Autogenerate rules and fix dashboard

d1fd346f · Jan Provaznik · fbe1daac · d1fd346f · d1fd346f · d1fd346f
Commit d1fd346f authored 1 year ago by Jan Provaznik
--- a/metrics-catalog/services/ai-gateway.jsonnet
+++ b/metrics-catalog/services/ai-gateway.jsonnet
@@ -40,7 +40,7 @@ metricsCatalog.serviceDefinition(

        apdex: histogramApdex(
          histogram='http_request_duration_seconds_bucket',
-          selector=baseSelector { status: { noneOf: ['4xx', '5xx'] } },
+          selector=baseSelector { status: { noneOf: ['4xx', '5xx'] }, handler: { noneOf: ['/v2/code/completions', '/v2/completions', '/v2/code/generations'] } },
          satisfiedThreshold=5,
          toleratedThreshold=10,
          metricsFormat='migrating'
@@ -48,13 +48,13 @@ metricsCatalog.serviceDefinition(

        requestRate: rateMetric(
          counter='http_request_duration_seconds_count',
-          selector=baseSelector,
+          selector=baseSelector { handler: { noneOf: ['/v2/code/completions', '/v2/completions', '/v2/code/generations'] } },
          useRecordingRuleRegistry=false,
        ),

        errorRate: rateMetric(
          counter='http_request_duration_seconds_count',
-          selector=baseSelector { status: '5xx' },
+          selector=baseSelector { status: '5xx', handler: { noneOf: ['/v2/code/completions', '/v2/completions', '/v2/code/generations'] } },
          useRecordingRuleRegistry=false,
        ),

@@ -69,7 +69,7 @@ metricsCatalog.serviceDefinition(
          ),
        ],
      },
-      server_completions: {
+      server_code_completions: {
        severity: 's2',
        userImpacting: true,
        serviceAggregation: false,
@@ -81,7 +81,7 @@ metricsCatalog.serviceDefinition(

        apdex: histogramApdex(
          histogram='http_request_duration_seconds_bucket',
-          selector=baseSelector { status: { noneOf: ['4xx', '5xx'] }, handler: '/v2/code/completions' },
+          selector=baseSelector { status: { noneOf: ['4xx', '5xx'] }, handler: { oneOf: ['/v2/code/completions', '/v2/completions'] } },
          satisfiedThreshold=1,
          toleratedThreshold=10,
          metricsFormat='migrating'
@@ -89,13 +89,13 @@ metricsCatalog.serviceDefinition(

        requestRate: rateMetric(
          counter='http_request_duration_seconds_count',
-          selector=baseSelector { handler: '/v2/code/completions' },
+          selector=baseSelector { handler: { oneOf: ['/v2/code/completions', '/v2/completions'] } },
          useRecordingRuleRegistry=false,
        ),

        errorRate: rateMetric(
          counter='http_request_duration_seconds_count',
-          selector=baseSelector { status: '5xx', handler: '/v2/code/completions' },
+          selector=baseSelector { status: '5xx', handler: { oneOf: ['/v2/code/completions', '/v2/completions'] } },
          useRecordingRuleRegistry=false,
        ),

@@ -110,7 +110,7 @@ metricsCatalog.serviceDefinition(
          ),
        ],
      },
-      server_generations: {
+      server_code_generations: {
        severity: 's2',
        userImpacting: true,
        serviceAggregation: false,

--- a/mimir-rules/gitlab-gprd/gprd/autogenerated-service-slos.yml
+++ b/mimir-rules/gitlab-gprd/gprd/autogenerated-service-slos.yml
@@ -94,6 +94,30 @@ groups:
      tier: sv
      type: ai-gateway
    expr: "0.020000"
+  - record: slo:min:events:gitlab_service_apdex:ratio
+    labels:
+      component: server_code_completions
+      tier: sv
+      type: ai-gateway
+    expr: "0.980000"
+  - record: slo:max:events:gitlab_service_errors:ratio
+    labels:
+      component: server_code_completions
+      tier: sv
+      type: ai-gateway
+    expr: "0.020000"
+  - record: slo:min:events:gitlab_service_apdex:ratio
+    labels:
+      component: server_code_generations
+      tier: sv
+      type: ai-gateway
+    expr: "0.980000"
+  - record: slo:max:events:gitlab_service_errors:ratio
+    labels:
+      component: server_code_generations
+      tier: sv
+      type: ai-gateway
+    expr: "0.020000"
  - record: slo:min:events:gitlab_service_apdex:ratio
    labels:
      component: waf

--- a/mimir-rules/gitlab-gstg/gstg/autogenerated-service-slos.yml
+++ b/mimir-rules/gitlab-gstg/gstg/autogenerated-service-slos.yml
@@ -94,6 +94,30 @@ groups:
      tier: sv
      type: ai-gateway
    expr: "0.020000"
+  - record: slo:min:events:gitlab_service_apdex:ratio
+    labels:
+      component: server_code_completions
+      tier: sv
+      type: ai-gateway
+    expr: "0.980000"
+  - record: slo:max:events:gitlab_service_errors:ratio
+    labels:
+      component: server_code_completions
+      tier: sv
+      type: ai-gateway
+    expr: "0.020000"
+  - record: slo:min:events:gitlab_service_apdex:ratio
+    labels:
+      component: server_code_generations
+      tier: sv
+      type: ai-gateway
+    expr: "0.980000"
+  - record: slo:max:events:gitlab_service_errors:ratio
+    labels:
+      component: server_code_generations
+      tier: sv
+      type: ai-gateway
+    expr: "0.020000"
  - record: slo:min:events:gitlab_service_apdex:ratio
    labels:
      component: waf

--- a/mimir-rules/gitlab-ops/ops/autogenerated-service-slos.yml
+++ b/mimir-rules/gitlab-ops/ops/autogenerated-service-slos.yml
@@ -94,6 +94,30 @@ groups:
      tier: sv
      type: ai-gateway
    expr: "0.020000"
+  - record: slo:min:events:gitlab_service_apdex:ratio
+    labels:
+      component: server_code_completions
+      tier: sv
+      type: ai-gateway
+    expr: "0.980000"
+  - record: slo:max:events:gitlab_service_errors:ratio
+    labels:
+      component: server_code_completions
+      tier: sv
+      type: ai-gateway
+    expr: "0.020000"
+  - record: slo:min:events:gitlab_service_apdex:ratio
+    labels:
+      component: server_code_generations
+      tier: sv
+      type: ai-gateway
+    expr: "0.980000"
+  - record: slo:max:events:gitlab_service_errors:ratio
+    labels:
+      component: server_code_generations
+      tier: sv
+      type: ai-gateway
+    expr: "0.020000"
  - record: slo:min:events:gitlab_service_apdex:ratio
    labels:
      component: waf

--- a/mimir-rules/gitlab-pre/pre/autogenerated-service-slos.yml
+++ b/mimir-rules/gitlab-pre/pre/autogenerated-service-slos.yml
@@ -94,6 +94,30 @@ groups:
      tier: sv
      type: ai-gateway
    expr: "0.020000"
+  - record: slo:min:events:gitlab_service_apdex:ratio
+    labels:
+      component: server_code_completions
+      tier: sv
+      type: ai-gateway
+    expr: "0.980000"
+  - record: slo:max:events:gitlab_service_errors:ratio
+    labels:
+      component: server_code_completions
+      tier: sv
+      type: ai-gateway
+    expr: "0.020000"
+  - record: slo:min:events:gitlab_service_apdex:ratio
+    labels:
+      component: server_code_generations
+      tier: sv
+      type: ai-gateway
+    expr: "0.980000"
+  - record: slo:max:events:gitlab_service_errors:ratio
+    labels:
+      component: server_code_generations
+      tier: sv
+      type: ai-gateway
+    expr: "0.020000"
  - record: slo:min:events:gitlab_service_apdex:ratio
    labels:
      component: waf

--- a/thanos-rules/autogenerated-feature-category-metrics-ai-gateway.yml
+++ b/thanos-rules/autogenerated-feature-category-metrics-ai-gateway.yml
--- a/thanos-rules/autogenerated-key-metrics-ai-gateway.yml
+++ b/thanos-rules/autogenerated-key-metrics-ai-gateway.yml
--- a/thanos-rules/autogenerated-service-level-alerts-ai-gateway-gprd.yml
+++ b/thanos-rules/autogenerated-service-level-alerts-ai-gateway-gprd.yml
--- a/thanos-rules/autogenerated-service-level-alerts-ai-gateway-ops.yml
+++ b/thanos-rules/autogenerated-service-level-alerts-ai-gateway-ops.yml
--- a/thanos-rules/autogenerated-service-level-alerts-ai-gateway-other.yml
+++ b/thanos-rules/autogenerated-service-level-alerts-ai-gateway-other.yml
--- a/thanos-rules/autogenerated-service-slos.yml
+++ b/thanos-rules/autogenerated-service-slos.yml
@@ -95,6 +95,30 @@ groups:
      tier: sv
      type: ai-gateway
    expr: "0.020000"
+  - record: slo:min:events:gitlab_service_apdex:ratio
+    labels:
+      component: server_code_completions
+      tier: sv
+      type: ai-gateway
+    expr: "0.980000"
+  - record: slo:max:events:gitlab_service_errors:ratio
+    labels:
+      component: server_code_completions
+      tier: sv
+      type: ai-gateway
+    expr: "0.020000"
+  - record: slo:min:events:gitlab_service_apdex:ratio
+    labels:
+      component: server_code_generations
+      tier: sv
+      type: ai-gateway
+    expr: "0.980000"
+  - record: slo:max:events:gitlab_service_errors:ratio
+    labels:
+      component: server_code_generations
+      tier: sv
+      type: ai-gateway
+    expr: "0.020000"
  - record: slo:min:events:gitlab_service_apdex:ratio
    labels:
      component: waf