Files
wangjia cf2bd93c93 feat(server): agent 热重载 sing-box(SIGHUP)替代全量重启(#3)
sing-box 支持 SIGHUP 热重载(cmd_run.go:进程内 check()验证→close 旧实例→
start 新实例;坏配置保留旧实例继续跑、不断网)。原 agent 用 systemctl restart
(全量重启:断所有连接 + 坏配置让 sing-box 起不来全断)。

- Restarter 接口加 Reload;SystemdRestarter.Reload 取 MainPID(systemctl show
  -p MainPID)发 SIGHUP——agent 与 sing-box 同 pangolin 用户,免 polkit/不改 unit;
  PID 取不到或发信号失败回退 Restart
- SingBox.started:首次/进程未起走 Restart(冷启动),之后配置变更走 Reload
- noopRestarter/fakeRestarter 补 Reload;加 TestWriteAndRestartFirstColdThenReload

价值:切节点/续期/加删用户不整进程重启、不全断流;坏配置不至于打死节点。

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-19 19:36:46 +08:00

267 lines
8.0 KiB
Go

package agentd
import (
"context"
"encoding/json"
"os"
"path/filepath"
"strings"
"sync"
"testing"
"time"
agentv1 "github.com/wangjia/pangolin/server/internal/pb/agentv1"
)
// fakeRestarter counts Restart calls.
type fakeRestarter struct {
mu sync.Mutex
n int // Restart 次数(冷启动)
reloads int // Reload 次数(热重载)
}
func (f *fakeRestarter) Restart(context.Context) error {
f.mu.Lock()
f.n++
f.mu.Unlock()
return nil
}
func (f *fakeRestarter) Reload(context.Context) error {
f.mu.Lock()
f.reloads++
f.mu.Unlock()
return nil
}
func (f *fakeRestarter) count() int {
f.mu.Lock()
defer f.mu.Unlock()
return f.n
}
func (f *fakeRestarter) reloadCount() int {
f.mu.Lock()
defer f.mu.Unlock()
return f.reloads
}
func testConfig(t *testing.T) Config {
t.Helper()
dir := t.TempDir()
return Config{
ControlPlaneAddr: "bufnet",
StateDir: dir,
SingboxConfigPath: filepath.Join(dir, "singbox", "config.json"),
DeriveKey: "test-derive-key",
DebounceWindow: 20 * time.Millisecond,
Insecure: true,
}.withDefaults()
}
func sampleSnapshot(creds ...*agentv1.Credential) *agentv1.ConfigSnapshot {
return &agentv1.ConfigSnapshot{
ConfigVersion: 7,
Credentials: creds,
Reality: &agentv1.RealityInbound{
ListenPort: 11443, PrivateKey: "pk", ShortID: "deadbeef",
ServerName: "www.apple.com", HandshakeServer: "www.apple.com", HandshakePort: 443,
},
Hy2: &agentv1.Hy2Inbound{ListenPort: 443, Masquerade: "https://www.bing.com", CertPath: "/c", KeyPath: "/k"},
LastCommandID: 3,
}
}
func TestRenderContainsCredentials(t *testing.T) {
sb := NewSingBox(testConfig(t), nil)
sb.ApplyConfig(sampleSnapshot(
&agentv1.Credential{DpUUID: "aaaa", Protocol: agentv1.ProtocolBoth},
&agentv1.Credential{DpUUID: "bbbb", Protocol: agentv1.ProtocolReality},
&agentv1.Credential{DpUUID: "cccc", Protocol: agentv1.ProtocolHy2},
), true)
data, err := sb.RenderConfig()
if err != nil {
t.Fatal(err)
}
s := string(data)
// REALITY inbound: aaaa + bbbb as VLESS uuids, with flow; cccc must NOT be there.
for _, want := range []string{`"uuid": "aaaa"`, `"uuid": "bbbb"`, `"flow": "xtls-rprx-vision"`} {
if !strings.Contains(s, want) {
t.Errorf("rendered config missing %q\n%s", want, s)
}
}
if strings.Contains(s, `"uuid": "cccc"`) {
t.Errorf("hy2-only credential cccc leaked into REALITY users")
}
// Hy2 inbound: derived passwords for aaaa + cccc, never the raw dp_uuid as pw.
var cfg map[string]any
if err := json.Unmarshal(data, &cfg); err != nil {
t.Fatalf("rendered config is not valid JSON: %v", err)
}
if !strings.Contains(s, DeriveHy2Password("aaaa", "test-derive-key")) {
t.Error("hy2 password for aaaa not found")
}
if !strings.Contains(s, DeriveHy2Password("cccc", "test-derive-key")) {
t.Error("hy2 password for cccc not found")
}
}
func TestRevokeRemovesFromRender(t *testing.T) {
sb := NewSingBox(testConfig(t), nil)
sb.ApplyConfig(sampleSnapshot(&agentv1.Credential{DpUUID: "aaaa", Protocol: agentv1.ProtocolBoth}), true)
sb.Revoke("aaaa")
data, _ := sb.RenderConfig()
if strings.Contains(string(data), "aaaa") {
t.Errorf("revoked credential still present:\n%s", data)
}
if sb.Has("aaaa") {
t.Error("Has reports revoked credential as present")
}
}
func TestRotateGraceBothPresentThenOldExpires(t *testing.T) {
cfg := testConfig(t)
sb := NewSingBox(cfg, nil)
now := time.Unix(1000, 0)
sb.clock = func() time.Time { return now }
sb.ApplyConfig(sampleSnapshot(&agentv1.Credential{DpUUID: "old", Protocol: agentv1.ProtocolBoth}), true)
grace := int64(1005)
sb.Rotate("old", &Cred{DpUUID: "new", Protocol: agentv1.ProtocolBoth}, grace)
// During grace both coexist.
if !sb.Has("old") || !sb.Has("new") {
t.Fatalf("expected both old+new during grace; old=%v new=%v", sb.Has("old"), sb.Has("new"))
}
// After grace, sweep removes old, keeps new.
now = time.Unix(1006, 0)
removed := sb.sweepExpired()
if len(removed) != 1 || removed[0] != "old" {
t.Fatalf("expected old removed by sweep, got %v", removed)
}
if sb.Has("old") || !sb.Has("new") {
t.Fatalf("after grace: old should be gone, new present; old=%v new=%v", sb.Has("old"), sb.Has("new"))
}
}
func TestStateRecovery(t *testing.T) {
cfg := testConfig(t)
now := time.Unix(2000, 0)
// First manager: add creds (one already expired), persist via mutation.
sb1 := NewSingBox(cfg, nil)
sb1.clock = func() time.Time { return now }
sb1.Upsert(&Cred{DpUUID: "live", Protocol: agentv1.ProtocolBoth, ExpiresAt: 3000})
sb1.Upsert(&Cred{DpUUID: "dead", Protocol: agentv1.ProtocolBoth, ExpiresAt: 1000}) // already expired
// state.json must contain ONLY dp_uuid + expires_at — no identity fields.
raw, err := os.ReadFile(cfg.StatePath())
if err != nil {
t.Fatal(err)
}
assertNoIdentityFields(t, raw)
// Second manager (simulated kill -9 + restart) restores from disk.
sb2 := NewSingBox(cfg, nil)
sb2.clock = func() time.Time { return now }
if err := sb2.LoadState(); err != nil {
t.Fatal(err)
}
if !sb2.Has("live") {
t.Error("live credential not recovered from state.json")
}
if sb2.Has("dead") {
t.Error("expired credential should be dropped on load")
}
// Recovered creds default to BOTH + default flow until Register reconciles.
for _, c := range sb2.Snapshot() {
if c.DpUUID == "live" && (c.Protocol != agentv1.ProtocolBoth || c.Flow != DefaultFlow) {
t.Errorf("recovered cred has unexpected defaults: %+v", c)
}
}
}
func TestTTLSweepReportsExpiry(t *testing.T) {
cfg := testConfig(t)
sb := NewSingBox(cfg, nil)
now := time.Unix(5000, 0)
sb.clock = func() time.Time { return now }
var mu sync.Mutex
var reported []string
sb.OnExpire = func(dp string) { mu.Lock(); reported = append(reported, dp); mu.Unlock() }
sb.Upsert(&Cred{DpUUID: "x", Protocol: agentv1.ProtocolBoth, ExpiresAt: 4999})
sb.Upsert(&Cred{DpUUID: "y", Protocol: agentv1.ProtocolBoth, ExpiresAt: 9999})
removed := sb.sweepExpired()
if len(removed) != 1 || removed[0] != "x" {
t.Fatalf("sweep removed = %v, want [x]", removed)
}
mu.Lock()
defer mu.Unlock()
if len(reported) != 1 || reported[0] != "x" {
t.Fatalf("OnExpire reported = %v, want [x]", reported)
}
}
func TestDebounceCoalescesRestarts(t *testing.T) {
cfg := testConfig(t)
cfg.DebounceWindow = 40 * time.Millisecond
fr := &fakeRestarter{}
sb := NewSingBox(cfg, fr)
sb.ApplyConfig(sampleSnapshot(), true) // set inbounds so render succeeds
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
go sb.Run(ctx)
// Burst of changes within one debounce window.
for i := 0; i < 5; i++ {
sb.Upsert(&Cred{DpUUID: string(rune('a' + i)), Protocol: agentv1.ProtocolBoth})
}
time.Sleep(150 * time.Millisecond)
if n := fr.count(); n != 1 {
t.Fatalf("expected 1 coalesced restart, got %d", n)
}
}
// assertNoIdentityFields fails if the blob contains any user-identity token.
func assertNoIdentityFields(t *testing.T, blob []byte) {
t.Helper()
lower := strings.ToLower(string(blob))
for _, banned := range []string{"email", "user_id", "userid", "phone", "device_id", "deviceid", "account", "destination", "dns"} {
if strings.Contains(lower, banned) {
t.Errorf("forbidden identity field %q found in persisted/rendered data:\n%s", banned, blob)
}
}
}
// 首次渲染走冷启动(Restart),之后的配置变更走 SIGHUP 热重载(Reload)。
func TestWriteAndRestartFirstColdThenReload(t *testing.T) {
fr := &fakeRestarter{}
sb := NewSingBox(testConfig(t), fr)
sb.ApplyConfig(sampleSnapshot(), true) // 设置 inbounds 使渲染成功
ctx := context.Background()
if err := sb.writeAndRestart(ctx); err != nil {
t.Fatalf("first writeAndRestart: %v", err)
}
if r, rl := fr.count(), fr.reloadCount(); r != 1 || rl != 0 {
t.Fatalf("首次后 restarts=%d reloads=%d, 期望 1/0", r, rl)
}
sb.Upsert(&Cred{DpUUID: "z", Protocol: agentv1.ProtocolBoth})
if err := sb.writeAndRestart(ctx); err != nil {
t.Fatalf("second writeAndRestart: %v", err)
}
if r, rl := fr.count(), fr.reloadCount(); r != 1 || rl != 1 {
t.Fatalf("变更后 restarts=%d reloads=%d, 期望 1/1(不再 Restart,走 Reload)", r, rl)
}
}