net/http: HTTP client hangs indefinitely if the request handler does not return

[tkashem]

Go version

go version go1.21.6 linux/amd64

Output of go env in your module/workspace:

GO111MODULE=''
GOARCH='amd64'
GOBIN=''
GOCACHE='/home/akashem/.cache/go-build'
GOENV='/home/akashem/.config/go/env'
GOEXE=''
GOEXPERIMENT=''
GOFLAGS=''
GOHOSTARCH='amd64'
GOHOSTOS='linux'
GOINSECURE=''
GOMODCACHE='/home/akashem/go/pkg/mod'
GONOPROXY=''
GONOSUMDB=''
GOOS='linux'
GOPATH='/home/akashem/go'
GOPRIVATE=''
GOPROXY='https://proxy.golang.org,direct'
GOROOT='/usr/lib/go'
GOSUMDB='sum.golang.org'
GOTMPDIR=''
GOTOOLCHAIN='auto'
GOTOOLDIR='/usr/lib/go/pkg/tool/linux_amd64'
GOVCS=''
GOVERSION='go1.21.6'
GCCGO='gccgo'
GOAMD64='v1'
AR='ar'
CC='gcc'
CXX='g++'
CGO_ENABLED='1'
GOMOD='/home/akashem/go/src/k8s.io/kubernetes/go.mod'
GOWORK=''
CGO_CFLAGS='-O2 -g'
CGO_CPPFLAGS=''
CGO_CXXFLAGS='-O2 -g'
CGO_FFLAGS='-O2 -g'
CGO_LDFLAGS='-O2 -g'
PKG_CONFIG='pkg-config'
GOGCCFLAGS='-fPIC -m64 -pthread -Wl,--no-gc-sections -fmessage-length=0 -ffile-prefix-map=/tmp/go-build3786731428=/tmp/go-build -gno-record-gcc-switches'

What did you do?

I have a test where the client hangs indefinitely waiting for a response/error from the server, this is the setup of the test:

• protocol: HTTP/1x
• the client does not set any timeout on its side, the client request context does not have a deadline
• on the server, per-request read (100ms) and write (200ms) deadline is set for the request
• the request handler blocks indefinitely on a channel that is closed by the client when it receives a response from the server

func TestClientHangingForeverWithHTTP1(t *testing.T) {
	clientDoneCh, handlerDoneCh := make(chan struct{}), make(chan struct{})
	server := httptest.NewUnstartedServer(http.HandlerFunc(func(w http.ResponseWriter, req *http.Request) {
		defer close(handlerDoneCh)

		ctrl := http.NewResponseController(w)
		if err := ctrl.SetWriteDeadline(time.Now().Add(200 * time.Millisecond)); err != nil {
			t.Errorf("expected no error from SetWriteDeadline, but got: %v", err)
			return
		}
		if err := ctrl.SetReadDeadline(time.Now().Add(100 * time.Millisecond)); err != nil {
			t.Errorf("expected no error from SetReadDeadline, but got: %v", err)
			return
		}

		<-clientDoneCh
	}))

	defer server.Close()
	server.StartTLS()

	client := server.Client()
	func() {
		defer close(clientDoneCh)
		client.Get(server.URL + "/foo")
	}()

	select {
	case <-handlerDoneCh:
	case <-time.After(time.Minute):
		t.Errorf("expected the request handler to have terminated")
	}
}

What did you see happen?

The client waits indefinitely for a response from the server. SetWriteDeadline does not result in the connection being reset or closed after the deadline exceeds.

It seems the request handler in http/1x is not executed in a new goroutine by design

go/src/net/http/server.go

Lines 2031 to 2039 in b8ac61e

	// HTTP cannot have multiple simultaneous active requests.[*]
	// Until the server replies to this request, it can't read another,
	// so we might as well run the handler in this goroutine.
	// [*] Not strictly true: HTTP pipelining. We could let them all process
	// in parallel even if their responses need to be serialized.
	// But we're not going to implement HTTP pipelining because it
	// was never deployed in the wild and the answer is HTTP/2.
	inFlightResponse = w
	serverHandler{c.server}.ServeHTTP(w, w.req)

What did you expect to see?

For http2, SetWriteDeadline will result in the stream being reset and thus unblocking the client after write timeout, even though the handler is blocked. The same version of the test does not fail with http2

Is this working as expected for http/1x?

There are clients that do not configure any timeout, these clients will hang indefinitely, or until the kernel TCP timeout takes effect. We are trying to replace the use of TimeoutHandler in the kubernetes apiserver with the ResponseController.
Please let us know if you have any suggestions, thanks!

[seankhliao]

cc @neild

I think timeout with a 503 status and writedeadline are different operations