feat(converge): 报告缺 RCON 的老用户服务器，-user-rcon 按新建时的配置补上

cmd/felis/converge.go

@@ -11,12 +11,14 @@ import (
 
 	"felis.lolicon.best/internal/apis/felis/v1alpha1"
 	"felis.lolicon.best/internal/config"
+	"felis.lolicon.best/internal/naming"
 	"felis.lolicon.best/internal/platform"
 	"sigs.k8s.io/controller-runtime/pkg/client"
 )
 
 // cmdConverge is the explicit convergence pass over already-installed system
-// servers (#1), plus the idle-stop default for user servers that predate it. Provisioning is create-if-absent, so a field the desired spec
+// servers (#1), plus the idle-stop default for user servers that predate it, and
+// with -user-rcon their RCON block (#3). Provisioning is create-if-absent, so a field the desired spec
 // gained after an install (spec.rcon, spec.startup.healthHTTPPort, a derived env
 // key) never reaches the existing CR — and nothing says so. This command fills
 // exactly those zero-value fields; see convergeSystemServers for the full contract
@@ -29,6 +31,7 @@ func cmdConverge(args []string, stdout, stderr io.Writer) int {
 	fs := flag.NewFlagSet("converge", flag.ContinueOnError)
 	fs.SetOutput(stderr)
 	cfgPath := fs.String("config", defaultSetupConfigPath, "path to felis.toml")
+	userRcon := fs.Bool("user-rcon", false, "also turn RCON on for user servers created before it was the default")
 	if err := fs.Parse(args); err != nil {
 		if errors.Is(err, flag.ErrHelp) {
 			return 0
@@ -59,6 +62,7 @@ func cmdConverge(args []string, stdout, stderr io.Writer) int {
 		defaultPanelHostname(cfg.Server.RootDomain, cfg.Auth.PanelHostname))
 
 	outcomes = append(outcomes, convergeUserServerIdle(context.Background(), cl, cfg.K8s.Namespace)...)
+	outcomes = append(outcomes, convergeUserServerRcon(context.Background(), cl, cfg.K8s.Namespace, *userRcon)...)
 
 	fmt.Fprintln(stdout, "felis converge: filling fields an installed server predates (operator-set values are never overwritten):")
 	exit := 0
@@ -113,3 +117,51 @@ func convergeUserServerIdle(ctx context.Context, cl client.Client, namespace str
 	}
 	return out
 }
+
+// convergeUserServerRcon handles user servers created before RCON was part of every
+// new server (694e3cb): spec.rcon entirely unset. Such a server has a dead console,
+// reports nobody online, and never idles out, because all three ride RCON.
+//
+// Only with fill does it turn RCON on, with the same block CreateServer writes
+// today; without it each such server gets a line saying so. The fill is opt-in
+// because the operator gates readiness on the RCON probe: a server whose image does
+// not open the listener RCON_PASSWORD asks for would sit in Starting until it is
+// marked Failed. Felis's own paper and lobby images open it; an image a user brought
+// may not, and only the operator running this can tell. A server that already
+// carries any RCON setting (on or off) is left alone and produces no line.
+func convergeUserServerRcon(ctx context.Context, cl client.Client, namespace string, fill bool) []systemServerOutcome {
+	var list v1alpha1.MinecraftServerList
+	if err := cl.List(ctx, &list, client.InNamespace(namespace)); err != nil {
+		return []systemServerOutcome{{name: "user servers", err: fmt.Errorf("list servers: %w", err)}}
+	}
+	var out []systemServerOutcome
+	for i := range list.Items {
+		ms := &list.Items[i]
+		if ms.Labels[v1alpha1.LabelSystemRole] != "" || ms.Spec.Rcon != (v1alpha1.RconSpec{}) {
+			continue
+		}
+		if !fill {
+			out = append(out, systemServerOutcome{name: ms.Name, available: true,
+				skipped: "no RCON (console, online count and idle stop are off); once its image serves RCON, sudo felis converge -user-rcon turns it on"})
+			continue
+		}
+		changed, err := patchOnConflictRetry(ctx, cl, ms, func() bool {
+			if ms.Spec.Rcon != (v1alpha1.RconSpec{}) {
+				return false
+			}
+			ms.Spec.Rcon = v1alpha1.RconSpec{
+				Enabled:   true,
+				SecretRef: v1alpha1.SecretKeyRef{Name: naming.RconSecretName(ms.Name), Key: naming.RconSecretKey},
+			}
+			return true
+		})
+		if err != nil {
+			out = append(out, systemServerOutcome{name: ms.Name, err: fmt.Errorf("converge %s: %w", ms.Name, err)})
+			continue
+		}
+		if changed {
+			out = append(out, systemServerOutcome{name: ms.Name, available: true, updated: true, changes: []string{"spec.rcon"}})
+		}
+	}
+	return out
+}

cmd/felis/converge_test.go

@@ -229,3 +229,67 @@ func TestConvergeUserServerIdle(t *testing.T) {
 		t.Fatalf("second pass = %+v, want nothing to do", again)
 	}
 }
+
+// TestConvergeUserServerRcon reports a user server with no RCON block at all and
+// fills it only when asked, with the block CreateServer writes. RCON turned off on
+// purpose, a server with its own secret, and a system server stay as they are and
+// produce no line.
+func TestConvergeUserServerRcon(t *testing.T) {
+	scheme := newSystemServerScheme(t)
+	ctx := context.Background()
+	mk := func(name string, rcon v1alpha1.RconSpec, role string) *v1alpha1.MinecraftServer {
+		ms := &v1alpha1.MinecraftServer{}
+		ms.Name, ms.Namespace = name, "minecraft"
+		ms.Spec.Rcon = rcon
+		if role != "" {
+			ms.Labels = map[string]string{v1alpha1.LabelSystemRole: role}
+		}
+		return ms
+	}
+	own := v1alpha1.RconSpec{Enabled: true, Port: 25580, SecretRef: v1alpha1.SecretKeyRef{Name: "own", Key: "pw"}}
+	off := v1alpha1.RconSpec{SecretRef: v1alpha1.SecretKeyRef{Name: "rcon-off", Key: naming.RconSecretKey}}
+	cl := fake.NewClientBuilder().WithScheme(scheme).WithObjects(
+		mk("demo", v1alpha1.RconSpec{}, ""),
+		mk("off", off, ""),
+		mk("own", own, ""),
+		mk(naming.SystemLobbyServer, v1alpha1.RconSpec{}, naming.SystemLobbyServer),
+	).Build()
+	get := func(name string) v1alpha1.RconSpec {
+		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)
+		}
+		return ms.Spec.Rcon
+	}
+
+	report := convergeUserServerRcon(ctx, cl, "minecraft", false)
+	if len(report) != 1 || report[0].name != "demo" || report[0].updated || report[0].err != nil ||
+		!strings.Contains(report[0].skipped, "-user-rcon") {
+		t.Fatalf("report = %+v, want one skipped line for demo naming -user-rcon", report)
+	}
+	if got := get("demo"); got != (v1alpha1.RconSpec{}) {
+		t.Fatalf("the report-only pass wrote demo's rcon: %+v", got)
+	}
+
+	filled := convergeUserServerRcon(ctx, cl, "minecraft", true)
+	if len(filled) != 1 || filled[0].name != "demo" || !filled[0].updated || filled[0].err != nil {
+		t.Fatalf("fill = %+v, want exactly one update for demo", filled)
+	}
+	want := map[string]v1alpha1.RconSpec{
+		"demo": {Enabled: true, SecretRef: v1alpha1.SecretKeyRef{
+			Name: naming.RconSecretName("demo"), Key: naming.RconSecretKey}},
+		"off":                    off,
+		"own":                    own,
+		naming.SystemLobbyServer: {},
+	}
+	for name, rcon := range want {
+		if got := get(name); got != rcon {
+			t.Errorf("%s rcon = %+v, want %+v", name, got, rcon)
+		}
+	}
+	for _, fill := range []bool{false, true} {
+		if again := convergeUserServerRcon(ctx, cl, "minecraft", fill); len(again) != 0 {
+			t.Fatalf("second pass (fill=%v) = %+v, want nothing to do", fill, again)
+		}
+	}
+}

docs/troubleshooting.md

@@ -1225,6 +1225,24 @@ empty. A server whose idle stop an admin turned off keeps a duration on its
 spec (`autoStopEnabled: false`, `emptySecondsBeforeStop` set), so converge
 leaves it off; only servers it actually filled get a line.
 
+A **user** server created before every new server got RCON (`spec.rcon` wholly
+unset) has a dead console, always shows 0 online, and never idles out, since all
+three ride RCON. Plain `felis converge` lists each such server and changes
+nothing; `-user-rcon` turns RCON on for them with the block a server created
+today gets (the `<name>-rcon` Secret the operator provisions):
+
+```
+sudo felis converge -user-rcon
+kubectl -n minecraft get minecraftserver -o custom-columns=NAME:.metadata.name,RCON:.spec.rcon.enabled
+```
+
+It is opt-in for the reason above: the new start waits on the RCON probe, and a
+server whose image does not open the listener that `RCON_PASSWORD` asks for
+stays in `Starting` until it is marked `Failed`. Felis's own paper and lobby
+images open it; check a server running an image a user brought before filling
+it. A server whose RCON someone set, on or off, is never touched. The change
+applies at the server's next start.
+
 ---
 
 ## 13. World PVC survives after I deleted the MinecraftServer