Added canaries for podgroup

engine/canaries.go

@@ -0,0 +1,84 @@
+package engine
+
+import (
+	"strings"
+
+	"github.com/laincloud/deployd/storage"
+	"github.com/mijia/sweb/log"
+)
+
+type IStrategy interface {
+	Type() string
+	Rules() []interface{}
+}
+
+type DefaultStrategy struct {
+	DivType  string        `json:"Type"`
+	DivDatas []interface{} `json:"Rule"`
+}
+
+func (ds *DefaultStrategy) Type() string {
+	return ds.DivType
+}
+
+func (ds *DefaultStrategy) Rules() []interface{} {
+	return ds.DivDatas
+}
+
+type Canary struct {
+	Strategies []IStrategy
+}
+
+func (canary *Canary) Equal(nCanary *Canary) bool {
+	if len(canary.Strategies) != len(nCanary.Strategies) {
+		return false
+	}
+	for i, strategy := range canary.Strategies {
+		if strategy.Type() != nCanary.Strategies[i].Type() {
+			return false
+		}
+	}
+	return true
+}
+
+type CanaryPodsWithSpec struct {
+	PodGroupSpec
+	canary *Canary
+}
+
+func (spec *CanaryPodsWithSpec) SaveCanary(store storage.Store) error {
+	key := strings.Join([]string{kLainDeploydRootKey, kLainCanaryKey, spec.Namespace, spec.Name}, "/")
+	if err := store.Set(key, spec.canary, true); err != nil {
+		log.Warnf("[Store] Failed to save pod group canary info %s, %s", key, err)
+		return err
+	}
+	return nil
+}
+
+func (spec *PodGroupSpec) RemoveCanary(store storage.Store) error {
+	key := strings.Join([]string{kLainDeploydRootKey, kLainCanaryKey, spec.Namespace, spec.Name}, "/")
+	if err := store.Remove(key); err != nil {
+		log.Warnf("[Store] Failed to remove pod group canary info %s, %s", key, err)
+		return err
+	}
+	return nil
+}
+
+func (spec *PodGroupSpec) FetchCanary(store storage.Store) (*Canary, error) {
+	key := strings.Join([]string{kLainDeploydRootKey, kLainCanaryKey, spec.Namespace, spec.Name}, "/")
+	var canary Canary
+	if err := store.Get(key, &canary); err != nil {
+		log.Warnf("[Store] Failed to fetch pod group canary info %s, %s", key, err)
+		return nil, err
+	}
+	return &canary, nil
+}
+
+func (spec *PodGroupSpec) UpdateCanary(store storage.Store, canary *Canary) error {
+	key := strings.Join([]string{kLainDeploydRootKey, kLainCanaryKey, spec.Namespace, spec.Name}, "/")
+	if err := store.Set(key, canary, true); err != nil {
+		log.Warnf("[Store] Failed to update pod group canary info %s, %s", key, err)
+		return err
+	}
+	return nil
+}

engine/canaries_test.go

@@ -0,0 +1,92 @@
+package engine
+
+import (
+	"encoding/json"
+	"testing"
+	"time"
+)
+
+func TestEngineCanaries(t *testing.T) {
+	etcdAddr := "http://127.0.0.1:2379"
+	ConfigPortsManager(etcdAddr)
+	c, store, err := initClusterAndStore()
+	if err != nil {
+		t.Fatalf("Cannot create the cluster and storage, %s", err)
+	}
+
+	engine, err := New(c, store)
+	if err != nil {
+		t.Fatalf("Cannot create the orc engine, %s", err)
+	}
+
+	namespace := "hello"
+	name := "hello.canary.web"
+	pgSpec := createPodGroupSpec(namespace, name, 1)
+	data := map[string]interface{}{"suffix": 1, "upstream": "beta1"}
+	strategies := []IStrategy{&DefaultStrategy{"uidsuffix", []interface{}{data}}}
+	canary := &Canary{strategies}
+	canarySpec := CanaryPodsWithSpec{pgSpec, canary}
+	if err := engine.NewCanary(canarySpec); err != nil {
+		t.Fatalf("Should not return error, %s", err)
+	}
+	if err := engine.NewCanary(canarySpec); err == nil {
+		t.Errorf("Should return exists error, but we got no problem")
+	}
+
+	time.Sleep(20 * time.Second)
+	if pg, ok := engine.InspectPodGroup(name); !ok {
+		t.Errorf("We should have the pod group, but we don't get it")
+	} else if pg.State != RunStateSuccess {
+		t.Errorf("We should have the pod deployed and running")
+	}
+
+	engine.RescheduleInstance(name, 3)
+	time.Sleep(20 * time.Second)
+	if pg, ok := engine.InspectPodGroup(name); !ok {
+		t.Errorf("We should have the pod group, but we don't get it")
+	} else if len(pg.Pods) != 3 {
+		t.Errorf("We should have 3 instance of the pods")
+	}
+
+	engine.RescheduleInstance(name, 1)
+	time.Sleep(30 * time.Second)
+	if pg, ok := engine.InspectPodGroup(name); !ok {
+		t.Errorf("We should have the pod group, but we don't get it")
+	} else if len(pg.Pods) != 1 {
+		bytes, err := json.Marshal(pg.Pods)
+		pods := ""
+		if err == nil {
+			pods = string(bytes)
+		}
+		t.Errorf("We should have 1 instance of the pods : %v", pods)
+	}
+
+	podSpec := createPodSpec(namespace, name)
+	podSpec.Containers[0].MemoryLimit = 24 * 1024 * 1024
+	engine.RescheduleSpec(name, podSpec)
+	time.Sleep(40 * time.Second)
+	if pg, ok := engine.InspectPodGroup(name); !ok {
+		t.Errorf("We should have the pod group, but we don't get it")
+	} else if pg.Spec.Version != 1 {
+		t.Errorf("We should have version 1 of the pods")
+	}
+
+	if nCanary := engine.FetchCanaryInfos(name); nCanary == nil {
+		t.Fatalf("Should return canary!")
+	} else {
+		if !nCanary.Equal(canary) {
+			t.Fatalf("Canary info should not changed")
+		}
+	}
+
+	if err := engine.RemovePodGroup(name); err != nil {
+		t.Errorf("We should be able to remove the pod group, %s", err)
+	} else if err := engine.NewPodGroup(pgSpec); err == nil {
+		t.Errorf("We should not be able to deploy pod group again in short time we remove it")
+	}
+	time.Sleep(20 * time.Second)
+
+	if nCanary := engine.FetchCanaryInfos(name); nCanary != nil {
+		t.Fatalf("Should not return nCanary")
+	}
+}

engine/engine.go

@@ -10,6 +10,7 @@ import (
 
 	"github.com/laincloud/deployd/cluster"
 	"github.com/laincloud/deployd/storage"
+	"github.com/laincloud/deployd/utils/util"
 	"github.com/mijia/adoc"
 	"github.com/mijia/sweb/log"
 )
@@ -43,18 +44,18 @@ type EngineConfig struct {
 type OrcEngine struct {
 	sync.RWMutex
 
-	config       EngineConfig
-	cluster      cluster.Cluster
-	store        storage.Store
-	eagleView    *RuntimeEagleView
-	pgCtrls      map[string]*podGroupController
-	rmPgCtrls    map[string]*podGroupController
-	dependsCtrls map[string]*dependsController
-	rmDepCtrls   map[string]*dependsController
-	opsChan      chan orcOperation
+	config         EngineConfig
+	cluster        cluster.Cluster
+	store          storage.Store
+	eagleView      *RuntimeEagleView
+	pgCtrls        map[string]*podGroupController
+	rmPgCtrls      map[string]*podGroupController
+	dependsCtrls   map[string]*dependsController
+	rmDepCtrls     map[string]*dependsController
+	opsChan        chan orcOperation
 	refreshAllChan chan bool
-	stop         chan struct{}
-	clstrFailCnt int
+	stop           chan struct{}
+	clstrFailCnt   int
 }
 
 const (
@@ -236,10 +237,16 @@ func (engine *OrcEngine) RemovePodGroup(name string) error {
 			return err
 		}
 		log.Infof("start delete %v\n", name)
+		// if is canary podgroup remove canary strategy
+		if util.PodGroupType(name) == PGCanaryType {
+			// remove key
+			pgCtrl.spec.RemoveCanary(engine.store)
+		}
 		engine.opsChan <- orcOperRemove{pgCtrl}
 		delete(engine.pgCtrls, name)
 		engine.rmPgCtrls[name] = pgCtrl
 		go engine.checkPodGroupRemoveResult(name, pgCtrl)
+
 		return nil
 	}
 }
@@ -284,7 +291,6 @@ func (engine *OrcEngine) RescheduleSpec(name string, podSpec PodSpec) error {
 			pgCtrl.opsChan <- pgOperSaveStore{true}
 			pgCtrl.opsChan <- pgOperOver{}
 		}
-
 		return nil
 	}
 }
@@ -329,6 +335,40 @@ func (engine *OrcEngine) ChangeState(pgName, op string, instance int) error {
 	return nil
 }
 
+func (engine *OrcEngine) NewCanary(spec CanaryPodsWithSpec) error {
+	if err := engine.NewPodGroup(spec.PodGroupSpec); err != nil {
+		return err
+	}
+	// gen canary strategy
+	spec.SaveCanary(engine.store)
+	return nil
+}
+
+func (engine *OrcEngine) FetchCanaryInfos(name string) *Canary {
+	engine.RLock()
+	defer engine.RUnlock()
+	if pgCtrl, ok := engine.pgCtrls[name]; !ok {
+		return nil
+	} else {
+		if canary, err := pgCtrl.spec.FetchCanary(engine.store); err != nil {
+			return nil
+		} else {
+			return canary
+		}
+	}
+	return nil
+}
+
+func (engine *OrcEngine) UpdateCanaryInfos(name string, canary *Canary) error {
+	engine.RLock()
+	defer engine.RUnlock()
+	if pgCtrl, ok := engine.pgCtrls[name]; !ok {
+		return ErrPodGroupNotExists
+	} else {
+		return pgCtrl.spec.UpdateCanary(engine.store, canary)
+	}
+}
+
 func (engine *OrcEngine) hasEnoughResource(pgCtrl *podGroupController, podSpec PodSpec) bool {
 	if resources, err := engine.cluster.GetResources(); err != nil {
 		return false

engine/podgroup.go

@@ -24,6 +24,16 @@ type PodGroupWithSpec struct {
 	PodGroup
 }
 
+// type PodGroupController interface {
+// 	Deploy()
+// 	Remove()
+// 	Inspect() PodGroupWithSpec
+// 	RescheduleSpec(podSpec PodSpec)
+// 	RescheduleInstance(numInstances int, restartPolicy ...RestartPolicy)
+// 	RescheduleDrift(fromNode, toNode string, instanceNo int, force bool)
+// 	ChangeState(op string, instance int)
+// }
+
 type podGroupController struct {
 	Publisher
 

engine/specs.go

@@ -22,6 +22,7 @@ const (
 	kLainNodesKey       = "nodes"
 	kLainLastPodSpecKey = "last_spec"
 	kLainPgOpingKey     = "operating"
+	kLainCanaryKey      = "canaries"
 
 	kLainLabelPrefix   = "cc.bdp.lain.deployd"
 	kLainLogVolumePath = "/lain/logs"
@@ -31,6 +32,8 @@ const (
 
 	MinPodKillTimeout = 10
 	MaxPodKillTimeout = 120
+
+	PGCanaryType = "canary"
 )
 
 var (

utils/util/util.go

@@ -48,6 +48,15 @@ func ParseContainerName(containerName string) (string, int, int, int, error) {
 	return g.Group(1), version, instance, driftCount, nil
 }
 
+func PodGroupType(procName string) string {
+	p := regex.MustCompile(`.*\.(.*)\..*`)
+	g := p.Match(procName)
+	if g == nil {
+		return ""
+	}
+	return g.Group(1)
+}
+
 func IpConflictErrorMatch(err string) string {
 	p := regex.MustCompile(`IP assignment error, data: {IP:([0-9.]+) HandleID:(.*)}: Address already assigned in block`)
 	g := p.Match(err)