fix(images): 服务器镜像在创建时固定到仓库 digest，更换镜像需确认备份，安装器重建前先固定旧服并推送不可变版本标签

cmd/felis/api.go

@@ -17,6 +17,7 @@ import (
 	"felis.lolicon.best/internal/build"
 	"felis.lolicon.best/internal/config"
 	"felis.lolicon.best/internal/fileedit"
+	"felis.lolicon.best/internal/imagepin"
 	"felis.lolicon.best/internal/mail"
 	"felis.lolicon.best/internal/naming"
 	"felis.lolicon.best/internal/panel"
@@ -268,6 +269,7 @@ func cmdAPI(args []string, stdout, stderr io.Writer) int {
 		BuildLogs:   api.NewK8sBuildLogStreamer(clientset, cfg.Registry.BuildNamespace),
 		Internal:    api.BearerTokenAuth{Token: token},
 		Builder:     builder,
+		Images:      imagePinner(cfg.Registry.URL),
 		Restorer:    restorer,
 		Backuper:    backuper,
 		JobStatus:   api.NewK8sJobStatus(cl, cfg.K8s.Namespace),
@@ -568,3 +570,13 @@ func mailLimit(perHour int) api.RateLimit {
 	}
 	return api.RateLimit{Burst: max(perHour/4, 5), PerMinute: float64(perHour) / 60}
 }
+
+// imagePinner resolves a new server's image against the platform registry
+// through its in-cluster Service, the address its refs already spell. An install
+// without a registry has no platform-built images to pin.
+func imagePinner(registry string) api.ImagePinner {
+	if registry == "" {
+		return nil
+	}
+	return imagepin.Resolver{Registry: registry}
+}

cmd/felis/pinimages.go

@@ -0,0 +1,126 @@
+package main
+
+import (
+	"context"
+	"errors"
+	"flag"
+	"fmt"
+	"io"
+	"strings"
+	"time"
+
+	"felis.lolicon.best/internal/apis/felis/v1alpha1"
+	"felis.lolicon.best/internal/imagepin"
+	"felis.lolicon.best/internal/platform"
+	"k8s.io/apimachinery/pkg/api/meta"
+	"sigs.k8s.io/controller-runtime/pkg/client"
+)
+
+// defaultRegistryURL is the [registry] url every install uses; deploy/bootstrap.sh
+// spells the same value as REGISTRY_URL.
+const defaultRegistryURL = "registry.felis.svc:5000"
+
+// cmdPinImages pins every user server whose spec.image still names a mutable tag
+// in the platform registry to the digest that tag names now (internal/imagepin).
+// felis-api pins on create, so this covers the servers created before it did.
+//
+// deploy/bootstrap.sh runs it before it rebuilds the game images and pushes them
+// over the same tags: run after the push, it would pin those servers to the new
+// build, which is exactly the silent Minecraft upgrade pinning exists to stop.
+// It reaches the registry through the node's loopback hostPort, the same way the
+// installer pushes.
+func cmdPinImages(args []string, stdout, stderr io.Writer) int {
+	fs := flag.NewFlagSet("pin-images", flag.ContinueOnError)
+	fs.SetOutput(stderr)
+	namespace := fs.String("namespace", platform.DefaultMinecraftNamespace, "namespace the MinecraftServers live in")
+	registry := fs.String("registry", defaultRegistryURL, "registry host[:port] the image refs spell")
+	endpoint := fs.String("endpoint", "", "host[:port] to reach the registry at (default: 127.0.0.1 on the registry's port, its hostPort on this node)")
+	if err := fs.Parse(args); err != nil {
+		if errors.Is(err, flag.ErrHelp) {
+			return 0
+		}
+		return 2
+	}
+	if *endpoint == "" {
+		*endpoint = loopbackEndpoint(*registry)
+	}
+	cl, err := buildSystemServerClient()
+	if err != nil {
+		fmt.Fprintf(stderr, "felis pin-images: %v\n", err)
+		return 1
+	}
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Minute)
+	defer cancel()
+	outcomes, err := pinUserServerImages(ctx, cl, *namespace, imagepin.Resolver{Registry: *registry, Endpoint: *endpoint})
+	if meta.IsNoMatchError(err) {
+		fmt.Fprintln(stdout, "felis pin-images: no MinecraftServer CRD yet, so no server to pin")
+		return 0
+	}
+	if err != nil {
+		fmt.Fprintf(stderr, "felis pin-images: %v\n", err)
+		return 1
+	}
+	if len(outcomes) == 0 {
+		fmt.Fprintln(stdout, "felis pin-images: every user server already runs a pinned image")
+		return 0
+	}
+	fmt.Fprintln(stdout, "felis pin-images: pinning user servers to the build their tag names now:")
+	exit := 0
+	for _, o := range outcomes {
+		if o.err != nil {
+			fmt.Fprintf(stdout, "  - %s: ERROR %v\n", o.name, o.err)
+			exit = 1
+			continue
+		}
+		fmt.Fprintf(stdout, "  - %s: %s\n", o.name, strings.Join(o.changes, ", "))
+	}
+	return exit
+}
+
+// loopbackEndpoint is the registry's port on 127.0.0.1: the registry Deployment
+// binds it as a hostPort, and containerd's mirror and the installer's pushes use
+// the same address.
+func loopbackEndpoint(registry string) string {
+	if i := strings.LastIndex(registry, ":"); i >= 0 {
+		return "127.0.0.1" + registry[i:]
+	}
+	return "127.0.0.1"
+}
+
+// pinUserServerImages patches spec.image of every user server whose image the
+// resolver covers and is not yet pinned. System servers are left on their tags:
+// the installer rebuilds and restarts them on purpose (restart_existing_system_servers).
+// A server that is already pinned, or runs an image from elsewhere, produces no
+// outcome, so a pinned fleet reports nothing. A running server restarts once as
+// the operator rolls its StatefulSet onto the pinned ref, which is the build it
+// already runs.
+func pinUserServerImages(ctx context.Context, cl client.Client, namespace string, r imagepin.Resolver) ([]systemServerOutcome, error) {
+	var list v1alpha1.MinecraftServerList
+	if err := cl.List(ctx, &list, client.InNamespace(namespace)); err != nil {
+		return nil, fmt.Errorf("list servers: %w", err)
+	}
+	var out []systemServerOutcome
+	for i := range list.Items {
+		ms := &list.Items[i]
+		if ms.Labels[v1alpha1.LabelSystemRole] != "" || imagepin.Pinned(ms.Spec.Image) || !r.Covers(ms.Spec.Image) {
+			continue
+		}
+		pinned, err := r.Pin(ctx, ms.Spec.Image)
+		if errors.Is(err, imagepin.ErrNotFound) {
+			err = fmt.Errorf("%s is not in the registry, so there is no build to pin it to; left unpinned: %w", ms.Spec.Image, err)
+		}
+		if err != nil {
+			out = append(out, systemServerOutcome{name: ms.Name, err: err})
+			continue
+		}
+		patch := client.MergeFrom(ms.DeepCopy())
+		ms.Spec.Image = pinned
+		if err := cl.Patch(ctx, ms, patch); err != nil {
+			out = append(out, systemServerOutcome{name: ms.Name, err: fmt.Errorf("patch %s: %w", ms.Name, err)})
+			continue
+		}
+		out = append(out, systemServerOutcome{name: ms.Name, available: true, updated: true,
+			changes: []string{"spec.image pinned to " + pinned}})
+	}
+	return out, nil
+}

cmd/felis/pinimages_test.go

@@ -0,0 +1,112 @@
+package main
+
+import (
+	"context"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	"felis.lolicon.best/internal/apis/felis/v1alpha1"
+	"felis.lolicon.best/internal/imagepin"
+	"felis.lolicon.best/internal/naming"
+	"k8s.io/apimachinery/pkg/api/meta"
+	"k8s.io/apimachinery/pkg/runtime/schema"
+	"sigs.k8s.io/controller-runtime/pkg/client"
+	"sigs.k8s.io/controller-runtime/pkg/client/fake"
+	"sigs.k8s.io/controller-runtime/pkg/client/interceptor"
+)
+
+const pinTestDigest = "sha256:2222222222222222222222222222222222222222222222222222222222222222"
+
+// TestPinUserServerImages pins exactly the user servers still on a platform tag,
+// reports a tag the registry lost as an error without touching that server, and
+// has nothing left to do on a second pass.
+func TestPinUserServerImages(t *testing.T) {
+	reg := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.URL.Path != "/v2/felis/paper/manifests/demo" {
+			http.NotFound(w, r)
+			return
+		}
+		w.Header().Set("Docker-Content-Digest", pinTestDigest)
+	}))
+	defer reg.Close()
+	res := imagepin.Resolver{Registry: defaultRegistryURL, Endpoint: strings.TrimPrefix(reg.URL, "http://")}
+
+	paper := defaultRegistryURL + "/felis/paper:demo"
+	mk := func(name, image, role string) *v1alpha1.MinecraftServer {
+		ms := &v1alpha1.MinecraftServer{}
+		ms.Name, ms.Namespace = name, "minecraft"
+		ms.Spec.Image = image
+		if role != "" {
+			ms.Labels = map[string]string{v1alpha1.LabelSystemRole: role}
+		}
+		return ms
+	}
+	cl := fake.NewClientBuilder().WithScheme(newSystemServerScheme(t)).WithObjects(
+		mk("legacy", paper, ""),
+		mk("pinned", paper+"@sha256:"+strings.Repeat("3", 64), ""),
+		mk("external", "docker.io/itzg/minecraft-server:java21", ""),
+		mk("gone", defaultRegistryURL+"/felis/paper:old", ""),
+		mk(naming.SystemLobbyServer, defaultRegistryURL+"/felis/felis-lobby:demo", naming.SystemLobbyServer),
+	).Build()
+	ctx := context.Background()
+
+	outcomes, err := pinUserServerImages(ctx, cl, "minecraft", res)
+	if err != nil {
+		t.Fatalf("pinUserServerImages: %v", err)
+	}
+	byName := map[string]systemServerOutcome{}
+	for _, o := range outcomes {
+		byName[o.name] = o
+	}
+	if len(outcomes) != 2 || byName["legacy"].err != nil || byName["gone"].err == nil {
+		t.Fatalf("outcomes = %+v, want legacy pinned and gone reported", outcomes)
+	}
+
+	want := map[string]string{
+		"legacy":                 paper + "@" + pinTestDigest,
+		"pinned":                 paper + "@sha256:" + strings.Repeat("3", 64),
+		"external":               "docker.io/itzg/minecraft-server:java21",
+		"gone":                   defaultRegistryURL + "/felis/paper:old",
+		naming.SystemLobbyServer: defaultRegistryURL + "/felis/felis-lobby:demo",
+	}
+	for name, image := range want {
+		var ms v1alpha1.MinecraftServer
+		if err := cl.Get(ctx, client.ObjectKey{Namespace: "minecraft", Name: name}, &ms); err != nil {
+			t.Fatalf("get %s: %v", name, err)
+		}
+		if ms.Spec.Image != image {
+			t.Errorf("%s image = %q, want %q", name, ms.Spec.Image, image)
+		}
+	}
+
+	again, err := pinUserServerImages(ctx, cl, "minecraft", res)
+	if err != nil || len(again) != 1 || again[0].name != "gone" {
+		t.Fatalf("second pass = %+v, %v; want only the unresolvable server again", again, err)
+	}
+}
+
+// A fresh install has no CRD yet; the command must read that as nothing to pin.
+func TestPinUserServerImagesNoCRD(t *testing.T) {
+	cl := fake.NewClientBuilder().WithScheme(newSystemServerScheme(t)).WithInterceptorFuncs(interceptor.Funcs{
+		List: func(context.Context, client.WithWatch, client.ObjectList, ...client.ListOption) error {
+			return &meta.NoKindMatchError{GroupKind: schema.GroupKind{Group: "felis.lolicon.best", Kind: "MinecraftServer"}}
+		},
+	}).Build()
+	_, err := pinUserServerImages(context.Background(), cl, "minecraft", imagepin.Resolver{Registry: defaultRegistryURL})
+	if !meta.IsNoMatchError(err) {
+		t.Fatalf("err = %v, want a NoMatch error the command can recognise", err)
+	}
+}
+
+func TestLoopbackEndpoint(t *testing.T) {
+	for in, want := range map[string]string{
+		"registry.felis.svc:5000": "127.0.0.1:5000",
+		"registry.example":        "127.0.0.1",
+	} {
+		if got := loopbackEndpoint(in); got != want {
+			t.Errorf("loopbackEndpoint(%q) = %q, want %q", in, got, want)
+		}
+	}
+}

cmd/felis/run.go

@@ -64,6 +64,7 @@ var commands = map[string]func(args []string, stdout, stderr io.Writer) int{
 	"bootstrap-assets": cmdBootstrapAssets,
 	"init-forwarding":  cmdInitForwarding,
 	"init-volume":      cmdInitVolume,
+	"pin-images":       cmdPinImages,
 	"version":          cmdVersion,
 	"update":           cmdUpdate,
 }

cmd/felis/run_test.go

@@ -43,6 +43,7 @@ func TestRunUnknownCommand(t *testing.T) {
 // decision rather than an oversight.
 var undocumentedCommands = map[string]bool{
 	"bootstrap-assets": true, "init-forwarding": true, "init-volume": true,
+	"pin-images": true,
 }
 
 // The usage text and the dispatch table must describe the same set of commands.