retry on 408 error

internal/network/network.go

@@ -71,7 +71,7 @@ func WithRetry(ctx context.Context, lim *rate.Limiter, maxAttempts int, fn func(
 			time.Sleep(rle.RetryAfter)
 			continue
 		} else if errors.As(cbErr, &sce) {
-			if sce.Code >= http.StatusInternalServerError && sce.Code <= 599 {
+			if isRecoverable(sce.Code) {
 				// possibly transient error
 				delay := waitFn(attempt)
 				tracelogf(ctx, "info", "got server error %d, sleeping %s", sce.Code, delay)
@@ -88,6 +88,11 @@ func WithRetry(ctx context.Context, lim *rate.Limiter, maxAttempts int, fn func(
 	return nil
 }
 
+// isRecoverable returns true if the status code is a recoverable error.
+func isRecoverable(statusCode int) bool {
+	return (statusCode >= http.StatusInternalServerError && statusCode <= 599) || statusCode == 408
+}
+
 // cubicWait is the wait time function.  Time is calculated as (x+2)^3 seconds,
 // where x is the current attempt number. The maximum wait time is capped at 5
 // minutes.

internal/network/network_test.go

@@ -253,3 +253,33 @@ func Test_cubicWait(t *testing.T) {
 		})
 	}
 }
+
+func Test_isRecoverable(t *testing.T) {
+	type args struct {
+		statusCode int
+	}
+	tests := []struct {
+		name string
+		args args
+		want bool
+	}{
+		{"500", args{500}, true},
+		{"502", args{502}, true},
+		{"503", args{503}, true},
+		{"504", args{504}, true},
+		{"598", args{598}, true},
+		{"599", args{599}, true},
+		{"200", args{200}, false},
+		{"400", args{400}, false},
+		{"404", args{404}, false},
+		{"408", args{408}, true},
+		{"429", args{429}, false},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			if got := isRecoverable(tt.args.statusCode); got != tt.want {
+				t.Errorf("isRecoverable() = %v, want %v", got, tt.want)
+			}
+		})
+	}
+}