= 0.0)
|
||||
})
|
||||
.unwrap_or(current)
|
||||
}
|
||||
|
||||
fn parse_pool_score_weights(
|
||||
raw_weights: Option<&Map<String, Value>>,
|
||||
current: PoolMemberScoreWeights,
|
||||
) -> PoolMemberScoreWeights {
|
||||
let Some(raw_weights) = raw_weights else {
|
||||
return current;
|
||||
};
|
||||
PoolMemberScoreWeights {
|
||||
manual_priority: pool_score_weight(
|
||||
raw_weights,
|
||||
&["manual_priority", "priority", "internal_priority"],
|
||||
current.manual_priority,
|
||||
),
|
||||
health: pool_score_weight(raw_weights, &["health"], current.health),
|
||||
probe_freshness: pool_score_weight(
|
||||
raw_weights,
|
||||
&["probe_freshness", "freshness", "probe"],
|
||||
current.probe_freshness,
|
||||
),
|
||||
quota_remaining: pool_score_weight(
|
||||
raw_weights,
|
||||
&["quota_remaining", "quota", "quota_available"],
|
||||
current.quota_remaining,
|
||||
),
|
||||
latency: pool_score_weight(raw_weights, &["latency"], current.latency),
|
||||
cost_lru: pool_score_weight(
|
||||
raw_weights,
|
||||
&["cost_lru", "cost_remaining", "cost", "lru"],
|
||||
current.cost_lru,
|
||||
),
|
||||
}
|
||||
}
|
||||
|
||||
fn parse_pool_score_rules(pool_advanced: &Map<String, Value>) -> PoolMemberScoreRules {
|
||||
let mut rules = PoolMemberScoreRules::default();
|
||||
|
||||
for key in ["score_weights", "pool_score_weights", "scoring_weights"] {
|
||||
rules.weights = parse_pool_score_weights(
|
||||
pool_advanced.get(key).and_then(Value::as_object),
|
||||
rules.weights,
|
||||
);
|
||||
}
|
||||
|
||||
if let Some(score_rules) = pool_advanced
|
||||
.get("score_rules")
|
||||
.or_else(|| pool_advanced.get("pool_score_rules"))
|
||||
.and_then(Value::as_object)
|
||||
{
|
||||
rules.weights = parse_pool_score_weights(
|
||||
score_rules.get("weights").and_then(Value::as_object),
|
||||
rules.weights,
|
||||
);
|
||||
if let Some(ttl_seconds) = score_rules
|
||||
.get("probe_freshness_ttl_seconds")
|
||||
.or_else(|| score_rules.get("score_probe_freshness_ttl_seconds"))
|
||||
.and_then(json_u64)
|
||||
.filter(|value| *value > 0)
|
||||
{
|
||||
rules.probe_freshness_ttl_seconds = ttl_seconds.min(7 * 24 * 3600);
|
||||
}
|
||||
if let Some(cap) = score_rules
|
||||
.get("unschedulable_score_cap")
|
||||
.or_else(|| score_rules.get("hard_state_score_cap"))
|
||||
.and_then(json_f64)
|
||||
.filter(|value| value.is_finite())
|
||||
{
|
||||
rules.unschedulable_score_cap = cap.clamp(0.0, 1.0);
|
||||
}
|
||||
if let Some(penalty) = score_rules
|
||||
.get("probe_failure_penalty")
|
||||
.and_then(json_f64)
|
||||
.filter(|value| value.is_finite())
|
||||
{
|
||||
rules.probe_failure_penalty = penalty.clamp(0.0, 1.0);
|
||||
}
|
||||
if let Some(penalty) = score_rules
|
||||
.get("request_failure_penalty")
|
||||
.or_else(|| score_rules.get("runtime_failure_penalty"))
|
||||
.and_then(json_f64)
|
||||
.filter(|value| value.is_finite())
|
||||
{
|
||||
rules.request_failure_penalty = penalty.clamp(0.0, 1.0);
|
||||
}
|
||||
if let Some(threshold) = score_rules
|
||||
.get("probe_failure_cooldown_threshold")
|
||||
.or_else(|| score_rules.get("probe_failure_hard_state_threshold"))
|
||||
.and_then(json_u64)
|
||||
{
|
||||
rules.probe_failure_cooldown_threshold = threshold.min(100);
|
||||
}
|
||||
}
|
||||
|
||||
rules.effective()
|
||||
}
|
||||
|
||||
fn normalize_pool_preset_mode(preset: &str, raw_mode: Option<&Value>) -> Option<String> {
|
||||
match preset {
|
||||
"free_first" | "team_first" | "plus_first" | "pro_first" => {
|
||||
@@ -273,6 +391,7 @@ pub(crate) fn admin_provider_pool_config_from_config_value(
|
||||
probe_concurrency: 4,
|
||||
score_top_n: 128,
|
||||
score_fallback_scan_limit: 1024,
|
||||
score_rules: PoolMemberScoreRules::default(),
|
||||
stream_timeout_threshold: 3,
|
||||
stream_timeout_window_seconds: 1800,
|
||||
stream_timeout_cooldown_seconds: 300,
|
||||
@@ -281,6 +400,7 @@ pub(crate) fn admin_provider_pool_config_from_config_value(
|
||||
|
||||
let scheduling_presets = parse_pool_scheduling_presets(pool_advanced);
|
||||
let unschedulable_rules = parse_pool_unschedulable_rules(pool_advanced);
|
||||
let score_rules = parse_pool_score_rules(pool_advanced);
|
||||
|
||||
Some(AdminProviderPoolConfig {
|
||||
lru_enabled: admin_provider_pool_lru_enabled(&scheduling_presets),
|
||||
@@ -354,6 +474,7 @@ pub(crate) fn admin_provider_pool_config_from_config_value(
|
||||
.filter(|value| *value > 0)
|
||||
.map(|value| value.min(50_000))
|
||||
.unwrap_or(1024),
|
||||
score_rules,
|
||||
stream_timeout_threshold: pool_advanced
|
||||
.get("stream_timeout_threshold")
|
||||
.and_then(json_u64)
|
||||
@@ -429,6 +550,21 @@ mod tests {
|
||||
"probe_concurrency": 6,
|
||||
"score_top_n": 256,
|
||||
"score_fallback_scan_limit": 2048,
|
||||
"score_rules": {
|
||||
"weights": {
|
||||
"manual_priority": 0.4,
|
||||
"health": 0.2,
|
||||
"probe_freshness": 0.2,
|
||||
"quota_remaining": 0.1,
|
||||
"latency": 0.05,
|
||||
"cost_lru": 0.05
|
||||
},
|
||||
"probe_freshness_ttl_seconds": 1200,
|
||||
"unschedulable_score_cap": 0.03,
|
||||
"probe_failure_penalty": 0.08,
|
||||
"request_failure_penalty": 0.01,
|
||||
"probe_failure_cooldown_threshold": 2
|
||||
},
|
||||
"stream_timeout_threshold": 4,
|
||||
"stream_timeout_window_seconds": 900,
|
||||
"stream_timeout_cooldown_seconds": 180
|
||||
@@ -451,6 +587,13 @@ mod tests {
|
||||
assert_eq!(config.probe_concurrency, 6);
|
||||
assert_eq!(config.score_top_n, 256);
|
||||
assert_eq!(config.score_fallback_scan_limit, 2048);
|
||||
assert_eq!(config.score_rules.weights.manual_priority, 0.4);
|
||||
assert_eq!(config.score_rules.weights.health, 0.2);
|
||||
assert_eq!(config.score_rules.probe_freshness_ttl_seconds, 1200);
|
||||
assert_eq!(config.score_rules.unschedulable_score_cap, 0.03);
|
||||
assert_eq!(config.score_rules.probe_failure_penalty, 0.08);
|
||||
assert_eq!(config.score_rules.request_failure_penalty, 0.01);
|
||||
assert_eq!(config.score_rules.probe_failure_cooldown_threshold, 2);
|
||||
assert_eq!(config.stream_timeout_threshold, 4);
|
||||
assert_eq!(config.stream_timeout_window_seconds, 900);
|
||||
assert_eq!(config.stream_timeout_cooldown_seconds, 180);
|
||||
@@ -489,6 +632,27 @@ mod tests {
|
||||
assert_eq!(config.sticky_session_ttl_seconds, 0);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn parses_legacy_pool_score_weights_from_pool_advanced() {
|
||||
let config = admin_provider_pool_config_from_config_value(Some(&json!({
|
||||
"pool_advanced": {
|
||||
"scoring_weights": {
|
||||
"manual_priority": 0,
|
||||
"health": 2,
|
||||
"probe": 1,
|
||||
"quota_remaining": 0,
|
||||
"latency": 0,
|
||||
"cost_remaining": 1
|
||||
}
|
||||
}
|
||||
})))
|
||||
.expect("pool config should parse");
|
||||
|
||||
assert_eq!(config.score_rules.weights.health, 0.5);
|
||||
assert_eq!(config.score_rules.weights.probe_freshness, 0.25);
|
||||
assert_eq!(config.score_rules.weights.cost_lru, 0.25);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn parses_pool_config_from_generic_config_value() {
|
||||
let config = admin_provider_pool_config_from_config_value(Some(&json!({
|
||||
|
||||
@@ -618,6 +618,7 @@ mod tests {
|
||||
probe_concurrency: 4,
|
||||
score_top_n: 128,
|
||||
score_fallback_scan_limit: 1024,
|
||||
score_rules: aether_ai_serving::PoolMemberScoreRules::default(),
|
||||
stream_timeout_threshold: 3,
|
||||
stream_timeout_window_seconds: 1800,
|
||||
stream_timeout_cooldown_seconds: 300,
|
||||
|
||||
@@ -1,5 +1,6 @@
|
||||
use crate::handlers::admin::request::AdminAppState;
|
||||
use crate::LocalProviderDeleteTaskState;
|
||||
use aether_ai_serving::PoolMemberScoreRules;
|
||||
use serde_json::json;
|
||||
use std::collections::BTreeMap;
|
||||
|
||||
@@ -42,6 +43,7 @@ pub(crate) struct AdminProviderPoolConfig {
|
||||
pub(crate) probe_concurrency: u64,
|
||||
pub(crate) score_top_n: u64,
|
||||
pub(crate) score_fallback_scan_limit: u64,
|
||||
pub(crate) score_rules: PoolMemberScoreRules,
|
||||
pub(crate) stream_timeout_threshold: u64,
|
||||
pub(crate) stream_timeout_window_seconds: u64,
|
||||
pub(crate) stream_timeout_cooldown_seconds: u64,
|
||||
|
||||
Reference in New Issue
Block a user