feat: add initial prometheus metrics support

Prometheus environments want metrics to be in specific formats. So the
metrics are slightly different from the monitoring endpoint.

Some metrics were not transfered over from the monitoring endpoint because in
a Prometheus environment one is likely to have other methods
(node_exporter, blackbox_exporter) to monitor some aspects.

Author: ykuksenko

api/tacticalrmm/core/decorators.py

@@ -29,3 +29,24 @@ def wrap(request, *args, **kwargs):
     wrap.__doc__ = function.__doc__
     wrap.__name__ = function.__name__
     return wrap
+
+def metrics_view(function):
+    def wrap(request, *args, **kwargs):
+        if request.method != "GET":
+            return HttpResponse("Invalid request type\n", status=400)
+
+        if "Authorization" not in request.headers:
+            return HttpResponse("Missing 'Authorization' header\n", status=400)
+
+        token = getattr(settings, "MON_TOKEN", "")
+        if not token:
+            return HttpResponse("Missing token config\n", status=401)
+
+        if request.headers["Authorization"] != 'Bearer ' + token:
+            return HttpResponse("Not authenticated\n", status=401)
+
+        return function(request, *args, **kwargs)
+
+    wrap.__doc__ = function.__doc__
+    wrap.__name__ = function.__name__
+    return wrap

api/tacticalrmm/core/views.py

@@ -5,7 +5,7 @@
 import pytz
 from cryptography import x509
 from django.conf import settings
-from django.http import JsonResponse
+from django.http import JsonResponse, HttpResponse
 from django.shortcuts import get_object_or_404
 from django.utils import timezone as djangotime
 from django.views.decorators.csrf import csrf_exempt
@@ -15,7 +15,7 @@
 from rest_framework.response import Response
 from rest_framework.views import APIView
 
-from core.decorators import monitoring_view
+from core.decorators import monitoring_view, metrics_view
 from core.utils import get_core_settings, sysd_svc_is_running, token_is_valid
 from logs.models import AuditLog
 from tacticalrmm.constants import AuditActionType, PAStatus
@@ -449,3 +449,45 @@ def status(request):
             "nginx": sysd_svc_is_running("nginx.service"),
         }
     return JsonResponse(ret, json_dumps_params={"indent": 2})
+
+@csrf_exempt
+@metrics_view
+def metrics(request):
+    from agents.models import Agent
+    from clients.models import Client, Site
+    cert_file, _ = get_certs()
+    cert_bytes = Path(cert_file).read_bytes()
+    cert = x509.load_pem_x509_certificate(cert_bytes)
+    cert_expiration = cert.not_valid_after.timestamp()
+    metrics = [
+        'trmm_buildinfo{{version="{}"}} 1'.format(settings.TRMM_VERSION),
+        'trmm_meshinfo{{version="{}"}} 1'.format(settings.MESH_VER),
+        'trmm_natsinfo{{version="{}"}} 1'.format(settings.NATS_SERVER_VER),
+        'trmm_appinfo{{version="{}"}} 1'.format(settings.APP_VER),
+        'trmm_agent{{latest_version="{}"}} 1'.format(settings.LATEST_AGENT_VER),
+        "trmm_agent_count {}".format(Agent.objects.count()),
+        "trmm_client_count {}".format(Client.objects.count()),
+        "trmm_site_count {}".format(Site.objects.count()),
+        "trmm_cert_expiration_time {}".format(cert_expiration),
+    ]
+
+    if settings.DOCKER_BUILD:
+        metrics.append("trmm_docker_build 1")
+    else:
+        metrics.append("trmm_docker_build 0")
+        services = {
+            "django": "rmm.service",
+            "mesh": "meshcentral.service",
+            "daphne": "daphne.service",
+            "celery": "celery.service",
+            "celerybeat": "celerybeat.service",
+            "redis": "redis-server.service",
+            "postgres": "postgresql.service",
+            "mongo": "mongod.service",
+            "nats": "nats.service",
+            "nats-api": "nats-api.service",
+            "nginx": "nginx.service",
+        }
+        for k,v in services.items():
+            metrics.append('trmm_service{{name="{}",unit="{}"}} {}'.format(k,v,int(sysd_svc_is_running(v))))
+    return HttpResponse(("\n").join(metrics),content_type="text/plain")

api/tacticalrmm/tacticalrmm/urls.py

@@ -5,6 +5,7 @@
 from accounts.views import CheckCreds, LoginView
 from agents.consumers import SendCMD
 from core.consumers import DashInfo
+from core import views as core_views
 
 
 class AgentIDConverter:
@@ -32,6 +33,7 @@ def to_url(self, value):
     path("winupdate/", include("winupdate.urls")),
     path("software/", include("software.urls")),
     path("core/", include("core.urls")),
+    path("metrics", core_views.metrics),
     path("automation/", include("automation.urls")),
     path("tasks/", include("autotasks.urls")),
     path("logs/", include("logs.urls")),