fix(builder): add timeouts to mirror pull client and probe real maninfest path

builder/mirror/prober.go

@@ -35,11 +35,19 @@ type probeResult struct {
 	latency time.Duration
 }
 
-// Probe checks every candidate's /v2/ registry endpoint concurrently and
-// returns only the alive ones, sorted by ascending latency. A mirror is alive
-// when /v2/ answers 200 or 401 (an auth challenge still proves a working
-// registry frontend). Candidates keep their scheme; entries without one are
-// probed via https.
+// probeManifestPath is a real, tiny docker.io manifest. Probing it instead of
+// the bare /v2/ ping exercises the mirror's actual proxy path, so a frontend
+// that answers pings but stalls on manifests fails the probe timeout.
+const probeManifestPath = "/v2/library/alpine/manifests/latest"
+
+// Probe fetches a real manifest from every candidate concurrently and returns
+// only the alive ones, sorted by ascending latency. 200 and 401 both count as
+// alive: token-auth mirrors (e.g. daocloud) answer 401 to anonymous manifest
+// requests, and the containerd/BuildKit token flow handles that during real
+// pulls. The probe cannot catch every stall (a mirror may serve alpine fine
+// and hang on another image) — the pull-side client timeout is the hard
+// safety net; the probe only filters and orders. Candidates keep their
+// scheme; entries without one are probed via https.
 func Probe(ctx context.Context, candidates []string, timeout time.Duration) []string {
 	if len(candidates) == 0 {
 		return nil
@@ -76,12 +84,18 @@ func probeOne(ctx context.Context, client *http.Client, mirrorURL string) (time.
 	if !strings.HasPrefix(endpoint, "http://") && !strings.HasPrefix(endpoint, "https://") {
 		endpoint = "https://" + endpoint
 	}
-	endpoint = strings.TrimSuffix(endpoint, "/") + "/v2/"
+	endpoint = strings.TrimSuffix(endpoint, "/") + probeManifestPath
 	req, err := http.NewRequestWithContext(ctx, http.MethodGet, endpoint, nil)
 	if err != nil {
 		logrus.Debugf("probe mirror %s: build request failure: %v", mirrorURL, err)
 		return 0, false
 	}
+	req.Header.Set("Accept", strings.Join([]string{
+		"application/vnd.docker.distribution.manifest.v2+json",
+		"application/vnd.docker.distribution.manifest.list.v2+json",
+		"application/vnd.oci.image.manifest.v1+json",
+		"application/vnd.oci.image.index.v1+json",
+	}, ", "))
 	start := time.Now()
 	resp, err := client.Do(req)
 	if err != nil {

builder/mirror/prober_test.go

@@ -11,12 +11,33 @@ import (
 // capability_id: rainbond.builder.dynamic-mirror-probe
 func TestProbeFiltersAndSortsByLatency(t *testing.T) {
 	slowOK := newRegistryStub(t, http.StatusOK, 300*time.Millisecond)
-	fastUnauthorized := newRegistryStub(t, http.StatusUnauthorized, 0)
+	fastOK := newRegistryStub(t, http.StatusOK, 0)
 	dead := newRegistryStub(t, http.StatusInternalServerError, 0)
 
-	got := Probe(context.Background(), []string{slowOK, fastUnauthorized, dead}, 2*time.Second)
+	got := Probe(context.Background(), []string{slowOK, fastOK, dead}, 2*time.Second)
 
-	assertStringSlice(t, got, []string{fastUnauthorized, slowOK})
+	assertStringSlice(t, got, []string{fastOK, slowOK})
+}
+
+// token 认证类 mirror（如 daocloud）对匿名 manifest 请求回 401，但真实拉取时
+// resolver 会走 token 流程，因此 401 必须判活。
+func TestProbeTokenAuthMirrorIsAlive(t *testing.T) {
+	authChallenge := newRegistryStub(t, http.StatusUnauthorized, 0)
+	got := Probe(context.Background(), []string{authChallenge}, time.Second)
+	assertStringSlice(t, got, []string{authChallenge})
+}
+
+// manifest 路径挂起不响应的“假活”源要在探活超时内排除
+// （docker.xuanyuan.me 卡死构建事故的场景之一）。
+func TestProbeManifestStallTreatedAsDead(t *testing.T) {
+	stalled := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		time.Sleep(2 * time.Second) // longer than probe timeout
+	}))
+	t.Cleanup(stalled.Close)
+	got := Probe(context.Background(), []string{stalled.URL}, 500*time.Millisecond)
+	if len(got) != 0 {
+		t.Fatalf("stalled mirror must be treated as dead, got %v", got)
+	}
 }
 
 func TestProbeUnreachableHostDropped(t *testing.T) {
@@ -32,12 +53,12 @@ func TestProbeEmptyInput(t *testing.T) {
 	}
 }
 
-// newRegistryStub serves /v2/ with the given status after an artificial delay
-// and returns the server base URL.
+// newRegistryStub serves the probe manifest path with the given status after
+// an artificial delay and returns the server base URL.
 func newRegistryStub(t *testing.T, status int, delay time.Duration) string {
 	t.Helper()
 	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		if r.URL.Path != "/v2/" {
+		if r.URL.Path != probeManifestPath {
 			w.WriteHeader(http.StatusNotFound)
 			return
 		}

builder/sources/mirror_hosts.go

@@ -20,8 +20,10 @@ package sources
 
 import (
 	"crypto/tls"
+	"net"
 	"net/http"
 	"strings"
+	"time"
 
 	refdocker "github.com/containerd/containerd/reference/docker"
 	"github.com/containerd/containerd/remotes/docker"
@@ -60,8 +62,16 @@ func mirrorRegistryHost(mirrorURL string) docker.RegistryHost {
 	}
 	host = strings.TrimSuffix(host, "/")
 	return docker.RegistryHost{
+		// 必须带超时：探活通过的 mirror 仍可能在取 manifest/blob 时挂起不响应，
+		// containerd 这条 pull 路径外层没有 ctx 超时，没有这里的超时会把构建卡死。
 		Client: &http.Client{
-			Transport: &http.Transport{TLSClientConfig: &tls.Config{InsecureSkipVerify: true}},
+			Transport: &http.Transport{
+				DialContext:           (&net.Dialer{Timeout: 10 * time.Second}).DialContext,
+				TLSClientConfig:       &tls.Config{InsecureSkipVerify: true},
+				TLSHandshakeTimeout:   10 * time.Second,
+				ResponseHeaderTimeout: 30 * time.Second,
+				IdleConnTimeout:       90 * time.Second,
+			},
 		},
 		Authorizer:   docker.NewDockerAuthorizer(),
 		Host:         host,