Merge pull request #36 from robusta-dev/prometheus_resolution

Prometheus resolution

docs/getting-started/uninstallation.rst

@@ -1,4 +1,4 @@
 Uninstalling Robusta
 #####################
 
-Everything Robusta installs is placed in the ``robusta`` namespace. To delete Robusta, just delete that namespace.
+Everything Robusta installs is placed in the ``robusta`` namespace. To delete Robusta, just delete that namespace.

playbooks/alerts_integration.py

@@ -97,10 +97,10 @@ def enrich(self, alert: PrometheusKubernetesAlert):
 class GraphEnricher(Enricher):
     def enrich(self, alert: PrometheusKubernetesAlert):
         url = urlparse(alert.alert.generatorURL)
-        prometheus_base_url = f"{url.scheme}://{url.netloc}"
         if alert.prometheus_url:
             prometheus_base_url = alert.prometheus_url
-
+        else:
+            prometheus_base_url = PrometheusDiscovery.find_prometheus_url()
         prom = PrometheusConnect(url=prometheus_base_url, disable_ssl=True)
 
         promql_query = re.match(

src/robusta/api/__init__.py

@@ -27,6 +27,7 @@
 from ..integrations.scheduled.triggers import *
 from ..integrations.git.git_repo import *
 from ..integrations.resource_analysis.kubernetes_node_analyzer import NodeAnalyzer
+from ..integrations.prometheus.utils import PrometheusDiscovery
 from ..core.persistency.in_memory import get_persistent_data
 from ..utils.rate_limiter import RateLimiter
 from ..runner.object_updater import *

src/robusta/cli/main.py

@@ -164,7 +164,6 @@ def examples_download(
     cluster_name: str = None,
     use_robusta_ui: bool = False,
     skip_robusta_sink: bool = False,
-    skip_new: bool = True,
     robusta_ui_token: str = None,
     url: str = None,
     skip_integrations: bool = False,
@@ -195,14 +194,12 @@ def examples_download(
             "Please specify a unique name for your cluster or press ENTER to use the default",
             default=default_name,
         )
-    # skip_new is used here, temporary, since we don't have the new fields in the released active_playbooks.yaml yet
-    # TODO remove on next release
-    if not skip_new and cluster_name is not None:
+    if cluster_name is not None:
         replace_in_file(
             "playbooks/active_playbooks.yaml", "<CLUSTER_NAME>", cluster_name.strip()
         )
 
-    if not skip_new and (
+    if not skip_robusta_sink and (
         use_robusta_ui or typer.confirm("Would you like to use Robusta UI?")
     ):
         if robusta_ui_token is None:
@@ -249,10 +246,6 @@ def examples(
         False,
         help="Enable Robusta sink?",
     ),
-    skip_new: bool = typer.Option(
-        True,
-        help="Skip new config replacements?",
-    ),
     robusta_ui_token: str = typer.Option(
         None,
         help="Robusta UI account token",
@@ -266,13 +259,13 @@ def examples(
         help="Skip integrations configuration",
     ),
 ):
+    """Download playbooks code and configuration defaults"""
     examples_download(
         slack_api_key,
         slack_channel,
         cluster_name,
         use_robusta_ui,
         skip_robusta_sink,
-        skip_new,
         robusta_ui_token,
         url,
         skip_integrations,

src/robusta/integrations/prometheus/utils.py

@@ -1,8 +1,22 @@
+import logging
+
 from ...utils.service_discovery import find_service_url
 
 
-def find_prometheus_url():
-    """
-    Try to autodiscover the url of an in-cluster grafana service
-    """
-    return find_service_url("app=kube-prometheus-stack-prometheus")
+class PrometheusDiscovery:
+    prometheus_url = None
+    @classmethod
+    def find_prometheus_url(cls):
+        """
+        Try to autodiscover the url of an in-cluster grafana service
+        """
+        if cls.prometheus_url:
+            return cls.prometheus_url
+        prometheus_selectors = ["app=kube-prometheus-stack-prometheus", "app.kubernetes.io/name=prometheus"]
+        for label_selector in prometheus_selectors:
+            service_url = find_service_url(label_selector)
+            if service_url:
+                cls.prometheus_url = service_url
+                return service_url
+        logging.error("Prometheus url could not be found. Add 'prometheus_url' under global_config")
+        return None

src/robusta/integrations/resource_analysis/kubernetes_node_analyzer.py

@@ -1,7 +1,7 @@
 from collections import OrderedDict
 from hikaru.model import Node
 from prometheus_api_client import PrometheusConnect
-from ..prometheus.utils import find_prometheus_url
+from ..prometheus.utils import PrometheusDiscovery
 
 
 class NodeAnalyzer:
@@ -17,7 +17,8 @@ def __init__(self, node: Node, prometheus_url: str, range_size="5m"):
             if addr.type == "InternalIP"
         )
         if prometheus_url is None:
-            prometheus_url = find_prometheus_url()
+            prometheus_url = PrometheusDiscovery.find_prometheus_url()
+
         self.prom = PrometheusConnect(url=prometheus_url, disable_ssl=True)
 
     def get_total_cpu_usage(self, other_method=False):

src/robusta/utils/service_discovery.py

@@ -1,5 +1,6 @@
 import logging
 from kubernetes import client
+from kubernetes.client import V1ServiceList
 from kubernetes.client.models.v1_service import V1Service
 
 
@@ -9,14 +10,17 @@ def find_service_url(label_selector):
     """
     # we do it this way because there is a weird issue with hikaru's ServiceList.listServiceForAllNamespaces()
     v1 = client.CoreV1Api()
-    svc: V1Service = v1.list_service_for_all_namespaces(
+    svc_list: V1ServiceList = v1.list_service_for_all_namespaces(
         label_selector=label_selector
-    ).items[0]
+    )
+    if not svc_list.items:
+        return None
+    svc: V1Service = svc_list.items[0]
     name = svc.metadata.name
     namespace = svc.metadata.namespace
     port = svc.spec.ports[0].port
-    url = f"http://{name}.{namespace}.svc:{port}"
-    logging.debug(
+    url = f"http://{name}.{namespace}.svc.cluster.local:{port}"
+    logging.info(
         f"discovered service with label-selector: `{label_selector}` at url: `{url}`"
     )
     return url

tests/utils/robusta_utils.py

@@ -47,7 +47,6 @@ def cli_examples(self, playbooks_url: str, slack_channel: str, slack_api_key: st
                 "--robusta-ui-token",
                 "test-token",
                 "--skip-robusta-sink",
-                "--skip-new",
             ],
         )
         assert b"examples downloaded into the playbooks/ directory" in logs