Compare commits

..
17 changed files with 405 additions and 367 deletions
@@ -0,0 +1,196 @@
package main
import (
"bytes"
"context"
"encoding/json"
"io"
"net/http"
"net/http/cookiejar"
"strings"
"testing"
"time"
"git.asio.asia/nixevol/NixMsg/internal/config"
)
// TestUplinkDisableFatalAndRevoked 验证停用在线端收到 fatal,已推送投递收到 revoked。
func TestUplinkDisableFatalAndRevoked(t *testing.T) {
dataDir := t.TempDir()
cfgPath := writeTestConfig(t, dataDir)
initAdminForTest(t, dataDir)
enableRegistration(t, dataDir, "uplink-code")
cfg, err := config.Load(cfgPath)
if err != nil {
t.Fatal(err)
}
if vErr := cfg.Validate(); vErr != nil {
t.Fatal(vErr)
}
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
errCh := make(chan error, 1)
go func() { errCh <- runServe(ctx, cfg) }()
defer func() {
cancel()
select {
case err := <-errCh:
if err != nil {
t.Errorf("serve exit: %v", err)
}
case <-time.After(15 * time.Second):
t.Error("serve did not stop")
}
}()
addr := waitListenAddr(t, dataDir, 15*time.Second)
base := "http://" + addr
registerEP(t, base, "alice", "password12", "Alice")
registerEP(t, base, "bob", "password12", "Bob")
alice := mqttSessionLogin(t, base, "alice", "password12")
defer alice.Close()
bob := mqttSessionLogin(t, base, "bob", "password12")
defer bob.Close()
delay0 := int64(0)
sendResp := alice.Request(t, map[string]any{
"v": 1, "type": "send", "rid": "s1", "id": "dm-fatal-1",
"to": map[string]any{"kind": "endpoint", "id": "bob"},
"body": map[string]any{"enc": "utf8", "data": "to-void"},
"delay_ms": delay0,
})
if !sendResp.OK {
t.Fatalf("send: %+v", sendResp)
}
msg := bob.WaitType(t, "msg", 8*time.Second)
if msg["id"] != "dm-fatal-1" {
t.Fatalf("bob msg=%v", msg)
}
admin := adminHTTPClient(t, base)
disableEP(t, admin, base, "bob")
fatal := bob.WaitType(t, "fatal", 8*time.Second)
if fatal["reason"] != "disabled" {
t.Fatalf("fatal=%v", fatal)
}
revoked := bob.WaitType(t, "revoked", 8*time.Second)
if revoked["id"] != "dm-fatal-1" || revoked["reason"] != "endpoint_disabled" {
t.Fatalf("revoked=%v", revoked)
}
}
// TestUplinkResetPasswordFatal 验证重置登录密码后在线端收到 fatal(password_reset)。
func TestUplinkResetPasswordFatal(t *testing.T) {
dataDir := t.TempDir()
cfgPath := writeTestConfig(t, dataDir)
initAdminForTest(t, dataDir)
enableRegistration(t, dataDir, "uplink-code")
cfg, err := config.Load(cfgPath)
if err != nil {
t.Fatal(err)
}
if vErr := cfg.Validate(); vErr != nil {
t.Fatal(vErr)
}
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
errCh := make(chan error, 1)
go func() { errCh <- runServe(ctx, cfg) }()
defer func() {
cancel()
select {
case err := <-errCh:
if err != nil {
t.Errorf("serve exit: %v", err)
}
case <-time.After(15 * time.Second):
t.Error("serve did not stop")
}
}()
addr := waitListenAddr(t, dataDir, 15*time.Second)
base := "http://" + addr
registerEP(t, base, "carol", "password12", "Carol")
carol := mqttSessionLogin(t, base, "carol", "password12")
defer carol.Close()
admin := adminHTTPClient(t, base)
resetLoginPassword(t, admin, base, "carol", "password99xx")
fatal := carol.WaitType(t, "fatal", 8*time.Second)
if fatal["reason"] != "password_reset" {
t.Fatalf("fatal=%v", fatal)
}
}
func adminHTTPClient(t *testing.T, base string) *http.Client {
t.Helper()
jar, err := cookiejar.New(nil)
if err != nil {
t.Fatal(err)
}
client := &http.Client{Jar: jar, Timeout: 10 * time.Second}
loginBody, _ := json.Marshal(map[string]string{
"username": "admin",
"password": "test-admin-password-xx",
})
resp, err := client.Post(base+"/api/admin/login", "application/json", bytes.NewReader(loginBody))
if err != nil {
t.Fatal(err)
}
raw, _ := io.ReadAll(resp.Body)
_ = resp.Body.Close()
if resp.StatusCode != http.StatusOK {
t.Fatalf("admin login: %d %s", resp.StatusCode, raw)
}
return client
}
func disableEP(t *testing.T, client *http.Client, base, id string) {
t.Helper()
req, err := http.NewRequest(http.MethodPatch, base+"/api/admin/endpoints/"+id,
strings.NewReader(`{"enabled":false}`))
if err != nil {
t.Fatal(err)
}
req.Header.Set("Content-Type", "application/json")
req.Header.Set("X-Nixmsg-Request", "1")
resp, err := client.Do(req)
if err != nil {
t.Fatal(err)
}
raw, _ := io.ReadAll(resp.Body)
_ = resp.Body.Close()
if resp.StatusCode != http.StatusOK {
t.Fatalf("disable %s: %d %s", id, resp.StatusCode, raw)
}
}
func resetLoginPassword(t *testing.T, client *http.Client, base, id, password string) {
t.Helper()
body := `{"login_password":"` + password + `"}`
req, err := http.NewRequest(http.MethodPost, base+"/api/admin/endpoints/"+id+"/reset-login-password",
strings.NewReader(body))
if err != nil {
t.Fatal(err)
}
req.Header.Set("Content-Type", "application/json")
req.Header.Set("X-Nixmsg-Request", "1")
resp, err := client.Do(req)
if err != nil {
t.Fatal(err)
}
raw, _ := io.ReadAll(resp.Body)
_ = resp.Body.Close()
if resp.StatusCode != http.StatusOK {
t.Fatalf("reset password %s: %d %s", id, resp.StatusCode, raw)
}
}
+36 -17
View File
@@ -71,10 +71,6 @@ func runServe(ctx context.Context, cfg config.Config) error {
loginLocks := auth.NewLoginLocks()
memConns := message.NewMemoryConns()
msgLim := message.LimitsFromFullConfig(cfg)
metricsReg := metrics.New()
db.Queue.OnBatchCommit = func(d time.Duration) {
metricsReg.WriteCommitSeconds.Observe(d.Seconds())
}
login := broker.NewLogin(broker.LoginOptions{
DB: db,
@@ -88,13 +84,11 @@ func runServe(ctx context.Context, cfg config.Config) error {
msgApp := message.New(db, msgLim, hashPool,
message.WithLocks(loginLocks),
message.WithConnRegistry(memConns),
message.WithMetrics(metricsReg),
)
uplink := &appUplink{
msg: msgApp,
conns: memConns,
log: slog.Default(),
metrics: metricsReg,
msg: msgApp,
conns: memConns,
log: slog.Default(),
}
sess := broker.NewSession(broker.SessionOptions{
Login: login,
@@ -115,7 +109,6 @@ func runServe(ctx context.Context, cfg config.Config) error {
Authenticator: login,
Uplink: sess,
Logger: slog.Default(),
Metrics: metricsReg,
OnPublishDropped: func(dropCtx context.Context, endpointID string, connID port.ConnID, payload []byte) {
if dropErr := msgApp.OnPublishDropped(dropCtx, endpointID, connID, payload); dropErr != nil {
slog.Error("on publish dropped", "endpoint", endpointID, "err", dropErr)
@@ -132,7 +125,6 @@ func runServe(ctx context.Context, cfg config.Config) error {
message.WithLocks(loginLocks),
message.WithConnRegistry(memConns),
message.WithDownlink(brk),
message.WithMetrics(metricsReg),
)
uplink.msg = msgApp
uplink.down = brk
@@ -153,6 +145,7 @@ func runServe(ctx context.Context, cfg config.Config) error {
MaxScheduleSeconds: int64(cfg.Limits.MaxScheduleSeconds),
Logger: slog.Default(),
ConnControl: brk,
Downlink: brk,
})
uplink.identity = idApp
@@ -181,6 +174,7 @@ func runServe(ctx context.Context, cfg config.Config) error {
Groups: groupApp,
Config: cfg,
Version: Version,
// Kick:只断开,令牌不变,SDK 重连(PRD 踢下线)。
KickEndpoint: func(kickCtx context.Context, endpointID string) (bool, error) {
if _, found := brk.ConnInfoOf(endpointID); !found {
return false, nil
@@ -190,8 +184,37 @@ func runServe(ctx context.Context, cfg config.Config) error {
}
return true, nil
},
// 停用/删除/重置:先 fatal 再断开(DEVELOPMENT 6.8)。
DisableKick: func(kickCtx context.Context, endpointID string) (bool, error) {
if _, found := brk.ConnInfoOf(endpointID); !found {
return false, nil
}
if err := sess.Disable(kickCtx, endpointID); err != nil {
return false, err
}
return true, nil
},
DeleteKick: func(kickCtx context.Context, endpointID string) (bool, error) {
if _, found := brk.ConnInfoOf(endpointID); !found {
return false, nil
}
if err := sess.Deleted(kickCtx, endpointID); err != nil {
return false, err
}
return true, nil
},
PasswordResetKick: func(kickCtx context.Context, endpointID string) (bool, error) {
if _, found := brk.ConnInfoOf(endpointID); !found {
return false, nil
}
if resetErr := sess.ResetPassword(kickCtx, endpointID); resetErr != nil {
return false, resetErr
}
return true, nil
},
})
metricsReg := metrics.New()
buildHandlers := func(proxies *listener.ProxySet) listener.Handlers {
return listener.Handlers{
MQTT: brk.WSHandler(proxies),
@@ -273,7 +296,7 @@ func runServe(ctx context.Context, cfg config.Config) error {
loopCtx, loopCancel := context.WithCancel(ctx)
defer loopCancel()
go messageLoops(loopCtx, msgApp, memConns, db, hashPool, metricsReg)
go messageLoops(loopCtx, msgApp, memConns)
<-ctx.Done()
loopCancel()
@@ -286,7 +309,7 @@ func runServe(ctx context.Context, cfg config.Config) error {
return nil
}
func messageLoops(ctx context.Context, msgApp *message.App, conns *message.MemoryConns, db *store.DB, hashPool auth.HashPool, met *metrics.Registry) {
func messageLoops(ctx context.Context, msgApp *message.App, conns *message.MemoryConns) {
t := time.NewTicker(time.Second)
defer t.Stop()
for {
@@ -306,10 +329,6 @@ func messageLoops(ctx context.Context, msgApp *message.App, conns *message.Memor
if err := msgApp.CleanupOnce(ctx, nowMs); err != nil {
slog.Error("cleanup once", "err", err)
}
if err := metrics.SampleStoreGauges(ctx, met, db.Read); err != nil {
slog.Debug("sample store gauges", "err", err)
}
metrics.SampleQueues(met, db.Queue.Len(), hashPool.QueueLen())
}
}
}
-5
View File
@@ -11,7 +11,6 @@ import (
"git.asio.asia/nixevol/NixMsg/internal/app/message"
"git.asio.asia/nixevol/NixMsg/internal/app/port"
"git.asio.asia/nixevol/NixMsg/internal/app/presence"
"git.asio.asia/nixevol/NixMsg/internal/metrics"
"git.asio.asia/nixevol/NixMsg/internal/protocol"
)
@@ -24,7 +23,6 @@ type appUplink struct {
conns *message.MemoryConns
down port.Downlink
log *slog.Logger
metrics *metrics.Registry
}
func (u *appUplink) OnSessionEstablished(ctx context.Context, conn port.ConnInfo) error {
@@ -257,9 +255,6 @@ func (u *appUplink) replyErr(ctx context.Context, conn port.ConnInfo, rid, code,
if rid == "" {
rid = "0"
}
if u.metrics != nil && code != "" {
u.metrics.ErrorsTotal.WithLabelValues(code).Inc()
}
resp := protocol.Resp{
V: protocol.Version,
Type: protocol.TypeResp,
+7 -7
View File
@@ -1093,11 +1093,11 @@
- 备选方案:broker 层对 Inline 发布做无锁队列。
- 影响:`group_event` 可能略晚于 `resp` 到达;业务结果仍以 `resp` 为准。
### fix-issue-5
### fix-issue-4
1. **指标在真实事件点打点,不新造名字**
- 原条款:PRD F22 / DEVELOPMENT 4.3 / issue #5;DEVIATIONS P4 已定名但从未接线。
- 实际做法:`nixmsg_connections{transport}` 在 broker `OnSessionEstablished`/`OnDisconnect` 末尾 Inc/Dec;`endpoints`/`deliveries_pending`/`messages_scheduled` 与写队列、哈希排队在 `messageLoops` 每秒按库/队列真实长度采样;`dispatch_to_push`/`ack` 直方图在成功推送与确认路径 Observe;写批提交耗时经 `store.Queue.OnBatchCommit`;`errors_total` 仅在上行 `replyErr` 时按错误码递增。门禁不变。
- 原因:空指标等于监控未交付;采样避免在每条写路径上改大段分发逻辑,并减小与 #3/#6 的合并面。
- 备选方案:全部改为纯事件加减(pending 等需在每处状态迁移维护计数)。
- 影响:仪表盘按既有名字即可看在线连接与待投递;无对应事件时计数保持 0,不做假数。
1. **接线补齐 Downlink 与停用/删除/重置密码 fatal**
- 原条款:DEVELOPMENT 6.8 / 7.6:停用、删除、重置密码先发 `fatal` 再断开;已推送作废投递尽力发 `revoked`。
- 实际做法:`serve` 给 `identity.New` 注入 `Downlink: brk`(作废后 `publishRevokes`);`DisableKick`/`DeleteKick`/`PasswordResetKick` 分别接到 `Session.Disable`/`Deleted`/`ResetPassword`;`KickEndpoint` 仍只 `Kick`。Identity 在未接 Kick 钩子时仍可用 `ConnControl` 异步断开兜底。
- 原因:原先 Downlink 未注入导致 revoked 丢失;管理路径只 `Kick`/`Disconnect` 不发 fatal。
- 备选方案:仅在 identity 内 `PublishDown(fatal)` 再断开;联调中该路径不如 Session.fatalKick 稳,故生产致命踢线统一走 Session。
- 影响:管理「踢下线」语义不变;SDK 可按 fatal 停止重连;接收方能收到已推送消息的 revoked。
+32 -5
View File
@@ -102,6 +102,33 @@ func (h *Handler) kickEndpoint(ctx context.Context, id string) (bool, error) {
return h.kick(ctx, id)
}
func (h *Handler) passwordResetKick(ctx context.Context, id string) (bool, error) {
if h.resetKick != nil {
return h.resetKick(ctx, id)
}
return h.kickEndpoint(ctx, id)
}
func (h *Handler) afterDisableKick(ctx context.Context, id string) {
if h.disableKick != nil {
_, _ = h.disableKick(ctx, id)
return
}
if h.identity == nil {
_, _ = h.kickEndpoint(ctx, id)
}
}
func (h *Handler) afterDeleteKick(ctx context.Context, id string) {
if h.deleteKick != nil {
_, _ = h.deleteKick(ctx, id)
return
}
if h.identity == nil {
_, _ = h.kickEndpoint(ctx, id)
}
}
func (h *Handler) handleEndpointList(w http.ResponseWriter, r *http.Request) {
q := r.URL.Query()
limit := defaultListLimit
@@ -350,7 +377,7 @@ func (h *Handler) handleEndpointPatch(w http.ResponseWriter, r *http.Request) {
return
}
if !*req.Enabled {
_, _ = h.kickEndpoint(r.Context(), id)
h.afterDisableKick(r.Context(), id)
}
} else if req.Enabled != nil && !*req.Enabled && wasEnabled {
_, _ = h.kickEndpoint(r.Context(), id)
@@ -381,7 +408,7 @@ func (h *Handler) handleEndpointDelete(w http.ResponseWriter, r *http.Request) {
httpx.WriteError(w, http.StatusNotFound, "not_found", "端不存在")
return
}
_, _ = h.kickEndpoint(r.Context(), id)
h.afterDeleteKick(r.Context(), id)
h.audit(actorString(p), "endpoint_delete", id, "ok", ip)
httpx.WriteOK(w, map[string]any{})
}
@@ -416,14 +443,14 @@ func (h *Handler) handleEndpointBatch(w http.ResponseWriter, r *http.Request) {
case "disable":
found, opErr = h.setEndpointEnabled(r.Context(), id, false)
if found && opErr == nil {
_, _ = h.kickEndpoint(r.Context(), id)
h.afterDisableKick(r.Context(), id)
}
case "enable":
found, opErr = h.setEndpointEnabled(r.Context(), id, true)
case "delete":
found, opErr = h.deleteEndpointBasic(r.Context(), id)
if found && opErr == nil {
_, _ = h.kickEndpoint(r.Context(), id)
h.afterDeleteKick(r.Context(), id)
}
}
if opErr != nil {
@@ -512,7 +539,7 @@ func (h *Handler) handleEndpointResetLoginPassword(w http.ResponseWriter, r *htt
httpx.WriteError(w, http.StatusNotFound, "not_found", "端不存在")
return
}
_, _ = h.kickEndpoint(r.Context(), id)
_, _ = h.passwordResetKick(r.Context(), id)
h.audit(actorString(p), "endpoint_reset_login_password", id, "ok", ip)
httpx.WriteOK(w, map[string]any{loginPasswordOnceKey: pw})
}
+42 -30
View File
@@ -41,6 +41,12 @@ type Deps struct {
SecureCookies bool
// KickEndpoint 踢下线钩子(只断开连接);nil 时踢线为 no-op。
KickEndpoint EndpointKickFunc
// PasswordResetKick 重置登录密码后踢线(应发 fatal);nil 时回退 KickEndpoint。
PasswordResetKick EndpointKickFunc
// DisableKick 停用后踢线(应发 fatal(disabled));nil 且已注入 Identity 时不再 Kick。
DisableKick EndpointKickFunc
// DeleteKick 删除后踢线(应发 fatal(deleted));nil 且已注入 Identity 时不再 Kick。
DeleteKick EndpointKickFunc
// Identity 端停用/启用/删除级联(I5);nil 时回退为仅改 enabled/删行。
Identity identity.Service
@@ -54,20 +60,23 @@ type Deps struct {
// Handler 是可挂载的管理接口(路由前缀 /api/admin/)。
type Handler struct {
db *store.DB
hash auth.HashPool
tokens auth.APITokens
locks auth.LoginLocks
log *slog.Logger
trusted []*net.IPNet
ttl time.Duration
forceSec bool
kick EndpointKickFunc
identity identity.Service
groups group.Service
cfg config.Config
version string
startedAt time.Time
db *store.DB
hash auth.HashPool
tokens auth.APITokens
locks auth.LoginLocks
log *slog.Logger
trusted []*net.IPNet
ttl time.Duration
forceSec bool
kick EndpointKickFunc
resetKick EndpointKickFunc
disableKick EndpointKickFunc
deleteKick EndpointKickFunc
identity identity.Service
groups group.Service
cfg config.Config
version string
startedAt time.Time
mux *http.ServeMux
@@ -96,22 +105,25 @@ func New(d Deps) *Handler {
ver = "dev"
}
h := &Handler{
db: d.DB,
hash: d.Hash,
tokens: d.Tokens,
locks: d.Locks,
log: d.Logger,
trusted: d.TrustedProxies,
ttl: ttl,
forceSec: d.SecureCookies,
kick: d.KickEndpoint,
identity: d.Identity,
groups: d.Groups,
cfg: cfg,
version: ver,
startedAt: time.Now(),
mux: http.NewServeMux(),
lastUsed: make(map[string]time.Time),
db: d.DB,
hash: d.Hash,
tokens: d.Tokens,
locks: d.Locks,
log: d.Logger,
trusted: d.TrustedProxies,
ttl: ttl,
forceSec: d.SecureCookies,
kick: d.KickEndpoint,
resetKick: d.PasswordResetKick,
disableKick: d.DisableKick,
deleteKick: d.DeleteKick,
identity: d.Identity,
groups: d.Groups,
cfg: cfg,
version: ver,
startedAt: time.Now(),
mux: http.NewServeMux(),
lastUsed: make(map[string]time.Time),
}
h.routes()
return h
+10 -1
View File
@@ -5,6 +5,7 @@ import (
"context"
"database/sql"
"errors"
"time"
"git.asio.asia/nixevol/NixMsg/internal/app/port"
"git.asio.asia/nixevol/NixMsg/internal/protocol"
@@ -22,6 +23,9 @@ const (
eventLeft = "left"
eventMemberRemoved = "member_removed"
eventDissolved = "dissolved"
// kickFlushDelay 给接线方 Session.Disable/Deleted 留出发 fatal 的窗口。
kickFlushDelay = 20 * time.Millisecond
)
type revokeItem struct {
@@ -124,8 +128,13 @@ WHERE id = ?`, endpointID); e != nil {
a.publishRevokes(ctx, revokes)
a.publishGroupEvents(ctx, notifies)
// fatal+断开由 admin DisableKick/DeleteKick(Session.Disable/Deleted)完成。
// 未接 Kick 钩子的单元测试仍可用 ConnControl 兜底断开。
if a.connCtrl != nil {
_ = a.connCtrl.Disconnect(ctx, endpointID, "", port.DisconnectFatal)
go func() {
time.Sleep(kickFlushDelay)
_ = a.connCtrl.Disconnect(context.Background(), endpointID, "", port.DisconnectFatal)
}()
}
return nil
}
+77
View File
@@ -3,6 +3,7 @@ package identity_test
import (
"context"
"database/sql"
"encoding/json"
"io"
"net/http"
"net/http/cookiejar"
@@ -64,6 +65,82 @@ VALUES(?,?,?,?,0,0,1,?,?)`, id, id, "stub$login", nil, 1_700_000_000_000, 1_700_
}
}
func TestDisableEmitsRevokedForPushed(t *testing.T) {
t.Parallel()
dir := t.TempDir()
db, err := store.Open(filepath.Join(dir, "data"), "FULL")
if err != nil {
t.Fatal(err)
}
t.Cleanup(func() { _ = db.Close() })
fixed := time.UnixMilli(1_700_000_000_000)
down := &message.RecordingDownlink{}
ctrl := &port.StubConnControl{}
idApp := identity.New(identity.Config{
DB: db,
Hash: auth.NewStubHashPool(),
Locks: auth.NewStubLoginLocks(),
Sessions: auth.NewSessionTokens(),
MaxScheduleSeconds: int64(config.Default().Limits.MaxScheduleSeconds),
Now: func() time.Time { return fixed },
ConnControl: ctrl,
Downlink: down,
})
ctx := context.Background()
insertEPFull(t, db, "alice")
insertEPFull(t, db, "bob")
err = db.Queue.Do(ctx, func(tx *sql.Tx) error {
res, e := tx.Exec(`
INSERT INTO messages(
id, sender_id, dest_kind, dest_id, meta, content_type, body_enc,
send_at, keep, ttl_seconds, receipt, state, reason, created_at)
VALUES('pushed-1','alice','endpoint','bob','{}','text/plain','utf8',?,1,0,0,'dispatched','',?)`,
fixed.UnixMilli(), fixed.UnixMilli())
if e != nil {
return e
}
seq, _ := res.LastInsertId()
_, e = tx.Exec(`
INSERT INTO deliveries(seq, endpoint_id, send_at, keep, state, reason, updated_at, pushed_at, pushed_conn)
VALUES(?,?,?,1,'pending','',?,?,?)`,
seq, "bob", fixed.UnixMilli(), fixed.UnixMilli(), fixed.UnixMilli(), "c-bob")
return e
})
if err != nil {
t.Fatal(err)
}
if err := idApp.Disable(ctx, "bob"); err != nil {
t.Fatal(err)
}
if down.FilterType(protocol.TypeRevoked) != 1 {
t.Fatalf("want 1 revoked, got snapshots=%v", down.Snapshots())
}
p := down.Snapshots()[0]
var head struct {
Type string `json:"type"`
Reason string `json:"reason"`
ID string `json:"id"`
}
_ = json.Unmarshal(p.Payload, &head)
if head.Type != protocol.TypeRevoked || head.Reason != "endpoint_disabled" || head.ID != "pushed-1" {
t.Fatalf("revoked=%+v", head)
}
if p.EndpointID != "bob" || p.QoS != 1 {
t.Fatalf("publish=%+v", p)
}
deadline := time.Now().Add(2 * time.Second)
for time.Now().Before(deadline) {
if len(ctrl.Calls) == 1 && ctrl.Calls[0] == "bob" {
return
}
time.Sleep(5 * time.Millisecond)
}
t.Fatalf("disconnect calls=%v", ctrl.Calls)
}
func TestF01DisableVoidsScheduledAndRejectsNew(t *testing.T) {
t.Parallel()
idApp, msgApp, db := openLifecycle(t)
-13
View File
@@ -17,8 +17,6 @@ func (a *App) Ack(ctx context.Context, endpointID string, req *protocol.Ack) (Ac
nowMs := a.now().UnixMilli()
var out AckResult
var seq int64
var ackLatencySec float64
var observeAck bool
err := a.db.Queue.Do(ctx, func(tx *sql.Tx) error {
err := tx.QueryRow(`SELECT seq FROM messages WHERE sender_id = ? AND id = ?`, req.From, req.ID).Scan(&seq)
if err == sql.ErrNoRows {
@@ -27,10 +25,6 @@ func (a *App) Ack(ctx context.Context, endpointID string, req *protocol.Ack) (Ac
if err != nil {
return err
}
var pushedAt sql.NullInt64
_ = tx.QueryRow(`
SELECT pushed_at FROM deliveries
WHERE seq = ? AND endpoint_id = ? AND state = 'pending'`, seq, endpointID).Scan(&pushedAt)
res, err := tx.Exec(`
UPDATE deliveries SET state = ?, reason = '', pushed_conn = NULL, updated_at = ?
WHERE seq = ? AND endpoint_id = ? AND state = 'pending'`,
@@ -41,10 +35,6 @@ WHERE seq = ? AND endpoint_id = ? AND state = 'pending'`,
aff, _ := res.RowsAffected()
if aff > 0 {
out.Result = DeliveryAccepted
if pushedAt.Valid && pushedAt.Int64 > 0 && nowMs >= pushedAt.Int64 {
ackLatencySec = float64(nowMs-pushedAt.Int64) / 1000.0
observeAck = true
}
if e := insertReceiptTx(tx, req.From, seq, endpointID, DeliveryAccepted, "", nowMs); e != nil {
return e
}
@@ -65,9 +55,6 @@ SELECT state FROM deliveries WHERE seq = ? AND endpoint_id = ?`, seq, endpointID
if err != nil {
return out, err
}
if observeAck && a.met != nil {
a.met.AckSeconds.Observe(ackLatencySec)
}
if out.Result == DeliveryAccepted {
a.releaseLarge(seq, endpointID)
}
-7
View File
@@ -7,7 +7,6 @@ import (
"git.asio.asia/nixevol/NixMsg/internal/app/port"
"git.asio.asia/nixevol/NixMsg/internal/auth"
"git.asio.asia/nixevol/NixMsg/internal/config"
"git.asio.asia/nixevol/NixMsg/internal/metrics"
"git.asio.asia/nixevol/NixMsg/internal/protocol"
"git.asio.asia/nixevol/NixMsg/internal/store"
)
@@ -87,7 +86,6 @@ type App struct {
down port.Downlink
conns ConnRegistry
met *metrics.Registry
mu sync.Mutex
largeSem chan struct{}
@@ -119,11 +117,6 @@ func WithConnRegistry(c ConnRegistry) Option {
return func(a *App) { a.conns = c }
}
// WithMetrics 注入 Prometheus 注册表(投递耗时直方图)。
func WithMetrics(m *metrics.Registry) Option {
return func(a *App) { a.met = m }
}
// New 创建消息服务实现。
func New(db *store.DB, lim Limits, hash auth.HashPool, opts ...Option) *App {
if lim.RequestBurst <= 0 {
-9
View File
@@ -229,20 +229,11 @@ WHERE seq = ? AND endpoint_id = ? AND state = 'pending' AND pushed_conn IS NULL`
a.releaseLarge(it.seq, endpointID)
}
a.scheduleRepush(endpointID, time.Second)
} else {
a.observeDispatchToPush(it.sendAt, nowMs)
}
}
return a.pushReceipts(ctx, endpointID, connID, nowMs)
}
func (a *App) observeDispatchToPush(sendAtMs, pushedAtMs int64) {
if a.met == nil || pushedAtMs < sendAtMs {
return
}
a.met.DispatchToPushSeconds.Observe(float64(pushedAtMs-sendAtMs) / 1000.0)
}
func (a *App) rejectTooLarge(ctx context.Context, seq int64, endpointID, senderID string, nowMs int64) error {
return a.db.Queue.Do(ctx, func(tx *sql.Tx) error {
res, err := tx.Exec(`
+5 -11
View File
@@ -12,7 +12,6 @@ import (
"time"
"git.asio.asia/nixevol/NixMsg/internal/app/port"
"git.asio.asia/nixevol/NixMsg/internal/metrics"
mqtt "github.com/mochi-mqtt/server/v2"
"github.com/mochi-mqtt/server/v2/packets"
)
@@ -69,18 +68,15 @@ type Options struct {
Logger *slog.Logger
// OnPublishDropped 可选;nil 时仅打 debug 日志。
OnPublishDropped PublishDroppedFunc
// Metrics 可选;会话建立/断开时更新 nixmsg_connections。
Metrics *metrics.Registry
}
// Broker 内置 mochi,不自带监听端口。
type Broker struct {
server *mqtt.Server
auth Authenticator
uplink port.UplinkHandler
log *slog.Logger
onDrop PublishDroppedFunc
metrics *metrics.Registry
server *mqtt.Server
auth Authenticator
uplink port.UplinkHandler
log *slog.Logger
onDrop PublishDroppedFunc
hook *nixHook
@@ -109,7 +105,6 @@ type connState struct {
handshook bool
subscribedDown bool
largeHeld int
metricsCounted bool
mu sync.Mutex
handshakeTimer *time.Timer
@@ -156,7 +151,6 @@ func New(opts Options) (*Broker, error) {
uplink: uplink,
log: log,
onDrop: opts.OnPublishDropped,
metrics: opts.Metrics,
current: make(map[string]*connState),
byClient: make(map[*mqtt.Client]*connState),
queues: make(map[string]*uplinkQueue),
-19
View File
@@ -190,7 +190,6 @@ func (h *nixHook) OnSessionEstablished(cl *mqtt.Client, _ packets.Packet) {
MaxPacketSize: st.maxPacketSize,
}
_ = h.b.uplink.OnSessionEstablished(context.Background(), info)
h.noteConnectionOpen(st)
}
func (h *nixHook) OnDisconnect(cl *mqtt.Client, err error, _ bool) {
@@ -230,27 +229,9 @@ func (h *nixHook) OnDisconnect(cl *mqtt.Client, err error, _ bool) {
}
if sess, ok := h.b.uplink.(*Session); ok {
sess.HandleDisconnect(context.Background(), info, reason, isCurrent)
h.noteConnectionClose(st)
return
}
h.b.uplink.OnDisconnect(context.Background(), info, reason)
h.noteConnectionClose(st)
}
func (h *nixHook) noteConnectionOpen(st *connState) {
if h.b.metrics == nil || st == nil || st.metricsCounted {
return
}
h.b.metrics.Connections.WithLabelValues(string(st.transport)).Inc()
st.metricsCounted = true
}
func (h *nixHook) noteConnectionClose(st *connState) {
if h.b.metrics == nil || st == nil || !st.metricsCounted {
return
}
h.b.metrics.Connections.WithLabelValues(string(st.transport)).Dec()
st.metricsCounted = false
}
func (h *nixHook) OnQosComplete(cl *mqtt.Client, pk packets.Packet) {
-118
View File
@@ -1,118 +0,0 @@
package broker
import (
"io"
"net"
"net/http"
"net/http/httptest"
"strings"
"testing"
"time"
"git.asio.asia/nixevol/NixMsg/internal/metrics"
dto "github.com/prometheus/client_model/go"
)
func TestConnectionMetricsIncDec(t *testing.T) {
reg := metrics.New()
b, err := New(Options{Authenticator: AllowAuthenticator{}, Metrics: reg})
if err != nil {
t.Fatal(err)
}
defer func() { _ = b.Close() }()
if got := gaugeValue(t, reg, "nixmsg_connections", "tcp"); got != 0 {
t.Fatalf("before connect tcp=%v", got)
}
r, w := net.Pipe()
done := make(chan struct{})
go func() {
defer close(done)
_ = b.AttachTCP(r)
}()
connectAndSubscribe(t, w, "ep-metrics", 0)
deadline := time.Now().Add(3 * time.Second)
for {
if _, ok := b.ConnInfoOf("ep-metrics"); ok {
break
}
if time.Now().After(deadline) {
t.Fatal("session not established")
}
time.Sleep(10 * time.Millisecond)
}
if got := gaugeValue(t, reg, "nixmsg_connections", "tcp"); got != 1 {
t.Fatalf("after connect tcp=%v want 1", got)
}
_ = w.Close()
select {
case <-done:
case <-time.After(3 * time.Second):
t.Fatal("attach did not finish")
}
deadline = time.Now().Add(3 * time.Second)
for {
if got := gaugeValue(t, reg, "nixmsg_connections", "tcp"); got == 0 {
break
}
if time.Now().After(deadline) {
t.Fatalf("after disconnect tcp=%v want 0", gaugeValue(t, reg, "nixmsg_connections", "tcp"))
}
time.Sleep(10 * time.Millisecond)
}
}
func TestWSConnectionMetrics(t *testing.T) {
reg := metrics.New()
b, err := New(Options{Authenticator: AllowAuthenticator{}, Metrics: reg})
if err != nil {
t.Fatal(err)
}
defer func() { _ = b.Close() }()
mux := http.NewServeMux()
mux.Handle("/mqtt", b.WSHandler(nil))
srv := httptest.NewServer(mux)
defer srv.Close()
// 仅验证 handler 暴露指标文本仍含初始标签;建连用 TCP 测即可。
req := httptest.NewRequest(http.MethodGet, "/metrics", nil)
rec := httptest.NewRecorder()
reg.Handler().ServeHTTP(rec, req)
body, _ := io.ReadAll(rec.Body)
if !strings.Contains(string(body), `nixmsg_connections{transport="ws"} 0`) {
t.Fatalf("missing ws series: %s", body)
}
}
func gaugeValue(t *testing.T, reg *metrics.Registry, name, transport string) float64 {
t.Helper()
mfs, err := reg.Gatherer().Gather()
if err != nil {
t.Fatal(err)
}
for _, mf := range mfs {
if mf.GetName() != name {
continue
}
for _, m := range mf.GetMetric() {
if matchLabel(m, "transport", transport) {
return m.GetGauge().GetValue()
}
}
}
t.Fatalf("metric %s transport=%s not found", name, transport)
return 0
}
func matchLabel(m *dto.Metric, key, val string) bool {
for _, lp := range m.GetLabel() {
if lp.GetName() == key && lp.GetValue() == val {
return true
}
}
return false
}
-36
View File
@@ -1,36 +0,0 @@
package metrics
import (
"context"
"database/sql"
)
// SampleStoreGauges 按库内真实计数刷新端总数、待投递、定时消息(无对应行则为 0)。
func SampleStoreGauges(ctx context.Context, r *Registry, db *sql.DB) error {
if r == nil || db == nil {
return nil
}
var endpoints, pending, scheduled int
if err := db.QueryRowContext(ctx, `SELECT COUNT(*) FROM endpoints`).Scan(&endpoints); err != nil {
return err
}
if err := db.QueryRowContext(ctx, `SELECT COUNT(*) FROM deliveries WHERE state = 'pending'`).Scan(&pending); err != nil {
return err
}
if err := db.QueryRowContext(ctx, `SELECT COUNT(*) FROM messages WHERE state = 'scheduled'`).Scan(&scheduled); err != nil {
return err
}
r.EndpointsTotal.Set(float64(endpoints))
r.DeliveriesPending.Set(float64(pending))
r.MessagesScheduled.Set(float64(scheduled))
return nil
}
// SampleQueues 刷新写队列与密码哈希排队长度(传入当前真实长度,不做估算)。
func SampleQueues(r *Registry, writeQueueLen, passwordHashQueueLen int) {
if r == nil {
return
}
r.WriteQueueLength.Set(float64(writeQueueLen))
r.PasswordHashQueue.Set(float64(passwordHashQueueLen))
}
-82
View File
@@ -1,82 +0,0 @@
package metrics
import (
"context"
"path/filepath"
"testing"
"git.asio.asia/nixevol/NixMsg/internal/store"
)
func TestSampleStoreGaugesPendingNonZero(t *testing.T) {
t.Parallel()
dir := t.TempDir()
db, err := store.Open(filepath.Join(dir, "data"), "FULL")
if err != nil {
t.Fatal(err)
}
t.Cleanup(func() { _ = db.Close() })
ctx := context.Background()
_, err = db.Write.ExecContext(ctx, `
INSERT INTO endpoints(id, name, login_hash, enabled, source, created_at)
VALUES ('alice', 'A', 'x', 1, 'admin', 1)`)
if err != nil {
t.Fatal(err)
}
_, err = db.Write.ExecContext(ctx, `
INSERT INTO messages(sender_id, id, dest_kind, dest_id, send_at, keep, ttl_seconds, receipt, state, reason, created_at, content_type, body_enc, meta)
VALUES ('alice', 'm1', 'endpoint', 'bob', 100, 1, 0, 0, 'dispatched', '', 1, 'text/plain', 'utf8', '{}')`)
if err != nil {
t.Fatal(err)
}
var seq int64
if err := db.Write.QueryRowContext(ctx, `SELECT seq FROM messages WHERE id='m1'`).Scan(&seq); err != nil {
t.Fatal(err)
}
_, err = db.Write.ExecContext(ctx, `
INSERT INTO deliveries(seq, endpoint_id, send_at, keep, state, reason, attempts, updated_at)
VALUES (?, 'bob', 100, 1, 'pending', '', 0, 1)`, seq)
if err != nil {
t.Fatal(err)
}
_, err = db.Write.ExecContext(ctx, `
INSERT INTO messages(sender_id, id, dest_kind, dest_id, send_at, keep, ttl_seconds, receipt, state, reason, created_at, content_type, body_enc, meta)
VALUES ('alice', 'm2', 'endpoint', 'bob', 999999, 0, 0, 0, 'scheduled', '', 1, 'text/plain', 'utf8', '{}')`)
if err != nil {
t.Fatal(err)
}
reg := New()
if err := SampleStoreGauges(ctx, reg, db.Read); err != nil {
t.Fatal(err)
}
SampleQueues(reg, 3, 1)
mfs, err := reg.Gatherer().Gather()
if err != nil {
t.Fatal(err)
}
got := map[string]float64{}
for _, mf := range mfs {
switch mf.GetName() {
case "nixmsg_endpoints", "nixmsg_deliveries_pending", "nixmsg_messages_scheduled",
"nixmsg_write_queue_length", "nixmsg_password_hash_queue_length":
if len(mf.GetMetric()) > 0 {
got[mf.GetName()] = mf.GetMetric()[0].GetGauge().GetValue()
}
}
}
if got["nixmsg_endpoints"] != 1 {
t.Fatalf("endpoints=%v", got["nixmsg_endpoints"])
}
if got["nixmsg_deliveries_pending"] != 1 {
t.Fatalf("pending=%v", got["nixmsg_deliveries_pending"])
}
if got["nixmsg_messages_scheduled"] != 1 {
t.Fatalf("scheduled=%v", got["nixmsg_messages_scheduled"])
}
if got["nixmsg_write_queue_length"] != 3 || got["nixmsg_password_hash_queue_length"] != 1 {
t.Fatalf("queues=%v", got)
}
}
-7
View File
@@ -43,9 +43,6 @@ type Queue struct {
ready bool
lastWriteErr error
pending int
// OnBatchCommit 可选;每次合并提交成功后回调耗时(秒级指标用)。
OnBatchCommit func(d time.Duration)
}
// NewQueue 创建合并写入队列并启动写 goroutine。
@@ -132,7 +129,6 @@ func (q *Queue) loop() {
}
func (q *Queue) runBatch(batch []writeJob) {
started := time.Now()
defer func() {
q.mu.Lock()
q.pending -= len(batch)
@@ -233,9 +229,6 @@ func (q *Queue) runBatch(batch []writeJob) {
}
return
}
if q.OnBatchCommit != nil {
q.OnBatchCommit(time.Since(started))
}
for _, o := range outcomes {
if o.success {
o.job.res <- nil