Refactor Cilium measurements queries

Azure · Jan 5, 2025 · 8ef35be · 8ef35be
1 parent fef9aad
commit 8ef35be
Showing 1 changed file with 40 additions and 23 deletions.
diff --git a/modules/python/clusterloader2/slo/config/modules/cilium-measurements.yaml b/modules/python/clusterloader2/slo/config/modules/cilium-measurements.yaml
@@ -153,17 +153,20 @@ steps:
           query: quantile(0.5, avg_over_time(cilium_operator_process_resident_memory_bytes[%v:]) / 1024 / 1024)
 
     # Cilium API Rate Limiter
-    - Identifier: CiliumAPIRateLimiterProcessedRequestsCount
+
+    - Identifier: CiliumAPIRateLimiterProcessedRequests
       Method: GenericPrometheusQuery
       Params:
         action: {{$action}}
-        metricName: Cilium API Rate Limiter Processed Requests Count
+        metricName: Cilium API Rate Limiter Processed Requests
         metricVersion: v1
         unit: requests/second
         enableViolations: true
         queries:
         - name: Total
-          query: sum(rate(cilium_api_limiter_processed_requests_total[%v:]))
+          query: sum(cilium_api_limiter_processed_requests_count)
+        - name: Rate
+          query: sum(rate(cilium_api_limiter_processed_requests_count[%v:]))
 
 
     - Identifier: CiliumAPIRateLimiterProcessingDuration
@@ -175,14 +178,11 @@ steps:
         unit: s
         enableViolations: true
         queries:
-        - name: Perc99
-          query: histogram_quantile(0.99, sum(rate(cilium_api_limiter_processing_duration_seconds_bucket[%v:])) by (le))
-        - name: Perc95
-          query: histogram_quantile(0.95, sum(rate(cilium_api_limiter_processing_duration_seconds_bucket[%v:])) by (le))
-        - name: Perc50
-          query: histogram_quantile(0.50, sum(rate(cilium_api_limiter_processing_duration_seconds_bucket[%v:])) by (le))
         - name: Avg
-          query: avg(rate(cilium_api_limiter_processing_duration_seconds_sum[%v:])) / avg(rate(cilium_api_limiter_processing_duration_seconds_count[%v:])) 
+          query: avg(sum_over_time(cilium_api_limiter_processing_duration_seconds[0][%v]))
+        - name: Max
+          query: max_over_time(max(cilium_api_limiter_processing_duration_seconds[0])[%v])
+
 
     - Identifier: CiliumAPIRateLimiterRequestsInFlight
       Method: GenericPrometheusQuery
@@ -196,14 +196,15 @@ steps:
         - name: Instantaneous
           query: sum(cilium_api_limiter_requests_in_flight)
         - name: MaxOverTimeSum
-          query: max_over_time(sum(cilium_api_limiter_requests_in_flight)[%v:])
+          query: max_over_time(cilium_api_limiter_requests_in_flight[0][%v:])
         - name: AvgOverTimeSum
-          query: avg_over_time(sum(cilium_api_limiter_requests_in_flight)[%v:])
+          query: avg_over_time(cilium_api_limiter_requests_in_flight[0][%v:])
         - name: SumOverTime
-          query: sum_over_time(sum(cilium_api_limiter_requests_in_flight)[%v:])
+          query: sum_over_time(cilium_api_limiter_requests_in_flight[0][%v:])
         - name: MinOverTimeSum
-          query: min_over_time(sum(cilium_api_limiter_requests_in_flight)[%v:])
-
+          query: min_over_time(cilium_api_limiter_requests_in_flight[0][%v:])
+
+
     - Identifier: CiliumAPIRateLimiterWaitDuration
       Method: GenericPrometheusQuery
       Params:
@@ -213,14 +214,30 @@ steps:
         unit: s
         enableViolations: true
         queries:
-        - name: Perc99
-          query: histogram_quantile(0.99, sum(rate(cilium_api_limiter_wait_duration_seconds_bucket[%v:])) by (le))
-        - name: Perc95
-          query: histogram_quantile(0.95, sum(rate(cilium_api_limiter_wait_duration_seconds_bucket[%v:])) by (le))
-        - name: Perc50
-          query: histogram_quantile(0.50, sum(rate(cilium_api_limiter_wait_duration_seconds_bucket[%v:])) by (le))
         - name: Avg
-          query: avg(rate(cilium_api_limiter_wait_duration_seconds_sum[%v:])) / avg(rate(cilium_api_limiter_wait_duration_seconds_count[%v:]))
+          query: avg(cilium_api_limiter_wait_duration_seconds{value="mean", api_call="endpoint-create"}[%v])
+        - name: Max
+          query: max_over_time(cilium_api_limiter_wait_duration_seconds{value="max", api_call="endpoint-create"}[%v])
+        - name: Min
+          query: min_over_time(cilium_api_limiter_wait_duration_seconds{value="min", api_call="endpoint-create"}[%v])
+
+
+    - Identifier: CiliumAPIRateLimiterRateLimit
+      Method: GenericPrometheusQuery
+      Params:
+        action: {{$action}}
+        metricName: Cilium API Rate Limiter Rate Limit
+        metricVersion: v1
+        unit: count
+        enableViolations: true
+        queries:
+        - name: Avg
+          query: avg(avg_over_time(cilium_api_limiter_rate_limit{value="limit", api_call="endpoint-create"}[%v]))
+        - name: Max
+          query: max_over_time(cilium_api_limiter_rate_limit{value="limit", api_call="endpoint-create"}[%v])
+        - name: Min
+          query: min_over_time(cilium_api_limiter_rate_limit{value="limit", api_call="endpoint-create"}[%v])
+
 
     - Identifier: CiliumAgentAPIProcessTime
       Method: GenericPrometheusQuery
@@ -238,4 +255,4 @@ steps:
         - name: Perc50
           query: histogram_quantile(0.50, sum(rate(cilium_agent_api_process_time_seconds_bucket[%v:])) by (le))
         - name: Avg
-          query: avg(rate(cilium_agent_api_process_time_seconds_sum[%v:])) / avg(rate(cilium_agent_api_process_time_seconds_count[%v:]))
+          query: avg(rate(cilium_agent_api_process_time_seconds_sum[%v:])) / avg(rate(cilium_agent_api_process_time_seconds_count[%v:]))