[marathon] Count running/pending deployments

checks.d/marathon.py

@@ -43,6 +43,7 @@ def check(self, instance):
             auth = (user,password)
         else:
             auth = None
+
         instance_tags = instance.get('tags', [])
         default_timeout = self.init_config.get('default_timeout', self.DEFAULT_TIMEOUT)
         timeout = float(instance.get('timeout', default_timeout))
@@ -56,6 +57,11 @@ def check(self, instance):
                     if attr in app:
                         self.gauge('marathon.' + attr, app[attr], tags=tags)
 
+        if instance.get('enable_deployment_metrics', False):
+            response = self.get_json(urljoin(url, "v2/deployments"), timeout, auth)
+            if response is not None:
+                self.gauge('marathon.deployments', len(response), tags=instance_tags)
+
     def get_json(self, url, timeout, auth):
         try:
             r = requests.get(url, timeout=timeout, auth=auth)

conf.d/marathon.yaml.example

@@ -9,3 +9,4 @@ instances:
   #   if marathon is protected by basic auth
   #   user: "username"
   #   password: "password"
+  #   enable_deployment_metrics: #Optional. Set to true if you want to receive send deployment related metrics. Adds an extra API call to Marathon.

tests/checks/common.py

@@ -14,6 +14,7 @@
 import time
 import traceback
 import unittest
+import json
 
 # project
 from checks import AgentCheck
@@ -130,6 +131,9 @@ def read_file(file_name, string_escape=True):
                 contents = contents.decode('string-escape')
             return contents.decode("utf-8")
 
+    @staticmethod
+    def read_json_file(file_name, string_escape=True):
+        return json.loads(Fixtures.read_file(file_name, string_escape=string_escape))
 
 class AgentCheckTest(unittest.TestCase):
     DEFAULT_AGENT_CONFIG = {

tests/checks/fixtures/marathon/apps.json

@@ -0,0 +1,88 @@
+{
+  "apps": [
+    {
+      "id": "/my-app",
+      "cmd": "/usr/bin/run-app",
+      "args": null,
+      "user": null,
+      "env": {
+        "ENVIRONMENT": "development"
+      },
+      "instances": 10,
+      "cpus": 1.0,
+      "mem": 512,
+      "disk": 0,
+      "executor": "",
+      "constraints": [],
+      "ports": [
+        10002
+      ],
+      "portDefinitions": [
+        {
+          "port": 10002,
+          "protocol": "tcp",
+          "labels": {}
+        }
+      ],
+      "requirePorts": false,
+      "backoffSeconds": 1,
+      "backoffFactor": 1.15,
+      "maxLaunchDelaySeconds": 3600,
+      "container": {
+        "type": "DOCKER",
+        "volumes": [],
+        "docker": {
+          "image": "my-org/my-image:latest",
+          "network": "BRIDGE",
+          "portMappings": [
+            {
+              "containerPort": 80,
+              "hostPort": 0,
+              "servicePort": 10002,
+              "protocol": "tcp",
+              "labels": {}
+            }
+          ],
+          "privileged": false,
+          "parameters": [],
+          "forcePullImage": true
+        }
+      },
+      "healthChecks": [
+        {
+          "path": "/health",
+          "protocol": "HTTP",
+          "portIndex": 0,
+          "gracePeriodSeconds": 5,
+          "intervalSeconds": 10,
+          "timeoutSeconds": 20,
+          "maxConsecutiveFailures": 3,
+          "ignoreHttp1xx": false
+        }
+      ],
+      "readinessChecks": [],
+      "dependencies": [],
+      "upgradeStrategy": {
+        "minimumHealthCapacity": 0.5,
+        "maximumOverCapacity": 0.5
+      },
+      "labels": {
+        "HAPROXY_0_VHOST": "my-app.my-org.net",
+        "HAPROXY_GROUP": "my-app"
+      },
+      "acceptedResourceRoles": null,
+      "ipAddress": null,
+      "version": "2016-08-25T18:13:34.079Z",
+      "residency": null,
+      "versionInfo": {
+        "lastScalingAt": "2016-08-25T18:13:34.079Z",
+        "lastConfigChangeAt": "2016-08-25T16:58:48.416Z"
+      },
+      "tasksStaged": 5,
+      "tasksRunning": 10,
+      "tasksHealthy": 5,
+      "tasksUnhealthy": 0,
+      "deployments": []
+    }
+  ]
+}

tests/checks/fixtures/marathon/deployments.json

@@ -0,0 +1,87 @@
+[
+  {
+    "affectedApps": [
+      "/test/service/srv1",
+      "/test/db/mongo1",
+      "/test/frontend/app1"
+    ],
+    "currentStep": 2,
+    "currentActions": [
+      {
+        "action": "RestartApplication",
+        "app": "/test/frontend/app1",
+        "readinessChecks": [
+          {
+            "lastResponse": {
+              "body": "{}",
+              "contentType": "application/json",
+              "status": 500
+            },
+            "name": "myReadyCheck",
+            "ready": false,
+            "taskId": "test_frontend_app1.c9de6033"
+          }
+        ]
+      }
+    ],
+    "totalSteps": 9,
+    "id": "2e72dbf1-2b2a-4204-b628-e8bd160945dd",
+    "steps": [
+      [
+        {
+          "action": "RestartApplication",
+          "app": "/test/service/srv1"
+        }
+      ],
+      [
+        {
+          "action": "RestartApplication",
+          "app": "/test/db/mongo1"
+        }
+      ],
+      [
+        {
+          "action": "RestartApplication",
+          "app": "/test/frontend/app1"
+        }
+      ],
+      [
+        {
+          "action": "KillAllOldTasksOf",
+          "app": "/test/frontend/app1"
+        }
+      ],
+      [
+        {
+          "action": "KillAllOldTasksOf",
+          "app": "/test/db/mongo1"
+        }
+      ],
+      [
+        {
+          "action": "KillAllOldTasksOf",
+          "app": "/test/service/srv1"
+        }
+      ],
+      [
+        {
+          "action": "ScaleApplication",
+          "app": "/test/service/srv1"
+        }
+      ],
+      [
+        {
+          "action": "ScaleApplication",
+          "app": "/test/db/mongo1"
+        }
+      ],
+      [
+        {
+          "action": "ScaleApplication",
+          "app": "/test/frontend/app1"
+        }
+      ]
+    ],
+    "version": "2014-07-09T11:14:11.477Z"
+  }
+]

tests/checks/mock/test_marathon.py

@@ -0,0 +1,68 @@
+# project
+from tests.checks.common import AgentCheckTest
+from tests.checks.common import Fixtures
+
+DEPLOYMENT_METRICS_CONFIG = {
+    'init_config': {
+        'default_timeout': 5
+    },
+    'instances': [
+        {
+            'url': 'http://localhost:8080',
+            'enable_deployment_metrics': True
+        }
+    ]
+}
+
+DEFAULT_CONFIG = {
+    'init_config': {
+        'default_timeout': 5
+    },
+    'instances': [
+        {
+            'url': 'http://localhost:8080'
+        }
+    ]
+}
+
+def getMetricNames(metrics):
+    return [metric[0] for metric in metrics]
+
+class MarathonCheckTest(AgentCheckTest):
+    CHECK_NAME = 'marathon'
+
+    def test_default_configuration(self):
+        def side_effect(url, timeout, auth):
+            if "v2/apps" in url:
+                return Fixtures.read_json_file("apps.json")
+            else:
+                raise Exception("unknown url:" + url)
+
+        self.run_check(DEFAULT_CONFIG, mocks={"get_json": side_effect})
+        self.assertMetric('marathon.apps', value=1)
+
+        # deployment-related metrics aren't included by default.
+        self.assertTrue('marathon.deployments' not in getMetricNames(self.metrics))
+
+    def test_empty_responses(self):
+        def side_effect(url, timeout, auth):
+            if "v2/apps" in url:
+                return {"apps": []}
+            else:
+                raise Exception("unknown url:" + url)
+
+        self.run_check(DEFAULT_CONFIG, mocks={"get_json": side_effect})
+        self.assertMetric('marathon.apps', value=0)
+
+    def test_enabled_deployment_metrics(self):
+        def side_effect(url, timeout, auth):
+            if "v2/apps" in url:
+                return Fixtures.read_json_file("apps.json")
+            elif "v2/deployments" in url:
+                return Fixtures.read_json_file("deployments.json")
+            else:
+                raise Exception("unknown url:" + url)
+
+        self.run_check(DEPLOYMENT_METRICS_CONFIG, mocks={"get_json": side_effect})
+        self.assertMetric('marathon.apps', value=1)
+        self.assertMetric('marathon.deployments', value=1)