[Feature] Config eviction_duration (PaddlePaddle#7125)

* [Feature] Config eviction_duration

* [Feature] Config eviction_duration

* [Feature] Config eviction_duration

* [Feature] Config eviction_duration

---------

Co-authored-by: mouxin <mouxin@baidu.com>

Author: mouxinqq

docs/online_serving/router.md

@@ -195,6 +195,7 @@ scheduler:
   prefill-policy: "cache_aware" # Prefill scheduling policy in PD mode
   decode-policy: "request_num" # Decode scheduling policy in PD mode
   eviction-interval-secs: 60 # Cache eviction interval for CacheAware scheduling
+  eviction-duration-mins: 30 # Eviction duration for cache-aware radix tree nodes (minutes); default: 30
   balance-abs-threshold: 1 # Absolute threshold for CacheAware balancing
   balance-rel-threshold: 0.2 # Relative threshold for CacheAware balancing
   hit-ratio-weight: 1.0 # Cache hit ratio weight

docs/zh/online_serving/router.md

@@ -195,6 +195,7 @@ scheduler:
   prefill-policy: "cache_aware" # pd分离模式下prefill节点调度策略; 默认: process_tokens
   decode-policy: "request_num" # pd分离模式下decode节点调度策略; 默认: request_num
   eviction-interval-secs: 60 # cache-aware策略清理过期cache的间隔时间
+  eviction-duration-mins: 30 # cache-aware策略radix tree节点驱逐时间(分钟); 默认: 30
   balance-abs-threshold: 1 # cache-aware策略绝对阈值
   balance-rel-threshold: 0.2 # cache-aware策略相对阈值
   hit-ratio-weight: 1.0 # cache-aware策略命中率权重

fastdeploy/golang_router/examples/run_with_config/config/config.example.yaml

@@ -9,6 +9,7 @@ scheduler:
   prefill-policy: "cache_aware"
   decode-policy: "request_num"
   eviction-interval-secs: 60
+  eviction-duration-mins: 30 # eviction duration for cache-aware radix tree nodes (minutes); default: 30
   balance-abs-threshold: 1
   balance-rel-threshold: 0.2
   hit-ratio-weight: 1.0

fastdeploy/golang_router/examples/run_with_default_workers/config/config.example.yaml

@@ -9,6 +9,7 @@ scheduler:
   prefill-policy: "cache_aware"
   decode-policy: "request_num"
   eviction-interval-secs: 60
+  eviction-duration-mins: 30 # eviction duration for cache-aware radix tree nodes (minutes); default: 30
   balance-abs-threshold: 1
   balance-rel-threshold: 0.2
   hit-ratio-weight: 1.0

fastdeploy/golang_router/internal/config/config.go

@@ -36,6 +36,7 @@ type SchedulerConfig struct {
 	PrefillPolicy        string  `yaml:"prefill-policy"`
 	DecodePolicy         string  `yaml:"decode-policy"`
 	EvictionIntervalSecs float64 `yaml:"eviction-interval-secs"`
+	EvictionDurationMins float64 `yaml:"eviction-duration-mins"`
 	CacheBlockSize       int     `yaml:"cache-block-size"`
 	TokenizerURL         string  `yaml:"tokenizer-url"`
 	TokenizerTimeoutSecs float64 `yaml:"tokenizer-timeout-secs"`
@@ -98,6 +99,9 @@ func Load(configPath, listenPort string, isSplitwise bool) (*Config, error) {
 	if cfg.Scheduler.EvictionIntervalSecs == 0 {
 		cfg.Scheduler.EvictionIntervalSecs = 60
 	}
+	if cfg.Scheduler.EvictionDurationMins == 0 {
+		cfg.Scheduler.EvictionDurationMins = 30
+	}
 	if cfg.Scheduler.CacheBlockSize == 0 {
 		cfg.Scheduler.CacheBlockSize = 64
 	}