feat: add scheduler host gc (#989)

Signed-off-by: Gaius <[email protected]>

docs/en/deployment/configuration/scheduler.yaml

@@ -34,14 +34,18 @@ scheduler:
   retryInterval: 1s
   # gc metadata configuration
   gc:
-    # peerGCInterval peer's gc interval
+    # peerGCInterval is peer's gc interval
     peerGCInterval: 10m
-    # peerTTL peer's TTL duration
+    # peerTTL is peer's TTL duration
     peerTTL: 24h
-    # taskGCInterval task's gc interval
+    # taskGCInterval is task's gc interval
     taskGCInterval: 10m
-    # taskTTL task's TTL duration
+    # taskTTL is task's TTL duration
     taskTTL: 24h
+    # hostGCInterval is host's gc interval
+    hostGCInterval: 30m
+    # hostTTL is host's TTL duration
+    hostTTL: 48h
 
 # dynamic data configuration
 dynConfig:

docs/zh-CN/deployment/configuration/scheduler.yaml

@@ -40,6 +40,10 @@ scheduler:
     taskGCInterval: 10m
     # 不活跃的 task 的存活时间
     taskTTL: 24h
+    # host 的回收间隔
+    hostGCInterval: 30m
+    # 不活跃的 host 的存活时间
+    hostTTL: 48h
 
 # 动态数据配置
 dynConfig:

scheduler/config/config.go

@@ -71,6 +71,8 @@ func New() *Config {
 				PeerTTL:        24 * time.Hour,
 				TaskGCInterval: 10 * time.Minute,
 				TaskTTL:        24 * time.Hour,
+				HostGCInterval: 30 * time.Minute,
+				HostTTL:        48 * time.Hour,
 			},
 		},
 		DynConfig: &DynConfig{
@@ -254,6 +256,12 @@ type GCConfig struct {
 
 	// Task time to live
 	TaskTTL time.Duration `yaml:"taskTTL" mapstructure:"taskTTL"`
+
+	// Host gc interval
+	HostGCInterval time.Duration `yaml:"hostGCInterval" mapstructure:"hostGCInterval"`
+
+	// Host time to live
+	HostTTL time.Duration `yaml:"hostTTL" mapstructure:"hostTTL"`
 }
 
 type DynConfig struct {

scheduler/config/config_test.go

@@ -48,6 +48,8 @@ func TestConfig_Load(t *testing.T) {
 				PeerTTL:        5 * time.Minute,
 				TaskGCInterval: 1 * time.Minute,
 				TaskTTL:        10 * time.Minute,
+				HostGCInterval: 1 * time.Minute,
+				HostTTL:        10 * time.Minute,
 			},
 		},
 		DynConfig: &DynConfig{

scheduler/config/testdata/scheduler.yaml

@@ -16,6 +16,8 @@ scheduler:
     peerTTL: 300000000000
     taskGCInterval: 60000000000
     taskTTL: 600000000000
+    hostGCInterval: 60000000000
+    hostTTL: 600000000000
 
 dynconfig:
   refreshInterval: 300000000000

scheduler/resource/host_manager.go

@@ -18,6 +18,15 @@ package resource
 
 import (
 	"sync"
+	"time"
+
+	pkggc "d7y.io/dragonfly/v2/pkg/gc"
+	"d7y.io/dragonfly/v2/scheduler/config"
+)
+
+const (
+	// GC host id
+	GCHostID = "host"
 )
 
 type HostManager interface {
@@ -34,16 +43,36 @@ type HostManager interface {
 
 	// Delete deletes host for a key
 	Delete(string)
+
+	// Try to reclaim host
+	RunGC() error
 }
 
 type hostManager struct {
 	// Host sync map
 	*sync.Map
+
+	// Host time to live
+	ttl time.Duration
 }
 
 // New host manager interface
-func newHostManager() HostManager {
-	return &hostManager{&sync.Map{}}
+func newHostManager(cfg *config.GCConfig, gc pkggc.GC) (HostManager, error) {
+	h := &hostManager{
+		Map: &sync.Map{},
+		ttl: cfg.HostTTL,
+	}
+
+	if err := gc.Add(pkggc.Task{
+		ID:       GCHostID,
+		Interval: cfg.HostGCInterval,
+		Timeout:  cfg.HostGCInterval,
+		Runner:   h,
+	}); err != nil {
+		return nil, err
+	}
+
+	return h, nil
 }
 
 func (h *hostManager) Load(key string) (*Host, bool) {
@@ -67,3 +96,19 @@ func (h *hostManager) LoadOrStore(host *Host) (*Host, bool) {
 func (h *hostManager) Delete(key string) {
 	h.Map.Delete(key)
 }
+
+func (h *hostManager) RunGC() error {
+	h.Map.Range(func(_, value interface{}) bool {
+		host := value.(*Host)
+		elapsed := time.Since(host.UpdateAt.Load())
+
+		if elapsed > h.ttl && host.LenPeers() == 0 {
+			host.Log.Info("host has been reclaimed")
+			h.Delete(host.ID)
+		}
+
+		return true
+	})
+
+	return nil
+}