Home

Awesome

Build Status GoDoc Go Report Card Coverage Status

Go-Marathon

Go-marathon is a API library for working with Marathon. It currently supports

Note: the library is still under active development; users should expect frequent (possibly breaking) API changes for the time being.

It requires Go version 1.6 or higher.

Code Examples

There is also an examples directory in the source which shows hints and snippets of code of how to use it — which is probably the best place to start.

You can use examples/docker-compose.yml in order to start a test cluster.

Creating a client

import (
	marathon "github.com/gambol99/go-marathon"
)

marathonURL := "http://10.241.1.71:8080"
config := marathon.NewDefaultConfig()
config.URL = marathonURL
client, err := marathon.NewClient(config)
if err != nil {
	log.Fatalf("Failed to create a client for marathon, error: %s", err)
}

applications, err := client.Applications(nil)
...

Note, you can also specify multiple endpoint for Marathon (i.e. you have setup Marathon in HA mode and having multiple running)

marathonURL := "http://10.241.1.71:8080,10.241.1.72:8080,10.241.1.73:8080"

The first one specified will be used, if that goes offline the member is marked as "unavailable" and a background process will continue to ping the member until it's back online.

You can also pass a custom path to the URL, which is especially needed in case of DCOS:

marathonURL := "http://10.241.1.71:8080/cluster,10.241.1.72:8080/cluster,10.241.1.73:8080/cluster"

If you specify a DCOSToken in the configuration file but do not pass a custom URL path, /marathon will be used.

Customizing the HTTP Clients

HTTP clients with reasonable timeouts are used by default. It is possible to pass custom clients to the configuration though if the behavior should be customized (e.g., to bypass TLS verification, load root CAs, or change timeouts).

Two clients can be given independently of each other:

If no HTTPSSEClient is given but an HTTPClient is, it will be used for SSE subscriptions as well (thereby overriding the default SSE HTTP client).

marathonURL := "http://10.241.1.71:8080"
config := marathon.NewDefaultConfig()
config.URL = marathonURL
config.HTTPClient = &http.Client{
    Timeout: (time.Duration(10) * time.Second),
    Transport: &http.Transport{
        Dial: (&net.Dialer{
            Timeout:   10 * time.Second,
            KeepAlive: 10 * time.Second,
        }).Dial,
        TLSClientConfig: &tls.Config{
            InsecureSkipVerify: true,
        },
    },
}
config.HTTPSSEClient = &http.Client{
    // Invalid to set Timeout as it contains timeout for reading a response body
    Transport: &http.Transport{
        Dial: (&net.Dialer{
            Timeout:   10 * time.Second,
            KeepAlive: 10 * time.Second,
        }).Dial,
        TLSClientConfig: &tls.Config{
            InsecureSkipVerify: true,
        },
    },
}

Listing the applications

applications, err := client.Applications(nil)
if err != nil {
	log.Fatalf("Failed to list applications: %s", err)
}

log.Printf("Found %d application(s) running", len(applications.Apps))
for _, application := range applications.Apps {
	log.Printf("Application: %s", application)
	appID := application.ID

	details, err := client.Application(appID)
	if err != nil {
		log.Fatalf("Failed to get application %s: %s", appID, err)
	}
	if details.Tasks != nil {
		for _, task := range details.Tasks {
			log.Printf("application %s has task: %s", appID, task)
		}
	}
}

Creating a new application

log.Printf("Deploying a new application")
application := marathon.NewDockerApplication().
  Name(applicationName).
  CPU(0.1).
  Memory(64).
  Storage(0.0).
  Count(2).
  AddArgs("/usr/sbin/apache2ctl", "-D", "FOREGROUND").
  AddEnv("NAME", "frontend_http").
  AddEnv("SERVICE_80_NAME", "test_http").
  CheckHTTP("/health", 10, 5)

application.
  Container.Docker.Container("quay.io/gambol99/apache-php:latest").
  Bridged().
  Expose(80).
  Expose(443)

if _, err := client.CreateApplication(application); err != nil {
	log.Fatalf("Failed to create application: %s, error: %s", application, err)
} else {
	log.Printf("Created the application: %s", application)
}

Note: Applications may also be defined by means of initializing a marathon.Application struct instance directly. However, go-marathon's DSL as shown above provides a more concise way to achieve the same.

Scaling application

Change the number of application instances to 4

log.Printf("Scale to 4 instances")
if err := client.ScaleApplicationInstances(application.ID, 4); err != nil {
	log.Fatalf("Failed to delete the application: %s, error: %s", application, err)
} else {
	client.WaitOnApplication(application.ID, 30 * time.Second)
	log.Printf("Successfully scaled the application")
}

Pods

Pods allow you to deploy groups of tasks as a unit. All tasks in a single instance of a pod share networking and storage. View the Marathon documentation for more details on this feature.

Examples of their usage can be seen in the examples/pods directory, and a smaller snippet is below.

// Initialize a single-container pod running nginx
pod := marathon.NewPod()

image := marathon.NewDockerPodContainerImage().SetID("nginx")

container := marathon.NewPodContainer().
	SetName("container", i).
	CPUs(0.1).
	Memory(128).
	SetImage(image)

pod.Name("mypod").AddContainer(container)

// Create it and wait for it to start up
pod, err := client.CreatePod(pod)
err = client.WaitOnPod(pod.ID, time.Minute*1)

// Scale it
pod.Count(5)
pod, err = client.UpdatePod(pod, true)

// Delete it
id, err := client.DeletePod(pod.ID, true)

Subscription & Events

Request to listen to events related to applications — namely status updates, health checks changes and failures. There are two different event transports controlled by EventsTransport setting with the following possible values: EventsTransportSSE and EventsTransportCallback (default value). See Event Stream and Event Subscriptions for details.

Event subscriptions can also be individually controlled with the Subscribe and Unsubscribe functions. See Controlling subscriptions for more details.

Event Stream

Only available in Marathon >= 0.9.0. Does not require any special configuration or prerequisites.

// Configure client
config := marathon.NewDefaultConfig()
config.URL = marathonURL
config.EventsTransport = marathon.EventsTransportSSE

client, err := marathon.NewClient(config)
if err != nil {
	log.Fatalf("Failed to create a client for marathon, error: %s", err)
}

// Register for events
events, err = client.AddEventsListener(marathon.EventIDApplications)
if err != nil {
	log.Fatalf("Failed to register for events, %s", err)
}

timer := time.After(60 * time.Second)
done := false

// Receive events from channel for 60 seconds
for {
	if done {
		break
	}
	select {
	case <-timer:
		log.Printf("Exiting the loop")
		done = true
	case event := <-events:
		log.Printf("Received event: %s", event)
	}
}

// Unsubscribe from Marathon events
client.RemoveEventsListener(events)

Event Subscriptions

Requires to start a built-in web server accessible by Marathon to connect and push events to. Consider the following additional settings:

// Configure client
config := marathon.NewDefaultConfig()
config.URL = marathonURL
config.EventsInterface = marathonInterface
config.EventsPort = marathonPort

client, err := marathon.NewClient(config)
if err != nil {
	log.Fatalf("Failed to create a client for marathon, error: %s", err)
}

// Register for events
events, err = client.AddEventsListener(marathon.EventIDApplications)
if err != nil {
	log.Fatalf("Failed to register for events, %s", err)
}

timer := time.After(60 * time.Second)
done := false

// Receive events from channel for 60 seconds
for {
	if done {
		break
	}
	select {
	case <-timer:
		log.Printf("Exiting the loop")
		done = true
	case event := <-events:
		log.Printf("Received event: %s", event)
	}
}

// Unsubscribe from Marathon events
client.RemoveEventsListener(events)

See events.go for a full list of event IDs.

Controlling subscriptions

If you simply want to (de)register event subscribers (i.e. without starting an internal web server) you can use the Subscribe and Unsubscribe methods.

// Configure client
config := marathon.NewDefaultConfig()
config.URL = marathonURL

client, err := marathon.NewClient(config)
if err != nil {
	log.Fatalf("Failed to create a client for marathon, error: %s", err)
}

// Register an event subscriber via a callback URL
callbackURL := "http://10.241.1.71:9494"
if err := client.Subscribe(callbackURL); err != nil {
	log.Fatalf("Unable to register the callbackURL [%s], error: %s", callbackURL, err)
}

// Deregister the same subscriber
if err := client.Unsubscribe(callbackURL); err != nil {
	log.Fatalf("Unable to deregister the callbackURL [%s], error: %s", callbackURL, err)
}

Contributing

See the contribution guidelines.

Development

Marathon Fake

go-marathon employs a fake Marathon implementation for testing purposes. It maintains a YML-encoded list of HTTP response messages which are returned upon a successful match based upon a number of attributes, the so-called message identifier:

Response Content

The response content can be provided in one of two forms:

An example for a trivial static response content is

- uri: /v2/apps
  method: POST
  content: |
		{
		"app": {
		}
		}

which would be returned for every POST request targetting /v2/apps.

An indexed response content would look like:

- uri: /v2/apps
  method: POST
  contentSequence:
		- index: 1
		- content: |
			{
			"app": {
				"id": "foo"
			}
			}
		- index: 3
		- content: |
			{
			"app": {
				"id": "bar"
			}
			}

What this means is that the first POST request to /v2/apps would yield a 404, the second one the foo app, the third one 404 again, the fourth one bar, and every following request thereafter a 404 again. Indexed responses enable more flexible testing required by some use cases.

Trying to define both a static and indexed response content constitutes an error and leads to panic.

Scope

By default, all responses are defined globally: Every message can be queried by any request across all tests. This enables reusability and allows to keep the YML definition fairly short. For certain cases, however, it is desirable to define a set of responses that are delivered exclusively for a particular test. Scopes offer a means to do so by representing a concept similar to namespaces. Combined with indexed responses, they allow to return different responses for message identifiers already defined at the global level.

Scopes do not have a particular format -- they are just strings. A scope must be defined in two places: The message specification and the server configuration. They are pure strings without any particular structure. Given the messages specification

- uri: /v2/apps
  method: GET
	# Note: no scope defined.
  content: |
		{
		"app": {
			"id": "foo"
		}
		}
- uri: /v2/apps
  method: GET
	scope: v1.1.1  # This one does have a scope.
  contentSequence:
		- index: 1
		- content: |
			{
			"app": {
				"id": "bar"
			}
			}

and the tests

func TestFoo(t * testing.T) {
	endpoint := newFakeMarathonEndpoint(t, nil)  // No custom configs given.
	defer endpoint.Close()
	app, err := endpoint.Client.Applications(nil)
	// Do something with "foo"
}

func TestFoo(t * testing.T) {
	endpoint := newFakeMarathonEndpoint(t, &configContainer{
		server: &serverConfig{
			scope: "v1.1.1",		// Matches the message spec's scope.
		},
	})
	defer endpoint.Close()
	app, err := endpoint.Client.Applications(nil)
	// Do something with "bar"
}

The "foo" response can be used by all tests using the default fake endpoint (such as TestFoo), while the "bar" response is only visible by tests that explicitly set the scope to 1.1.1 (as TestBar does) and query the endpoint twice.