Add Metrics to RetryableHTTPClient (#4545)

* add response body size metric to InstrumentedTransport, add InstrumentedTransport to RetryableHTTPClient

* add doc to responseSizeCounterReadCloser, remove unused field from test

* record response using ContentLength

Author: mustansir14

pkg/common/http.go

@@ -133,8 +133,8 @@ func (t *InstrumentedTransport) RoundTrip(req *http.Request) (*http.Response, er
 	}
 
 	if resp != nil {
-		// record latency and increment counter for non-200 status code
-		recordHTTPResponse(sanitizedURL, resp.StatusCode, duration.Seconds())
+		// record latency, response size and increment counter for non-200 status code
+		recordHTTPResponse(sanitizedURL, resp.StatusCode, duration.Seconds(), resp.ContentLength)
 	}
 
 	return resp, err
@@ -198,7 +198,7 @@ func WithRetryWaitMax(wait time.Duration) ClientOption {
 func PinnedRetryableHttpClient() *http.Client {
 	httpClient := retryablehttp.NewClient()
 	httpClient.Logger = nil
-	httpClient.HTTPClient.Transport = NewCustomTransport(&http.Transport{
+	httpClient.HTTPClient.Transport = NewInstrumentedTransport(NewCustomTransport(&http.Transport{
 		TLSClientConfig: &tls.Config{
 			RootCAs: PinnedCertPool(),
 		},
@@ -212,15 +212,15 @@ func PinnedRetryableHttpClient() *http.Client {
 		IdleConnTimeout:       90 * time.Second,
 		TLSHandshakeTimeout:   10 * time.Second,
 		ExpectContinueTimeout: 1 * time.Second,
-	})
+	}))
 	return httpClient.StandardClient()
 }
 
 func RetryableHTTPClient(opts ...ClientOption) *http.Client {
 	httpClient := retryablehttp.NewClient()
 	httpClient.RetryMax = 3
 	httpClient.Logger = nil
-	httpClient.HTTPClient.Transport = NewCustomTransport(nil)
+	httpClient.HTTPClient.Transport = NewInstrumentedTransport(NewCustomTransport(nil))
 
 	for _, opt := range opts {
 		opt(httpClient)
@@ -234,7 +234,7 @@ func RetryableHTTPClientTimeout(timeOutSeconds int64, opts ...ClientOption) *htt
 	httpClient.RetryMax = 3
 	httpClient.Logger = nil
 	httpClient.HTTPClient.Timeout = time.Duration(timeOutSeconds) * time.Second
-	httpClient.HTTPClient.Transport = NewCustomTransport(nil)
+	httpClient.HTTPClient.Transport = NewInstrumentedTransport(NewCustomTransport(nil))
 
 	for _, opt := range opts {
 		opt(httpClient)

pkg/common/http_metrics.go

@@ -39,6 +39,17 @@ var (
 		},
 		[]string{"url", "status_code"},
 	)
+
+	httpResponseBodySizeBytes = promauto.NewHistogramVec(
+		prometheus.HistogramOpts{
+			Namespace: MetricsNamespace,
+			Subsystem: "http_client",
+			Name:      "response_body_size_bytes",
+			Help:      "Size of HTTP response bodies in bytes, labeled by URL.",
+			Buckets:   prometheus.ExponentialBuckets(100, 10, 5), // [100B, 1KB, 10KB, 100KB, 1MB]
+		},
+		[]string{"url"},
+	)
 )
 
 // sanitizeURL sanitizes a URL to avoid high cardinality metrics.
@@ -92,14 +103,19 @@ func recordHTTPRequest(sanitizedURL string) {
 }
 
 // recordHTTPResponse records metrics for an HTTP response.
-func recordHTTPResponse(sanitizedURL string, statusCode int, durationSeconds float64) {
+func recordHTTPResponse(sanitizedURL string, statusCode int, durationSeconds float64, contentLength int64) {
 	// Record latency
 	httpRequestDuration.WithLabelValues(sanitizedURL).Observe(durationSeconds)
 
 	// Record non-200 responses
 	if statusCode != 200 {
 		httpNon200ResponsesTotal.WithLabelValues(sanitizedURL, strconv.Itoa(statusCode)).Inc()
 	}
+
+	// Record response body size if known
+	if contentLength >= 0 {
+		httpResponseBodySizeBytes.WithLabelValues(sanitizedURL).Observe(float64(contentLength))
+	}
 }
 
 // recordNetworkError records metrics for failed HTTP response

pkg/common/http_test.go

@@ -405,6 +405,73 @@ func TestSaneHttpClientMetrics(t *testing.T) {
 	}
 }
 
+func TestRetryableHttpClientMetrics(t *testing.T) {
+	// Create a test server that returns different status codes
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		switch r.URL.Path {
+		case "/success":
+			w.WriteHeader(http.StatusOK)
+			_, _ = w.Write([]byte("success"))
+		case "/error":
+			w.WriteHeader(http.StatusInternalServerError)
+			_, _ = w.Write([]byte("error"))
+		case "/notfound":
+			w.WriteHeader(http.StatusNotFound)
+			_, _ = w.Write([]byte("not found"))
+		default:
+			w.WriteHeader(http.StatusOK)
+			_, _ = w.Write([]byte("default"))
+		}
+	}))
+	defer server.Close()
+
+	// Create a RetryableHttpClient
+	client := RetryableHTTPClient()
+
+	testCases := []struct {
+		name               string
+		path               string
+		expectedStatusCode int
+	}{
+		{
+			name:               "successful request",
+			path:               "/success",
+			expectedStatusCode: 200,
+		},
+		{
+			name:               "not found request",
+			path:               "/notfound",
+			expectedStatusCode: 404,
+		},
+	}
+
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			var requestURL string
+			if strings.HasPrefix(tc.path, "http") {
+				requestURL = tc.path
+			} else {
+				requestURL = server.URL + tc.path
+			}
+
+			// Get initial metric values
+			sanitizedURL := sanitizeURL(requestURL)
+			initialRequestsTotal := testutil.ToFloat64(httpRequestsTotal.WithLabelValues(sanitizedURL))
+
+			// Make the request
+			resp, err := client.Get(requestURL)
+
+			require.NoError(t, err)
+			defer resp.Body.Close()
+			assert.Equal(t, tc.expectedStatusCode, resp.StatusCode)
+
+			// Check that request counter was incremented
+			requestsTotal := testutil.ToFloat64(httpRequestsTotal.WithLabelValues(sanitizedURL))
+			assert.Equal(t, initialRequestsTotal+1, requestsTotal)
+		})
+	}
+}
+
 func TestInstrumentedTransport(t *testing.T) {
 	// Create a mock transport that we can control
 	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {