From 5b560d77eccd921fdb3a083353470647775a9f60 Mon Sep 17 00:00:00 2001
From: Vivek Kumar <vkumar3@salesforce.com>
Date: Thu, 22 Feb 2024 20:27:18 +0530
Subject: [PATCH] add telemetry endpoint

---
 cmd/adapter/adapter.go        | 12 -------
 pkg/client/metrics/metrics.go | 64 +++++++----------------------------
 2 files changed, 13 insertions(+), 63 deletions(-)

diff --git a/cmd/adapter/adapter.go b/cmd/adapter/adapter.go
index 55961eaf..dd2a799b 100644
--- a/cmd/adapter/adapter.go
+++ b/cmd/adapter/adapter.go
@@ -271,23 +271,11 @@ func (cmd *PrometheusAdapter) addResourceMetricsAPI(promClient prom.Client, stop
 		return err
 	}
 
-	config, err := cmd.Config()
-	if err != nil {
-		return err
-	}
-	config.GenericConfig.EnableMetrics = false
-
 	server, err := cmd.Server()
 	if err != nil {
 		return err
 	}
 
-	metricsHandler, err := mprom.MetricsHandler()
-	if err != nil {
-		return err
-	}
-	server.GenericAPIServer.Handler.NonGoRestfulMux.HandleFunc("/metrics", metricsHandler)
-
 	if err := api.Install(provider, podInformer.Lister(), informer.Core().V1().Nodes().Lister(), server.GenericAPIServer, nil); err != nil {
 		return err
 	}
diff --git a/pkg/client/metrics/metrics.go b/pkg/client/metrics/metrics.go
index 8067872a..fd7660da 100644
--- a/pkg/client/metrics/metrics.go
+++ b/pkg/client/metrics/metrics.go
@@ -18,16 +18,11 @@ package metrics
 
 import (
 	"context"
-	"net/http"
 	"net/url"
 	"time"
 
 	"github.com/prometheus/client_golang/prometheus"
 
-	apimetrics "k8s.io/apiserver/pkg/endpoints/metrics"
-	"k8s.io/component-base/metrics"
-	"k8s.io/component-base/metrics/legacyregistry"
-
 	"sigs.k8s.io/prometheus-adapter/pkg/client"
 )
 
@@ -35,47 +30,18 @@ var (
 	// queryLatency is the total latency of any query going through the
 	// various endpoints (query, range-query, series).  It includes some deserialization
 	// overhead and HTTP overhead.
-	queryLatency = metrics.NewHistogramVec(
-		&metrics.HistogramOpts{
-			Namespace: "prometheus_adapter",
-			Subsystem: "prometheus_client",
-			Name:      "request_duration_seconds",
-			Help:      "Prometheus client query latency in seconds.  Broken down by target prometheus endpoint and target server",
-			Buckets:   prometheus.DefBuckets,
+	queryLatency = prometheus.NewHistogramVec(
+		prometheus.HistogramOpts{
+			Name:    "cmgateway_prometheus_query_latency_seconds",
+			Help:    "Prometheus client query latency in seconds.  Broken down by target prometheus endpoint and target server",
+			Buckets: prometheus.ExponentialBuckets(0.0001, 2, 10),
 		},
-		[]string{"path", "server"},
+		[]string{"endpoint", "server"},
 	)
-
-	// define a counter for API errors for various ErrorTypes
-    apiErrorCount = metrics.NewCounterVec(
-    	&metrics.CounterOpts{
-    		Namespace: "prometheus_adapter",
-    		Subsystem: "prometheus_client",
-    		Name:      "api_errors_total",
-    		Help:      "Total number of API errors",
-    	},
-    	[]string{"error_code", "path", "server"},
-    )
 )
 
-func MetricsHandler() (http.HandlerFunc, error) {
-	registry := metrics.NewKubeRegistry()
-
-	errRegisterQueryLatency := registry.Register(queryLatency)
-	if errRegisterQueryLatency != nil {
-		return nil, errRegisterQueryLatency
-	}
-
-	errRegisterAPIErrorCount := registry.Register(apiErrorCount)
-    if errRegisterAPIErrorCount != nil {
-        return nil, errRegisterAPIErrorCount
-    }
-
-	apimetrics.Register()
-	return func(w http.ResponseWriter, req *http.Request) {
-		legacyregistry.Handler().ServeHTTP(w, req)
-		metrics.HandlerFor(registry, metrics.HandlerOpts{}).ServeHTTP(w, req)
-	}, nil
+func init() {
+	prometheus.MustRegister(queryLatency)
 }
 
 // instrumentedClient is a client.GenericAPIClient which instruments calls to Do,
@@ -92,16 +58,12 @@ func (c *instrumentedGenericClient) Do(ctx context.Context, verb, endpoint strin
 		endTime := time.Now()
 		// skip calls where we don't make the actual request
 		if err != nil {
-			if apiErr, wasAPIErr := err.(*client.Error); wasAPIErr {
-				// Measure API errors
-                apiErrorCount.With(prometheus.Labels{"error_code": string(apiErr.Type), "path": endpoint, "server": c.serverName}).Inc()
-			} else {
-                // Increment a generic error code counter
-                apiErrorCount.With(prometheus.Labels{"error_code": "generic", "path": endpoint, "server": c.serverName}).Inc()
-             }
-			return
+			if _, wasAPIErr := err.(*client.Error); !wasAPIErr {
+				// TODO: measure API errors by code?
+				return
+			}
 		}
-		queryLatency.With(prometheus.Labels{"path": endpoint, "server": c.serverName}).Observe(endTime.Sub(startTime).Seconds())
+		queryLatency.With(prometheus.Labels{"endpoint": endpoint, "server": c.serverName}).Observe(endTime.Sub(startTime).Seconds())
 	}()
 
 	var resp client.APIResponse