add detailed failed task logging

pkg/deploy/mock_ecs.go

@@ -170,3 +170,32 @@ func (c MockECSClient) UpdateService(ctx context.Context, params *ecs.UpdateServ
 
 	return &out, nil
 }
+
+func (c MockECSClient) ListTasks(ctx context.Context, params *ecs.ListTasksInput, optFns ...func(*ecs.Options)) (*ecs.ListTasksOutput, error) {
+	output := &ecs.ListTasksOutput{
+		NextToken: aws.String("FOO"),
+		TaskArns: []string{
+			"arn-1",
+			"arn-2",
+		},
+	}
+
+	return output, nil
+}
+
+func (c MockECSClient) DescribeTasks(ctx context.Context, params *ecs.DescribeTasksInput, optFns ...func(*ecs.Options)) (*ecs.DescribeTasksOutput, error) {
+	output := &ecs.DescribeTasksOutput{
+		Tasks: []ecstypes.Task{
+			{
+				StoppedReason: aws.String("Error pulling image"),
+				TaskArn:       aws.String("arn-1"),
+			},
+			{
+				StoppedReason: aws.String("Essential container in task exited"),
+				TaskArn:       aws.String("arn-2"),
+			},
+		},
+	}
+
+	return output, nil
+}

pkg/deploy/service.go

@@ -90,6 +90,9 @@ func CheckDeploymentStatus(ctx context.Context, c types.ECSClient, service strin
 
 	if out.Services[0].Deployments[0].FailedTasks > 0 {
 		log.Printf("Deployment %d has failed tasks\n", out.Services[0].Deployments[0].FailedTasks)
+		// This is helpful for debugging
+		// ECS will clear all stopped tasks after a deployment finishes
+		showFailedTasks(c, service, cluster, deploymentID)
 		return true, errors.New("deployment failed")
 	}
 
@@ -101,6 +104,42 @@ func CheckDeploymentStatus(ctx context.Context, c types.ECSClient, service strin
 		// The only other status is FAILED
 		return true, errors.New("deployment failed")
 	}
+}
+
+func showFailedTasks(c types.ECSClient, service string, cluster string, deploymentID string) {
+	input := ecs.ListTasksInput{
+		Cluster:       &cluster,
+		DesiredStatus: "STOPPED",
+		StartedBy:     &deploymentID,
+		ServiceName:   &service,
+	}
+
+	resp, err := c.ListTasks(context.TODO(), &input)
+
+	if err != nil {
+		log.Printf("Error listing tasks: %s", err.Error())
+		return
+	}
+
+	tasks, err := c.DescribeTasks(
+		context.TODO(),
+		&ecs.DescribeTasksInput{
+			Tasks:   resp.TaskArns,
+			Cluster: &cluster,
+		},
+	)
+
+	if err != nil {
+		log.Printf("Error describing tasks: %s", err.Error())
+		return
+	}
+
+	for _, task := range tasks.Tasks {
+		stoppedReason := task.StoppedReason
+		taskARN := task.TaskArn
+
+		log.Printf("Task '%s' failure reason: %s", *taskARN, *stoppedReason)
+	}
 
 }
 

pkg/deploy/service_test.go

@@ -246,3 +246,31 @@ func TestGetServiceDesiredCount(t *testing.T) {
 		})
 	}
 }
+
+func Test_showFailedTasks(t *testing.T) {
+	type args struct {
+		c            types.ECSClient
+		service      string
+		cluster      string
+		deploymentID string
+	}
+	tests := []struct {
+		name string
+		args args
+	}{
+		{
+			name: "test-failed-task-logging",
+			args: args{
+				c:            MockECSClient{WantError: false, TestingT: t},
+				service:      "test-service",
+				cluster:      "test-cluster",
+				deploymentID: "ecs/deploy-foo",
+			},
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			showFailedTasks(tt.args.c, tt.args.service, tt.args.cluster, tt.args.deploymentID)
+		})
+	}
+}

pkg/types/types.go

@@ -12,6 +12,8 @@ type ECSClient interface {
 	RegisterTaskDefinition(ctx context.Context, params *ecs.RegisterTaskDefinitionInput, optFns ...func(*ecs.Options)) (*ecs.RegisterTaskDefinitionOutput, error)
 	DescribeServices(ctx context.Context, params *ecs.DescribeServicesInput, optFns ...func(*ecs.Options)) (*ecs.DescribeServicesOutput, error)
 	UpdateService(ctx context.Context, params *ecs.UpdateServiceInput, optFns ...func(*ecs.Options)) (*ecs.UpdateServiceOutput, error)
+	ListTasks(ctx context.Context, params *ecs.ListTasksInput, optFns ...func(*ecs.Options)) (*ecs.ListTasksOutput, error)
+	DescribeTasks(ctx context.Context, params *ecs.DescribeTasksInput, optFns ...func(*ecs.Options)) (*ecs.DescribeTasksOutput, error)
 }
 
 type AppAutoscalingClient interface {