Merge pull request #68 from appuio/fix/ingress-sli

Fix ingress error ratio query to only consider the canary route for the cluster's `appsDomain`
appuio · May 15, 2024 · 2f5e031 · 2f5e031
2 parents e03df70 + aaf49a3
commit 2f5e031
Show file tree

Hide file tree

Showing 2 changed files with 15 additions and 11 deletions.
diff --git a/component/slos.libsonnet b/component/slos.libsonnet
@@ -93,8 +93,10 @@ local defaultSlos = {
   },
   ingress: {
     local config = params.slos.ingress,
-    local os = com.getValueOrDefault(inv.parameters, 'openshift', {}),
-    local appsDomain = com.getValueOrDefault(os, 'appsDomain', ''),
+    local os = std.get(inv.parameters, 'openshift', {}),
+    // NOTE: appsDomain should always be present if we have parameter `openshift`.
+    local appsDomain = std.get(os, 'appsDomain', ''),
+    local canaryRoute = 'canary-openshift-ingress-canary.%s' % appsDomain,
 
     extra_rules: [
       {
@@ -111,14 +113,16 @@ local defaultSlos = {
           description: 'OpenShift ingress SLO based on canary availability',
           sli: {
             raw: {
-              error_ratio_query: '1 - avg_over_time(appuio_ocp4_slo:ingress_canary_route_reachable:no_instance[{{.window}}])',
+              error_ratio_query:
+                '1 - avg_over_time(appuio_ocp4_slo:ingress_canary_route_reachable:no_instance{%s}[{{.window}}])'
+                % [ if appsDomain != '' then 'host="%s"' % canaryRoute else '' ],
             },
           },
           alerting: {
             name: 'SLO_ClusterIngressFailure',
             annotations: {
               summary: 'Probes to ingress canary fail',
-              [if appsDomain != '' then 'canary_url']: 'canary-openshift-ingress-canary.%s' % appsDomain,
+              [if appsDomain != '' then 'canary_url']: canaryRoute,
             },
             page_alert: {},
             ticket_alert: {},

diff --git a/tests/golden/defaults/openshift4-slos/openshift4-slos/ingress.yaml b/tests/golden/defaults/openshift4-slos/openshift4-slos/ingress.yaml
@@ -10,49 +10,49 @@ spec:
   groups:
     - name: sloth-slo-sli-recordings-ingress-canary
       rules:
-        - expr: (1 - avg_over_time(appuio_ocp4_slo:ingress_canary_route_reachable:no_instance[5m]))
+        - expr: (1 - avg_over_time(appuio_ocp4_slo:ingress_canary_route_reachable:no_instance{host="canary-openshift-ingress-canary.apps.foo.example.com"}[5m]))
           labels:
             sloth_id: ingress-canary
             sloth_service: ingress
             sloth_slo: canary
             sloth_window: 5m
           record: slo:sli_error:ratio_rate5m
-        - expr: (1 - avg_over_time(appuio_ocp4_slo:ingress_canary_route_reachable:no_instance[30m]))
+        - expr: (1 - avg_over_time(appuio_ocp4_slo:ingress_canary_route_reachable:no_instance{host="canary-openshift-ingress-canary.apps.foo.example.com"}[30m]))
           labels:
             sloth_id: ingress-canary
             sloth_service: ingress
             sloth_slo: canary
             sloth_window: 30m
           record: slo:sli_error:ratio_rate30m
-        - expr: (1 - avg_over_time(appuio_ocp4_slo:ingress_canary_route_reachable:no_instance[1h]))
+        - expr: (1 - avg_over_time(appuio_ocp4_slo:ingress_canary_route_reachable:no_instance{host="canary-openshift-ingress-canary.apps.foo.example.com"}[1h]))
           labels:
             sloth_id: ingress-canary
             sloth_service: ingress
             sloth_slo: canary
             sloth_window: 1h
           record: slo:sli_error:ratio_rate1h
-        - expr: (1 - avg_over_time(appuio_ocp4_slo:ingress_canary_route_reachable:no_instance[2h]))
+        - expr: (1 - avg_over_time(appuio_ocp4_slo:ingress_canary_route_reachable:no_instance{host="canary-openshift-ingress-canary.apps.foo.example.com"}[2h]))
           labels:
             sloth_id: ingress-canary
             sloth_service: ingress
             sloth_slo: canary
             sloth_window: 2h
           record: slo:sli_error:ratio_rate2h
-        - expr: (1 - avg_over_time(appuio_ocp4_slo:ingress_canary_route_reachable:no_instance[6h]))
+        - expr: (1 - avg_over_time(appuio_ocp4_slo:ingress_canary_route_reachable:no_instance{host="canary-openshift-ingress-canary.apps.foo.example.com"}[6h]))
           labels:
             sloth_id: ingress-canary
             sloth_service: ingress
             sloth_slo: canary
             sloth_window: 6h
           record: slo:sli_error:ratio_rate6h
-        - expr: (1 - avg_over_time(appuio_ocp4_slo:ingress_canary_route_reachable:no_instance[1d]))
+        - expr: (1 - avg_over_time(appuio_ocp4_slo:ingress_canary_route_reachable:no_instance{host="canary-openshift-ingress-canary.apps.foo.example.com"}[1d]))
           labels:
             sloth_id: ingress-canary
             sloth_service: ingress
             sloth_slo: canary
             sloth_window: 1d
           record: slo:sli_error:ratio_rate1d
-        - expr: (1 - avg_over_time(appuio_ocp4_slo:ingress_canary_route_reachable:no_instance[3d]))
+        - expr: (1 - avg_over_time(appuio_ocp4_slo:ingress_canary_route_reachable:no_instance{host="canary-openshift-ingress-canary.apps.foo.example.com"}[3d]))
           labels:
             sloth_id: ingress-canary
             sloth_service: ingress