Merge pull request #5 from /issues/4

feat: --skipCanaryオプション追加

cli/cage/main.go

@@ -28,11 +28,12 @@ func main() {
 		AvailabilityThreshold:       aws.Float64(-1.0),
 		ResponseTimeThreshold:       aws.Float64(-1.0),
 		RollOutPeriod:               aws.Int64(-1),
+		SkipCanary:                  aws.Bool(false),
 	}
 	configPath := ""
 	app := cli.NewApp()
 	app.Name = "canarycage"
-	app.Version = "1.0.0-alpha2"
+	app.Version = "1.1.0-alpha"
 	app.Description = "A gradual roll-out deployment tool for AWS ECS"
 	app.Flags = []cli.Flag{
 		cli.StringFlag{
@@ -46,7 +47,7 @@ func main() {
 			Usage: "generate config file skeleton json",
 		},
 		cli.BoolFlag{
-			Name: "dryRun",
+			Name:  "dryRun",
 			Usage: "describe roll out plan without affecting any resources",
 		},
 		cli.StringFlag{
@@ -113,10 +114,16 @@ func main() {
 			Value:       300,
 			Destination: envars.RollOutPeriod,
 		},
+		cli.BoolFlag{
+			Name:        "skipCanary",
+			EnvVar:      cage.SkipCanaryKey,
+			Usage:       "skip canary test. ensuring only healthy tasks.",
+			Destination: envars.SkipCanary,
+		},
 	}
 	app.Action = func(ctx *cli.Context) {
 		if ctx.Bool("skeleton") {
-			d, err := json.MarshalIndent(envars,"","\t")
+			d, err := json.MarshalIndent(envars, "", "\t")
 			if err != nil {
 				log.Fatalf("failed to marshal json due to: %s", err)
 			}
@@ -151,7 +158,7 @@ func DryRun(envars *cage.Envars) {
 	log.Infof("== [DRY RUN] ==")
 	d, _ := json.MarshalIndent(envars, "", "\t")
 	log.Infof("envars = \n%s", string(d))
-	if envars.NextTaskDefinitionArn == nil{
+	if envars.NextTaskDefinitionArn == nil {
 		log.Info("create NEXT task definition with provided json")
 	}
 	log.Infof("create NEXT service '%s' with desiredCount=1", *envars.NextServiceName)

env.go

@@ -17,6 +17,7 @@ type Envars struct {
 	AvailabilityThreshold       *float64 `json:"availabilityThreshold" type:"double"`
 	ResponseTimeThreshold       *float64 `json:"responseTimeThreshold" type:"double"`
 	RollOutPeriod               *int64   `json:"rollOutPeriod" type:"integer"`
+	SkipCanary                  *bool    `json:"skipCanary" type:"bool"`
 }
 
 // required
@@ -35,6 +36,7 @@ const RegionKey = "CAGE_AWS_REGION"
 const AvailabilityThresholdKey = "CAGE_AVAILABILITY_THRESHOLD"
 const ResponseTimeThresholdKey = "CAGE_RESPONSE_TIME_THRESHOLD"
 const RollOutPeriodKey = "CAGE_ROLL_OUT_PERIOD"
+const SkipCanaryKey = "CAGE_SKIP_CANARY"
 
 const kAvailabilityThresholdDefaultValue = 0.9970
 const kResponseTimeThresholdDefaultValue = 1.0
@@ -81,5 +83,8 @@ func EnsureEnvars(
 	if period := *dest.RollOutPeriod; !(60 <= period && float64(period) != math.NaN() && float64(period) != math.Inf(0)) {
 		return NewErrorf("--rollOutPeriod [%s] must be lesser than 60, but got '%d'", RollOutPeriodKey, period)
 	}
+	if dest.SkipCanary == nil {
+		dest.SkipCanary = aws.Bool(false)
+	}
 	return nil
 }

rollout.go

@@ -158,7 +158,9 @@ func (envars *Envars) RollOut(
 			return err
 		}
 		// Phase2: service-nextのperiodic healthを計測
-		if err := envars.CanaryTest(ctx.Cw, tg.LoadBalancerArns[0], tg.TargetGroupArn, totalRollOutCnt); err != nil {
+		if *envars.SkipCanary {
+			log.Infof("🤫 %dth canary test skipped.", totalRollOutCnt)
+		} else if err := envars.CanaryTest(ctx.Cw, tg.LoadBalancerArns[0], tg.TargetGroupArn, totalRollOutCnt); err != nil {
 			return err
 		}
 		// Phase3: service-currentからタスクを指定数消す
@@ -195,7 +197,9 @@ func (envars *Envars) RollOut(
 		if oldTaskCount == 0 && newTaskCount >= originalDesiredCount {
 			// ロールアウトが終わったら最終検証を行う
 			log.Infof("estimated roll out completed. Do final canary test...")
-			if err := envars.CanaryTest(ctx.Cw, tg.LoadBalancerArns[0], tg.TargetGroupArn, totalRollOutCnt); err != nil {
+			if *envars.SkipCanary {
+				log.Infof("😑 final canary test skipped...")
+			} else if err := envars.CanaryTest(ctx.Cw, tg.LoadBalancerArns[0], tg.TargetGroupArn, totalRollOutCnt); err != nil {
 				log.Errorf("final canary test has failed due to: %s", err)
 				return err
 			}

rollout_test.go

@@ -102,7 +102,7 @@ func TestEnvars_StartGradualRollOut(t *testing.T) {
 			t.Fatalf("%s", err)
 		}
 		assert.Nil(t, result.HandledError)
-		assert.False(t,  *result.Rolledback)
+		assert.False(t, *result.Rolledback)
 		assert.Equal(t, int64(1), mctx.ServiceSize())
 		assert.Equal(t, v, mctx.TaskSize())
 	}
@@ -189,6 +189,22 @@ func TestEnvars_StartGradualRollOut3(t *testing.T) {
 	assert.Equal(t, int64(4), mocker.TaskSize())
 }
 
+func TestEnvars_StartGradualRollOut4(t *testing.T) {
+	// skipCanaryオプションを追加した場合canaryTestは行わない
+	envars := DefaultEnvars()
+	envars.SkipCanary = aws.Bool(true)
+	ctrl := gomock.NewController(t)
+	_, ctx := envars.Setup(ctrl, 2)
+	cwMock := mock_cloudwatch.NewMockCloudWatchAPI(ctrl)
+	cwMock.EXPECT().GetMetricStatistics(gomock.Any()).Times(0)
+	ctx.Cw = cwMock
+	if res, err := envars.StartGradualRollOut(ctx); err != nil {
+		t.Fatalf(err.Error())
+	} else if res.HandledError != nil {
+		t.Fatalf(err.Error())
+	}
+}
+
 func TestEnvars_Rollback(t *testing.T) {
 	log.SetLevel(log.DebugLevel)
 	newTimer = fakeTimer

test-integration/integration_test.go

@@ -234,6 +234,25 @@ func TestHealthyToHealthy(t *testing.T) {
 	assert.False(t, *result.Rolledback)
 }
 
+func TestHealthyToHealthySkipCanary(t *testing.T) {
+	ctx := setup()
+	envars := setupEnvars()
+	envars.SkipCanary = aws.Bool(true)
+	envars.NextTaskDefinitionArn = aws.String(kHealthyTDArn)
+	envars.CurrentServiceName = aws.String(kCurrentServiceName)
+	envars.NextServiceName = aws.String(kNextServiceName)
+	defer func() {
+		cleanupService(ctx.Ecs, envars, envars.CurrentServiceName)
+		cleanupService(ctx.Ecs, envars, envars.NextServiceName)
+	}()
+	result, err := testInternal(t, envars)
+	if err != nil {
+		t.Fatalf(err.Error())
+	}
+	assert.Nil(t, result.HandledError)
+	assert.False(t, *result.Rolledback)
+}
+
 func testAbnormal(t *testing.T, tdarn string, servicePostfix string) error {
 	log.SetLevel(log.InfoLevel)
 	ctx := setup()