test(e2e): retry transient network errors on idempotent requests

The TestEGUpgrade teardown intermittently fails with a transient
connection-level error when the gateway-api conformance suite's cleanup
deletes the upgrade GatewayClass:

    Delete ".../gatewayclasses/upgrade": EOF

This is a test-infra flake, not a regression in the code under test: it
shows up on unrelated PRs, and the failing DELETE runs in the suite's
registered t.Cleanup after the test body already succeeded.

Root cause: client-go's rest.Request only retries GET on transient
errors (IsProbableEOF/IsConnectionReset/IsHTTP2ConnectionLost); DELETE
and other write verbs surface the error immediately. So when a DELETE
reaches the apiserver but the response is lost (EOF / connection reset /
HTTP/2 GOAWAY / timeout), teardown fails even though the resource was
already deleted.

Fix: install a retrying http.RoundTripper via rest.Config.WrapTransport in
test/utils/kubernetes NewClient. WrapTransport is the only hook inherited
by the conformance suite's per-test client rebuild (setClientsetForTest
calls client.New(suite.RestConfig, ...)), so a client.Client wrapper would
be bypassed while the transport wrapper is not.

Retry policy (conservative):
- Methods: GET, HEAD, DELETE only. Never POST/PATCH/PUT.
- Body: only retry when replayable (nil / http.NoBody / GetBody != nil).
  client-go builds DELETE bodies from bytes.NewReader, and the stdlib
  auto-populates GetBody for *bytes.Reader, so DeleteOptions bodies are
  replayable. http.NoBody must be checked explicitly because the stdlib
  substitutes it for a zero-length body instead of nil.
- Before each retry after the first: re-derive the body via GetBody and
  assign to a Clone (Clone copies GetBody but does not reset Body).
- Errors retried: EOF, ErrUnexpectedEOF, connection reset, HTTP/2
  GOAWAY/connection-lost, net.Error timeouts.
- HTTP status codes are NOT retried (5xx/429/404 returned as-is), so a
  404 from a retried DELETE flows through to MustApplyWithCleanup's
  existing IsNotFound -> success handling.
- Close any partial resp.Body before retrying; honor context cancellation
  in the backoff select.
- Compose with any pre-existing cfg.WrapTransport via transport.Wrappers.

Adds unit tests covering: EOF retry on GET, DELETE-with-body replay via
GetBody, non-retry of POST/non-replayable bodies, non-retry of HTTP
status codes (incl. 404 pass-through), attempt exhaustion, context
cancellation, and timeout (net.Error) retry.

Fixes #9467

Signed-off-by: liuhy <liuhongyu@apache.org>

Author: Aias00

test/utils/kubernetes/client.go

@@ -6,11 +6,17 @@
 package kubernetes
 
 import (
+	"errors"
+	"net"
+	"net/http"
 	"testing"
+	"time"
 
 	"github.com/stretchr/testify/require"
 	batchv1 "k8s.io/api/batch/v1"
+	netutil "k8s.io/apimachinery/pkg/util/net"
 	"k8s.io/client-go/rest"
+	"k8s.io/client-go/transport"
 	"sigs.k8s.io/controller-runtime/pkg/client"
 	"sigs.k8s.io/controller-runtime/pkg/client/config"
 	gwapiv1 "sigs.k8s.io/gateway-api/apis/v1"
@@ -25,6 +31,23 @@ func NewClient(t *testing.T) (client.Client, *rest.Config) {
 	cfg, err := config.GetConfig()
 	require.NoError(t, err)
 
+	// Install a transport wrapper that retries transient network errors
+	// (e.g. EOF, connection reset, HTTP/2 GOAWAY, timeout) for idempotent
+	// requests whose body can be replayed. This is necessary because the
+	// gateway-api conformance suite rebuilds its client per test via
+	// client.New(suite.RestConfig, ...) (see setClientsetForTest), so a
+	// controller-runtime client wrapper returned from this function would be
+	// bypassed. The transport wrapper is the only hook inherited by those
+	// rebuilt clients, and it also covers the raw client returned here.
+	//
+	// Notably, client-go only retries GET on transient errors; DELETE and other
+	// write verbs surface the error immediately (e.g. "Delete ... : EOF"),
+	// which is the source of teardown flakes. Retrying replayable DELETEs here
+	// closes that gap: if the first DELETE reached the apiserver but the
+	// response was lost, the retry may return 404, which conformance cleanup
+	// already treats as success.
+	cfg.Wrap(transport.Wrappers(cfg.WrapTransport, newRetryTransportWrapper))
+
 	c, err := client.New(cfg, client.Options{})
 	require.NoError(t, err)
 
@@ -42,3 +65,126 @@ func CheckInstallScheme(t *testing.T, c client.Client) {
 	require.NoError(t, egv1a1.AddToScheme(c.Scheme()))
 	require.NoError(t, batchv1.AddToScheme(c.Scheme()))
 }
+
+// retryTransportMaxAttempts bounds the total attempts (1 initial + retries).
+const retryTransportMaxAttempts = 4
+
+// retryTransportBaseBackoff is the initial backoff between retries; it grows
+// by retryTransportBackoffFactor on each attempt.
+const (
+	retryTransportBaseBackoff   = 100 * time.Millisecond
+	retryTransportBackoffFactor = 2
+)
+
+// newRetryTransportWrapper returns a transport.WrapperFunc that layers a
+// retrying RoundTripper over the base transport.
+func newRetryTransportWrapper(base http.RoundTripper) http.RoundTripper {
+	if base == nil {
+		base = http.DefaultTransport
+	}
+	return &retryTransport{base: base}
+}
+
+// retryTransport retries transient, connection-level errors for idempotent
+// requests whose body can be safely replayed.
+type retryTransport struct {
+	base http.RoundTripper
+}
+
+func (r *retryTransport) RoundTrip(req *http.Request) (*http.Response, error) {
+	// Only retry idempotent verbs with a replayable body. POST/PATCH/PUT are
+	// never retried; a request with a body and no GetBody cannot be replayed
+	// safely after the body has been (partially) consumed.
+	if !retryableMethod(req.Method) || !replayableBody(req) {
+		return r.base.RoundTrip(req)
+	}
+
+	var (
+		lastErr    error
+		backoff    = retryTransportBaseBackoff
+		attemptReq = req
+	)
+
+	for attempt := 0; attempt < retryTransportMaxAttempts; attempt++ {
+		// On every attempt after the first, back off before retrying. The sleep
+		// is done BEFORE creating a fresh body so that a context cancellation
+		// observed here does not leak a body ReadCloser obtained from GetBody.
+		if attempt > 0 {
+			select {
+			case <-req.Context().Done():
+				return nil, req.Context().Err()
+			case <-time.After(backoff):
+			}
+			backoff *= retryTransportBackoffFactor
+
+			// Reset the body from GetBody so the base transport reads a fresh
+			// copy; the original req.Body may have been consumed by the previous
+			// attempt.
+			if req.GetBody != nil {
+				body, err := req.GetBody()
+				if err != nil {
+					return nil, err
+				}
+				attemptReq = req.Clone(req.Context())
+				attemptReq.Body = body
+			}
+		}
+
+		resp, err := r.base.RoundTrip(attemptReq)
+		if !isTransientNetworkError(err) {
+			// Either success (err == nil) or a non-retryable error: return as-is.
+			return resp, err
+		}
+
+		// Transient error: drain and close any partial response before retrying
+		// to avoid leaking the underlying connection.
+		if resp != nil {
+			_ = resp.Body.Close()
+		}
+		lastErr = err
+	}
+
+	return nil, lastErr
+}
+
+// retryableMethod reports whether the HTTP method is safe to retry without
+// risking duplicate side effects.
+func retryableMethod(method string) bool {
+	switch method {
+	case http.MethodGet, http.MethodHead, http.MethodDelete:
+		return true
+	default:
+		return false
+	}
+}
+
+// replayableBody reports whether the request body can be re-sent on a retry.
+// A request with no body, an explicit http.NoBody, or a body that can be
+// obtained again via GetBody is replayable.
+func replayableBody(req *http.Request) bool {
+	return req.Body == nil || req.Body == http.NoBody || req.GetBody != nil
+}
+
+// isTransientNetworkError reports whether err is a connection-level error that
+// may have occurred before or during the response, making a retry safe. It does
+// not treat HTTP status codes (e.g. 5xx, 429) as retryable: those are returned
+// to the caller, preserving the apiserver's status semantics (e.g. a 404 on a
+// retried DELETE flows through to conformance cleanup's IsNotFound handling).
+func isTransientNetworkError(err error) bool {
+	if err == nil {
+		return false
+	}
+	if netutil.IsProbableEOF(err) || netutil.IsConnectionReset(err) || netutil.IsHTTP2ConnectionLost(err) {
+		return true
+	}
+	// net.Error covers request timeouts (DeadlineExceeded), which are safe to
+	// retry for idempotent verbs.
+	var netErr net.Error
+	if errors.As(err, &netErr) && netErr.Timeout() {
+		return true
+	}
+	return false
+}
+
+// compile-time assertion that retryTransport satisfies http.RoundTripper.
+var _ http.RoundTripper = &retryTransport{}