[sgl-router] refactor - move policy-required states under src/state (#40272)
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
This commit is contained in:
co-authored by
Claude Fable 5.1
parent
acd20a516e
commit
4a9dc5c4af
@@ -10,17 +10,17 @@ use axum::body::Body;
|
||||
use axum::http::{Request, StatusCode};
|
||||
use sgl_kv_indexer::{PrefixIndex, PrefixIndexError, PrefixMatch, PrefixOutcome};
|
||||
use sgl_router::config::{
|
||||
ActiveLoadConfig, AffinityConfig, BucketConfig, BucketSpec, BucketStage, CacheAwareConfig,
|
||||
CachePrefixProvider, Config, DiscoveryBackend, KvIndexerEndpointConfig, ModelConfig,
|
||||
AffinityConfig, BucketConfig, BucketSpec, BucketStage, CacheAwareConfig, CachePrefixProvider,
|
||||
Config, DiscoveryBackend, InflightLoadConfig, KvIndexerEndpointConfig, ModelConfig,
|
||||
ObservabilityConfig, PolicyKind, ProxyConfig, ServerConfig, SessionAffinityMode,
|
||||
SloBucketPolicy, StaticUrlsDiscoveryConfig,
|
||||
};
|
||||
use sgl_router::discovery::{ModelId, WorkerId, WorkerMode, WorkerSpec};
|
||||
use sgl_router::policies::engine_load::{LoadStat, NativeCacheRankLoad};
|
||||
use sgl_router::policies::factory::build_registry_with_defaults;
|
||||
use sgl_router::proxy::Proxy;
|
||||
use sgl_router::server::app::build_router;
|
||||
use sgl_router::server::app_context::AppContext;
|
||||
use sgl_router::state::load_monitor::engine_reported_load::{LoadStat, NativeCacheRankLoad};
|
||||
use sgl_router::tokenizer::TokenizerRegistry;
|
||||
use sgl_router::workers::WorkerRegistry;
|
||||
use std::sync::atomic::{AtomicUsize, Ordering};
|
||||
@@ -77,7 +77,7 @@ fn build_app_context(
|
||||
urls: vec!["http://placeholder:0".into()],
|
||||
}),
|
||||
proxy: ProxyConfig::default(),
|
||||
active_load: ActiveLoadConfig::default(),
|
||||
router_inflight_load: InflightLoadConfig::default(),
|
||||
};
|
||||
let tokenizers = Arc::new(TokenizerRegistry::load_from_config(&config).unwrap());
|
||||
let registry = Arc::new(WorkerRegistry::default());
|
||||
@@ -236,7 +236,7 @@ fn set_native_load_with_waiting(
|
||||
max_total_num_tokens: u64,
|
||||
num_waiting_uncached_tokens: u64,
|
||||
) {
|
||||
ctx.engine_load.set(
|
||||
ctx.engine_reported_load.set(
|
||||
worker_url,
|
||||
0,
|
||||
LoadStat {
|
||||
|
||||
Reference in New Issue
Block a user