[Router] Add composable scoring and eligibility policies (#37731)
Co-authored-by: inkcherry <mingzhi.liu@amd.com>
This commit is contained in:
@@ -12,10 +12,18 @@
|
||||
//! doesn't render tool schemas, so its ids would diverge from the engine).
|
||||
//! * A request with multimodal (array) content → `input_ids` omitted (a text
|
||||
//! tokenizer can't represent image content).
|
||||
//!
|
||||
//! The model id contains `deepseek-v4` so the tokenizer registry auto-attaches
|
||||
//! the built-in V4 chat encoder — the engine-equivalent path — without a
|
||||
//! template fixture.
|
||||
|
||||
use axum::body::Body;
|
||||
use axum::http::{Request, StatusCode};
|
||||
use serde_json::{json, Value};
|
||||
use sgl_router::config::{
|
||||
ActiveLoadConfig, CacheAwareConfig, Config, DiscoveryBackend, ModelConfig, ObservabilityConfig,
|
||||
PolicyKind, ProxyConfig, ServerConfig, StaticUrlsDiscoveryConfig,
|
||||
};
|
||||
use sgl_router::discovery::{ModelId, WorkerId, WorkerMode, WorkerSpec};
|
||||
use sgl_router::policies::factory::build_registry;
|
||||
use sgl_router::policies::kv_events::{BlockSizeOracle, HashTree};
|
||||
@@ -28,9 +36,35 @@ use std::sync::Arc;
|
||||
use std::time::Duration;
|
||||
use tower::ServiceExt;
|
||||
|
||||
use crate::common::cache_aware_fixture::{config, MODEL};
|
||||
use crate::common::mock_worker::MockWorker;
|
||||
|
||||
const MODEL: &str = "deepseek-v4-tiny";
|
||||
|
||||
fn config() -> Config {
|
||||
Config {
|
||||
server: ServerConfig {
|
||||
host: "0".into(),
|
||||
port: 0,
|
||||
},
|
||||
observability: ObservabilityConfig::default(),
|
||||
model: ModelConfig {
|
||||
id: MODEL.into(),
|
||||
tokenizer_path: "tests/fixtures/tiny_tokenizer.json".into(),
|
||||
policy: PolicyKind::CacheAwareZmq,
|
||||
circuit_breaker: None,
|
||||
cache_aware: Some(CacheAwareConfig::default()),
|
||||
sticky: None,
|
||||
fused: None,
|
||||
eligibility: None,
|
||||
},
|
||||
discovery: DiscoveryBackend::StaticUrls(StaticUrlsDiscoveryConfig {
|
||||
urls: vec!["http://placeholder:0".into()],
|
||||
}),
|
||||
proxy: ProxyConfig::default(),
|
||||
active_load: ActiveLoadConfig::default(),
|
||||
}
|
||||
}
|
||||
|
||||
fn build_ctx(url: String) -> Arc<AppContext> {
|
||||
let cfg = config();
|
||||
let tokenizers = Arc::new(TokenizerRegistry::load_from_config(&cfg).unwrap());
|
||||
|
||||
@@ -37,6 +37,8 @@ fn config_for(_worker_url: &str) -> Config {
|
||||
circuit_breaker: None,
|
||||
cache_aware: None,
|
||||
sticky: None,
|
||||
fused: None,
|
||||
eligibility: None,
|
||||
},
|
||||
discovery: DiscoveryBackend::StaticUrls(StaticUrlsDiscoveryConfig {
|
||||
urls: vec!["http://placeholder:0".into()],
|
||||
|
||||
@@ -29,6 +29,8 @@ pub fn config() -> Config {
|
||||
circuit_breaker: None,
|
||||
cache_aware: Some(CacheAwareConfig::default()),
|
||||
sticky: None,
|
||||
fused: None,
|
||||
eligibility: None,
|
||||
},
|
||||
discovery: DiscoveryBackend::StaticUrls(StaticUrlsDiscoveryConfig {
|
||||
urls: vec!["http://placeholder:0".into()],
|
||||
|
||||
@@ -41,6 +41,8 @@ async fn failover_when_one_worker_dies() {
|
||||
}),
|
||||
cache_aware: None,
|
||||
sticky: None,
|
||||
fused: None,
|
||||
eligibility: None,
|
||||
},
|
||||
discovery: DiscoveryBackend::StaticUrls(StaticUrlsDiscoveryConfig {
|
||||
urls: vec![w1.url.clone(), w2.url.clone(), w3.url.clone()],
|
||||
|
||||
@@ -47,6 +47,8 @@ fn build_ctx_with_worker(worker_url: &str) -> Arc<AppContext> {
|
||||
circuit_breaker: None,
|
||||
cache_aware: None,
|
||||
sticky: None,
|
||||
fused: None,
|
||||
eligibility: None,
|
||||
},
|
||||
discovery: DiscoveryBackend::StaticUrls(StaticUrlsDiscoveryConfig {
|
||||
urls: vec!["http://placeholder:0".into()],
|
||||
|
||||
@@ -34,6 +34,8 @@ async fn forwards_whitelisted_headers_strips_others() {
|
||||
circuit_breaker: None,
|
||||
cache_aware: None,
|
||||
sticky: None,
|
||||
fused: None,
|
||||
eligibility: None,
|
||||
},
|
||||
discovery: DiscoveryBackend::StaticUrls(StaticUrlsDiscoveryConfig {
|
||||
urls: vec!["http://placeholder:0".into()],
|
||||
|
||||
@@ -49,6 +49,8 @@ fn config() -> Config {
|
||||
circuit_breaker: None,
|
||||
cache_aware: None,
|
||||
sticky: None,
|
||||
fused: None,
|
||||
eligibility: None,
|
||||
},
|
||||
discovery: DiscoveryBackend::StaticUrls(StaticUrlsDiscoveryConfig {
|
||||
urls: vec!["http://placeholder:0".into()],
|
||||
|
||||
@@ -48,6 +48,8 @@ fn config() -> Config {
|
||||
circuit_breaker: None,
|
||||
cache_aware: None,
|
||||
sticky: None,
|
||||
fused: None,
|
||||
eligibility: None,
|
||||
},
|
||||
discovery: DiscoveryBackend::StaticUrls(StaticUrlsDiscoveryConfig {
|
||||
urls: vec!["http://placeholder:0".into()],
|
||||
|
||||
@@ -45,6 +45,8 @@ fn config() -> Config {
|
||||
circuit_breaker: None,
|
||||
cache_aware: None,
|
||||
sticky: None,
|
||||
fused: None,
|
||||
eligibility: None,
|
||||
},
|
||||
discovery: DiscoveryBackend::StaticUrls(StaticUrlsDiscoveryConfig {
|
||||
urls: vec!["http://placeholder:0".into()],
|
||||
|
||||
@@ -25,7 +25,7 @@ use axum::http::{Request, StatusCode};
|
||||
use serde_json::{json, Value};
|
||||
use sgl_router::config::{
|
||||
ActiveLoadConfig, Config, DiscoveryBackend, ModelConfig, ObservabilityConfig, PolicyKind,
|
||||
ProxyConfig, ServerConfig, StaticUrlsDiscoveryConfig, StickyConfig,
|
||||
ProxyConfig, ServerConfig, StaticUrlsDiscoveryConfig, StickyConfig, StickyFallbackKind,
|
||||
};
|
||||
use sgl_router::discovery::{ModelId, WorkerId, WorkerMode, WorkerSpec};
|
||||
use sgl_router::policies::factory::build_registry_with_defaults as build_policy_registry;
|
||||
@@ -60,10 +60,12 @@ fn config() -> Config {
|
||||
// mid-test; round-robin fallback for the initial pin of a key.
|
||||
sticky: Some(StickyConfig {
|
||||
header_name: HEADER.to_string(),
|
||||
fallback_policy: PolicyKind::RoundRobin,
|
||||
fallback_policy: StickyFallbackKind::RoundRobin,
|
||||
idle_secs: 3600,
|
||||
eviction_interval_secs: 3600,
|
||||
}),
|
||||
fused: None,
|
||||
eligibility: None,
|
||||
},
|
||||
discovery: DiscoveryBackend::StaticUrls(StaticUrlsDiscoveryConfig {
|
||||
urls: vec!["http://placeholder:0".into()],
|
||||
|
||||
@@ -8,7 +8,7 @@
|
||||
|
||||
use sgl_router::config::{
|
||||
ActiveLoadConfig, Config, DiscoveryBackend, ModelConfig, ObservabilityConfig, PolicyKind,
|
||||
ProxyConfig, ServerConfig, StaticUrlsDiscoveryConfig, StickyConfig,
|
||||
ProxyConfig, ServerConfig, StaticUrlsDiscoveryConfig, StickyConfig, StickyFallbackKind,
|
||||
};
|
||||
use sgl_router::discovery::{ModelId, WorkerId, WorkerMode, WorkerSpec};
|
||||
use sgl_router::policies::factory::build_registry_with_defaults as build_policy_registry;
|
||||
@@ -46,10 +46,12 @@ fn build_sticky_ctx(header_name: &str, worker_urls: &[String]) -> Arc<AppContext
|
||||
cache_aware: None,
|
||||
sticky: Some(StickyConfig {
|
||||
header_name: header_name.to_string(),
|
||||
fallback_policy: PolicyKind::RoundRobin,
|
||||
fallback_policy: StickyFallbackKind::RoundRobin,
|
||||
idle_secs: 3600,
|
||||
eviction_interval_secs: 3600,
|
||||
}),
|
||||
fused: None,
|
||||
eligibility: None,
|
||||
},
|
||||
discovery: DiscoveryBackend::StaticUrls(StaticUrlsDiscoveryConfig {
|
||||
urls: vec!["http://placeholder:0".into()],
|
||||
|
||||
@@ -41,6 +41,8 @@ fn config(_worker_url: &str) -> Config {
|
||||
circuit_breaker: None,
|
||||
cache_aware: None,
|
||||
sticky: None,
|
||||
fused: None,
|
||||
eligibility: None,
|
||||
},
|
||||
discovery: DiscoveryBackend::StaticUrls(StaticUrlsDiscoveryConfig {
|
||||
urls: vec!["http://placeholder:0".into()],
|
||||
|
||||
Reference in New Issue
Block a user