fix(otel): Map 500 errors to 503 (#13173)

Author: salvacorts

pkg/distributor/http.go

@@ -23,7 +23,26 @@ func (d *Distributor) PushHandler(w http.ResponseWriter, r *http.Request) {
 }
 
 func (d *Distributor) OTLPPushHandler(w http.ResponseWriter, r *http.Request) {
-	d.pushHandler(w, r, push.ParseOTLPRequest)
+	interceptor := newOtelErrorHeaderInterceptor(w)
+	d.pushHandler(interceptor, r, push.ParseOTLPRequest)
+}
+
+// otelErrorHeaderInterceptor maps 500 errors to 503.
+// According to the OTLP specification, 500 errors are never retried on the client side, but 503 are.
+type otelErrorHeaderInterceptor struct {
+	http.ResponseWriter
+}
+
+func newOtelErrorHeaderInterceptor(w http.ResponseWriter) *otelErrorHeaderInterceptor {
+	return &otelErrorHeaderInterceptor{ResponseWriter: w}
+}
+
+func (i *otelErrorHeaderInterceptor) WriteHeader(statusCode int) {
+	if statusCode == http.StatusInternalServerError {
+		statusCode = http.StatusServiceUnavailable
+	}
+
+	i.ResponseWriter.WriteHeader(statusCode)
 }
 
 func (d *Distributor) pushHandler(w http.ResponseWriter, r *http.Request, pushRequestParser push.RequestParser) {

pkg/distributor/http_test.go

@@ -82,6 +82,38 @@ func TestRequestParserWrapping(t *testing.T) {
 	require.True(t, called)
 }
 
+func Test_OtelErrorHeaderInterceptor(t *testing.T) {
+	for _, tc := range []struct {
+		name         string
+		inputCode    int
+		expectedCode int
+	}{
+		{
+			name:         "500",
+			inputCode:    http.StatusInternalServerError,
+			expectedCode: http.StatusServiceUnavailable,
+		},
+		{
+			name:         "400",
+			inputCode:    http.StatusBadRequest,
+			expectedCode: http.StatusBadRequest,
+		},
+		{
+			name:         "204",
+			inputCode:    http.StatusNoContent,
+			expectedCode: http.StatusNoContent,
+		},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			r := httptest.NewRecorder()
+			i := newOtelErrorHeaderInterceptor(r)
+
+			http.Error(i, "error", tc.inputCode)
+			require.Equal(t, tc.expectedCode, r.Code)
+		})
+	}
+}
+
 func stubParser(_ string, _ *http.Request, _ push.TenantsRetention, _ push.Limits, _ push.UsageTracker) (*logproto.PushRequest, *push.Stats, error) {
 	return &logproto.PushRequest{}, &push.Stats{}, nil
 }