fix(plugin): fix ai latency metric doesn't correctly calculated issue (#14639)

KAG-7113

Author: ProBrian

kong/plugins/prometheus/exporter.lua

@@ -401,15 +401,15 @@ local function log(message, serialized)
         metrics.ai_llm_cost:inc(ai_metrics.usage.cost, labels_table_ai_llm_status)
       end
 
-      if ai_metrics.meta and ai_metrics.meta.llm_latency and ai_metrics.meta.llm_latency > 0 then
+      if ai_metrics.meta and ai_metrics.meta.llm_latency and ai_metrics.meta.llm_latency >= 0 then
         metrics.ai_llm_provider_latency:observe(ai_metrics.meta.llm_latency, labels_table_ai_llm_status)
       end
 
-      if ai_metrics.cache and ai_metrics.cache.fetch_latency and ai_metrics.cache.fetch_latency > 0 then
+      if ai_metrics.cache and ai_metrics.cache.fetch_latency and ai_metrics.cache.fetch_latency >= 0 then
         metrics.ai_cache_fetch_latency:observe(ai_metrics.cache.fetch_latency, labels_table_ai_llm_status)
       end
 
-      if ai_metrics.cache and ai_metrics.cache.embeddings_latency and ai_metrics.cache.embeddings_latency > 0 then
+      if ai_metrics.cache and ai_metrics.cache.embeddings_latency and ai_metrics.cache.embeddings_latency >= 0 then
         metrics.ai_cache_embeddings_latency:observe(ai_metrics.cache.embeddings_latency, labels_table_ai_llm_status)
       end
 

spec/03-plugins/26-prometheus/02-access_spec.lua

@@ -1,6 +1,8 @@
 local helpers = require "spec.helpers"
 local shell = require "resty.shell"
 local pl_file = require "pl.file"
+local timeout = 10
+local step = 1
 
 local UUID_PATTERN = "%x%x%x%x%x%x%x%x%-%x%x%x%x%-%x%x%x%x%-%x%x%x%x%-%x%x%x%x%x%x%x%x%x%x%x%x"
 
@@ -820,14 +822,16 @@ describe("Plugin: prometheus (access) AI metrics", function()
     assert.res_status(200, res)
 
     local body
+    -- wait until the histogram observe finished and get the correct metrics.
     helpers.wait_until(function()
       local res = assert(admin_client:send {
         method  = "GET",
         path    = "/metrics",
       })
       body = assert.res_status(200, res)
-      return res.status == 200
-    end)
+      return body:find('ai_llm_provider_latency_ms_bucket{ai_provider="openai",ai_model="gpt-3.5-turbo",cache_status="",vector_db="",embeddings_provider="",embeddings_model="",workspace="default",le="+Inf"} 1',
+          nil, true)
+    end, timeout, step)
 
     assert.matches('kong_nginx_metric_errors_total 0', body, nil, true)
     assert.matches('http_requests_total{service="empty_service",route="http-route",code="200",source="service",workspace="default",consumer=""} 2', body, nil, true)
@@ -864,8 +868,9 @@ describe("Plugin: prometheus (access) AI metrics", function()
         path    = "/metrics",
       })
       body = assert.res_status(200, res)
-      return res.status == 200
-    end)
+      return body:find('ai_llm_provider_latency_ms_bucket{ai_provider="openai",ai_model="gpt-3.5-turbo",cache_status="",vector_db="",embeddings_provider="",embeddings_model="",workspace="default",le="+Inf"} 2',
+          nil, true)
+    end, timeout, step)
 
     assert.matches('kong_nginx_metric_errors_total 0', body, nil, true)
     assert.matches('http_requests_total{service="empty_service",route="http-route",code="200",source="service",workspace="default",consumer=""} 3', body, nil, true)
@@ -898,8 +903,9 @@ describe("Plugin: prometheus (access) AI metrics", function()
         path    = "/metrics",
       })
       body = assert.res_status(200, res)
-      return res.status == 200
-    end)
+      return body:find('ai_llm_provider_latency_ms_bucket{ai_provider="openai",ai_model="gpt-3.5-turbo",cache_status="",vector_db="",embeddings_provider="",embeddings_model="",workspace="default",le="+Inf"} 2',
+          nil, true)
+    end, timeout, step)
 
     assert.matches('http_requests_total{service="empty_service",route="http-route",code="400",source="kong",workspace="default",consumer=""} 1', body, nil, true)
     assert.matches('kong_nginx_metric_errors_total 0', body, nil, true)