feat(operator): 平台升级不再重启运行中的游戏服，registry 清理保留游戏 pod 在用的镜像

cmd/felis/api.go

@@ -284,9 +284,10 @@ func cmdAPI(args []string, stdout, stderr io.Writer) int {
 		rcfg = reaper.DefaultConfig()
 	}
 
+	cluster := api.NewK8sCluster(cl, cfg.K8s.Namespace)
 	a := &api.API{
 		Repo:    repo,
-		Cluster: api.NewK8sCluster(cl, cfg.K8s.Namespace),
+		Cluster: cluster,
 		Console: api.NewK8sConsole(cl, cfg.K8s.Namespace),
 		Logs:    api.NewK8sLogStreamer(clientset, cfg.K8s.Namespace),
 		// Build-log stream (spec §16) is scoped to the BUILD namespace — the same
@@ -408,7 +409,7 @@ func cmdAPI(args []string, stdout, stderr io.Writer) int {
 	// reconciles it, but this loop converges builds nobody is polling.
 	go reconcileBuilds(ctx, builder, stderr)
 
-	if pruner := registryPruner(cfg, builder.Store, a.Cluster, stderr); pruner != nil {
+	if pruner := registryPruner(cfg, builder.Store, cluster, stderr); pruner != nil {
 		go pruner.Loop(ctx, registryPruneInterval)
 	}
 	go reapRejectedContexts(ctx, submissions, stderr)
@@ -683,13 +684,19 @@ type imageRefStore interface {
 
 type serverLister interface {
 	ListServers(ctx context.Context) ([]api.ServerInfo, error)
+	PodImages(ctx context.Context) ([]string, error)
 }
 
 // inUseImageRefs lists every image reference the platform still depends on: the
 // whitelist (disabled rows too, an admin may enable them again), every server's
-// spec, builds still running, and the images the control plane and the build
-// Jobs run. Any source failing fails the whole list, so the pruner never decides
-// on a partial view.
+// spec, the images the game pods run, builds still running, and the images the
+// control plane and the build Jobs run. Any source failing fails the whole list,
+// so the pruner never decides on a partial view.
+//
+// The pods matter for the felis image: a running server keeps the one it started
+// with across platform upgrades (operator.PodTemplateAnnotation), which after a
+// few releases is no longer among the newest tags the pruner keeps anyway, and
+// the pod needs it again whenever it is recreated.
 func inUseImageRefs(ctx context.Context, store imageRefStore, servers serverLister, static []string) ([]string, error) {
 	refs := append([]string(nil), static...)
 	images, err := store.ListImages(ctx)
@@ -706,6 +713,11 @@ func inUseImageRefs(ctx context.Context, store imageRefStore, servers serverList
 	for _, s := range srvs {
 		refs = append(refs, s.Image)
 	}
+	podImages, err := servers.PodImages(ctx)
+	if err != nil {
+		return nil, fmt.Errorf("game pods: %w", err)
+	}
+	refs = append(refs, podImages...)
 	builds, err := store.ListUnfinishedBuilds(ctx)
 	if err != nil {
 		return nil, fmt.Errorf("running builds: %w", err)

cmd/felis/api_test.go

@@ -108,9 +108,14 @@ func (f fakeRefStore) ListUnfinishedBuilds(context.Context) ([]build.Build, erro
 	return f.builds, nil
 }
 
-type fakeServers []api.ServerInfo
+type fakeServers struct {
+	list    []api.ServerInfo
+	pods    []string
+	podsErr error
+}
 
-func (f fakeServers) ListServers(context.Context) ([]api.ServerInfo, error) { return f, nil }
+func (f fakeServers) ListServers(context.Context) ([]api.ServerInfo, error) { return f.list, nil }
+func (f fakeServers) PodImages(context.Context) ([]string, error)           { return f.pods, f.podsErr }
 
 // The registry pruner deletes whatever this list does not name, so every source of
 // a reference has to be in it, and a failing source must fail the list.
@@ -120,7 +125,10 @@ func TestInUseImageRefsCoversEverySource(t *testing.T) {
 		images: []build.Image{{ImageRef: reg + "modpacks/pack:*"}, {ImageRef: reg + "felis/paper:demo"}},
 		builds: []build.Build{{ImageRef: reg + "user-uploads/sub-9:latest"}},
 	}
-	servers := fakeServers{{Name: "s1", Image: reg + "felis/paper:demo@sha256:" + fmt.Sprintf("%064d", 1)}}
+	servers := fakeServers{
+		list: []api.ServerInfo{{Name: "s1", Image: reg + "felis/paper:demo@sha256:" + fmt.Sprintf("%064d", 1)}},
+		pods: []string{reg + "felis/felis:v1.0.0"},
+	}
 	got, err := inUseImageRefs(context.Background(), store, servers, []string{reg + "felis/felis:b60"})
 	if err != nil {
 		t.Fatal(err)
@@ -129,12 +137,19 @@ func TestInUseImageRefsCoversEverySource(t *testing.T) {
 		reg + "felis/felis:b60",
 		reg + "modpacks/pack:*", reg + "felis/paper:demo",
 		reg + "felis/paper:demo@sha256:" + fmt.Sprintf("%064d", 1),
+		reg + "felis/felis:v1.0.0",
 		reg + "user-uploads/sub-9:latest",
 	}
 	if fmt.Sprint(got) != fmt.Sprint(want) {
 		t.Fatalf("refs = %v\nwant %v", got, want)
 	}
 
+	servers.podsErr = errors.New("apiserver down")
+	if _, err := inUseImageRefs(context.Background(), store, servers, nil); err == nil {
+		t.Fatal("a failing pod list produced a reference list")
+	}
+	servers.podsErr = nil
+
 	store.err = errors.New("db down")
 	if _, err := inUseImageRefs(context.Background(), store, servers, nil); err == nil {
 		t.Fatal("a failing whitelist read produced a reference list")