Merge ~andreserl/maas:prometheus_publish_stats into maas:master

[Andres Rodriguez]

Publish more stats for prometheus.

[Lee Trager (ltrager)]

LGTM!

review: Approve

[Andres Rodriguez (andreserl)]

review: Approve

[MAAS Lander (maas-lander)]

UNIT TESTS
-b prometheus_publish_stats lp:~andreserl/maas/+git/maas into -b master lp:~maas-committers/maas

STATUS: SUCCESS
COMMIT: 1e757599782607cdbde91424fe0448addf8b754a

review: Approve

diff --git a/src/maasserver/prometheus.py b/src/maasserver/prometheus.py
index 55c58df..fe69f4a 100644
--- a/src/maasserver/prometheus.py
+++ b/src/maasserver/prometheus.py
@@ -16,7 +16,11 @@ from django.http import (
     HttpResponseNotFound,
 )
 from maasserver.models import Config
-from maasserver.stats import get_maas_stats
+from maasserver.stats import (
+    get_kvm_pods_stats,
+    get_maas_stats,
+    get_machines_by_architecture,
+)
 from maasserver.utils.orm import transactional
 from maasserver.utils.threads import deferToDatabase
 from provisioningserver.logger import LegacyLogger
@@ -49,6 +53,8 @@ def prometheus_handler(request):
 def get_stats_for_prometheus():
     registry = CollectorRegistry()
     stats = json.loads(get_maas_stats())
+    architectures = get_machines_by_architecture()
+    pods = get_kvm_pods_stats()
 
     # Gather counter for machines per status
     counter = Gauge(
@@ -57,6 +63,48 @@ def get_stats_for_prometheus():
     for status, machines in stats['machine_status'].items():
         counter.labels(status).set(machines)
 
+    # Gather counter for number of nodes (controllers/machine/devices)
+    counter = Gauge(
+        "nodes", "Number of nodes per type (e.g. racks, machines, etc).",
+        ["type"], registry=registry)
+    for ctype, number in stats['controllers'].items():
+        counter.labels(ctype).set(number)
+    for ctype, number in stats['nodes'].items():
+        counter.labels(ctype).set(number)
+
+    # Gather counter for networks
+    counter = Gauge(
+        "networks", "General statistics for subnets.",
+        ["type"], registry=registry)
+    for stype, number in stats['network_stats'].items():
+        counter.labels(stype).set(number)
+
+    # Gather overall amount of machine resources
+    counter = Gauge(
+        "machine_resources", "Amount of combined resources for all machines",
+        ["resource"], registry=registry)
+    for resource, value in stats['machine_stats'].items():
+        counter.labels(resource).set(value)
+
+    # Gather all stats for pods
+    counter = Gauge(
+        "kvm_pods", "General stats for KVM pods",
+        ["type"], registry=registry)
+    for resource, value in pods.items():
+        if isinstance(value, dict):
+            for r, v in value.items():
+                counter.labels("%s_%s" % (resource, r)).set(v)
+        else:
+            counter.labels(resource).set(value)
+
+    # Gather statistics for architectures
+    if len(architectures.keys()) > 0:
+        counter = Gauge(
+            "machine_arches", "Number of machines per architecture.",
+            ["arches"], registry=registry)
+        for arch, machines in architectures.items():
+            counter.labels(arch).set(machines)
+
     return registry
 
 
diff --git a/src/maasserver/tests/test_prometheus.py b/src/maasserver/tests/test_prometheus.py
index fc93872..8637ede 100644
--- a/src/maasserver/tests/test_prometheus.py
+++ b/src/maasserver/tests/test_prometheus.py
@@ -68,16 +68,47 @@ class TestPrometheus(MAASServerTestCase):
     def test_get_stats_for_prometheus(self):
         self.patch(prometheus, "CollectorRegistry")
         self.patch(prometheus, "Gauge")
+        # general values
         values = {
             "machine_status": {
                 "random_status": 0,
             },
+            "controllers": {
+                "regions": 0,
+            },
+            "nodes": {
+                "machines": 0,
+            },
+            "network_stats": {
+                "spaces": 0,
+            },
+            "machine_stats": {
+                "total_cpus": 0,
+            },
         }
         mock = self.patch(prometheus, "get_maas_stats")
         mock.return_value = json.dumps(values)
+        # architecture
+        arches = {
+            "amd64": 0,
+            "i386": 0,
+        }
+        mock_arches = self.patch(prometheus, "get_machines_by_architecture")
+        mock_arches.return_value = arches
+        # pods
+        pods = {
+            "kvm_pods": 0,
+            "kvm_machines": 0,
+        }
+        mock_pods = self.patch(prometheus, "get_kvm_pods_stats")
+        mock_pods.return_value = pods
         get_stats_for_prometheus()
         self.assertThat(
             mock, MockCalledOnce())
+        self.assertThat(
+            mock_arches, MockCalledOnce())
+        self.assertThat(
+            mock_pods, MockCalledOnce())
 
     def test_push_stats_to_prometheus(self):
         factory.make_RegionRackController()