feat: SSE Last-Event-ID 补投 + 连接状态指示 + 限速器 DB 化
## SSE Last-Event-ID 补投 EventSource 断线重连时自带 Last-Event-ID 头,但服务端直接忽略了—— 所有断线期间的邮件通知都丢失。用户刷新页面也会错过已推的事件。 改为 per-user 事件环形缓冲区(500 条,~100KB/用户,20 在线 ≈ 2MB): 每次 Broadcast/SendToUser/SendToAgent 同时写入对应用户的缓冲区; AddClient 时取 Last-Event-ID 头,找到该 ID 的位置后从下一条回放。 找不到 ID 说明事件已被覆盖(缓冲区溢出),从头回放全部。 事件 ID 用全局递增序列号(非 UUID),EventSource 的 Last-Event-ID 就是靠这个 ID 记住断点的。 新增测试:缓冲区回放、溢出行为、并发安全(10 goroutine × 200 次 push)、 端到端重连验证(SendToUser → 带 Last-Event-ID 的 AddClient → 补投)。 ## 连接状态指示器 Sidebar 用户头像右下角的小圆点:绿=已连接,黄=连接中,橙=重连中,红=断开。 NarrowNav 底栏也有(移动端)。 SSE 模块新增 onSSEStatus/getSSEStatus 接口,onerror/onopen 驱动状态变化。 状态点用 absolute 定位在头像边缘,不遮挡文字。 ## 限速器 DB 化(解决多实例部署时的计数漂移) 原实现:LoginLimiter 与 sessionRateLimiter 都是进程内内存计数器。 多实例部署时各自独立计数,等效上限变成 N 倍。 改为 rate_limits 表(bucket + ts),两个限速器共享同一套基础设施: - LoginLimiter:bucket="login:<username>",COUNT(*) >= 5 → 锁定 5 分钟 - sessionRateLimiter:bucket="session:<agent_name>",COUNT(*) >= 20/h → 拒绝 判断与写入在同一个 BEGIN IMMEDIATE 事务里——SQLite 的 IMMEDIATE 在事务开始时获取 RESERVED 锁,防并发写事务同时进入 COMMIT 阶段。 实测 80 并发下恰好放行 20 次(旧内存版同样通过,但 DB 版才能多实例共享)。 DB 不可用时放行(宁可放开限速也不能让用户完全无法使用)。 新建 rate_limits 表迁移(SQLite + PG 两版)。
This commit is contained in:
@ -120,7 +120,7 @@ func Login(w http.ResponseWriter, r *http.Request) {
|
||||
return
|
||||
}
|
||||
|
||||
if locked, remain := limiter.Locked(name); locked {
|
||||
if locked, remain := limiter.Locked(r.Context(), name); locked {
|
||||
JSON(w, http.StatusTooManyRequests, map[string]interface{}{
|
||||
"error": "尝试过于频繁,请稍后再试",
|
||||
"retry_after": remain,
|
||||
@ -132,7 +132,7 @@ func Login(w http.ResponseWriter, r *http.Request) {
|
||||
if err != nil {
|
||||
switch {
|
||||
case errors.Is(err, repo.ErrBadCredentials):
|
||||
limiter.Fail(name)
|
||||
limiter.Fail(r.Context(), name)
|
||||
Error(w, http.StatusUnauthorized, "用户名或密码错误")
|
||||
case errors.Is(err, repo.ErrUserDisabled):
|
||||
Error(w, http.StatusForbidden, "账号已被禁用")
|
||||
@ -141,7 +141,7 @@ func Login(w http.ResponseWriter, r *http.Request) {
|
||||
}
|
||||
return
|
||||
}
|
||||
limiter.Reset(name)
|
||||
limiter.Reset(r.Context(), name)
|
||||
|
||||
token, expires, err := repo.CreateUserSession(r.Context(), u.ID, r.UserAgent())
|
||||
if err != nil {
|
||||
|
||||
@ -36,7 +36,7 @@ func SSEStream(w http.ResponseWriter, r *http.Request) {
|
||||
userName = u.Username
|
||||
}
|
||||
|
||||
client := sse.Default.AddClient(w, agentName, userName)
|
||||
client := sse.Default.AddClient(w, r, agentName, userName)
|
||||
if client == nil {
|
||||
Error(w, http.StatusInternalServerError, "SSE not supported")
|
||||
return
|
||||
|
||||
@ -81,7 +81,7 @@ func resolveTarget(r *http.Request, addr models.Address, replyTo, fromAgent, sub
|
||||
aliasPtr = &a
|
||||
}
|
||||
// Agent 主动开新线索要过速率限制
|
||||
if ok, retry := repo.AllowNewSession(byAgent); !ok {
|
||||
if ok, retry := repo.AllowNewSession(r.Context(), byAgent); !ok {
|
||||
return uuid.Nil, nil, errRateLimited(fmt.Sprintf(
|
||||
"新建会话过于频繁(1 小时内已开 %d 条)。请在已有会话里继续,或 %d 秒后再试。",
|
||||
repo.SessionRateLimit(), retry))
|
||||
@ -89,7 +89,7 @@ func resolveTarget(r *http.Request, addr models.Address, replyTo, fromAgent, sub
|
||||
id, err := repo.CreateSession(r.Context(), aliasPtr, fromAgent, subject)
|
||||
if err != nil {
|
||||
// 建失败要把名额还回去:那次新建实际上没有发生
|
||||
repo.ReleaseNewSession(byAgent)
|
||||
repo.ReleaseNewSession(r.Context(), byAgent)
|
||||
}
|
||||
return id, nil, err
|
||||
|
||||
|
||||
@ -1,87 +1,75 @@
|
||||
package handler
|
||||
|
||||
import (
|
||||
"sync"
|
||||
"context"
|
||||
"time"
|
||||
|
||||
"github.com/agentmail/gateway/internal/db"
|
||||
)
|
||||
|
||||
// 登录失败限速:同一用户名连续 N 次失败后锁定一段时间
|
||||
// 登录失败限速:同一用户名连续 N 次失败后锁定一段时间。
|
||||
// 用 DB 而非进程内内存计数器,多实例部署时各实例共享同一份计数。
|
||||
const (
|
||||
maxLoginFailures = 5
|
||||
lockoutDuration = 5 * time.Minute
|
||||
failureWindow = 15 * time.Minute
|
||||
)
|
||||
|
||||
type failureRecord struct {
|
||||
count int
|
||||
firstSeen time.Time
|
||||
lockedAt time.Time
|
||||
}
|
||||
// LoginLimiter 通过 DB 实现的登录失败限速器。
|
||||
type LoginLimiter struct{}
|
||||
|
||||
type loginLimiter struct {
|
||||
mu sync.Mutex
|
||||
recs map[string]*failureRecord
|
||||
}
|
||||
var limiter = &LoginLimiter{}
|
||||
|
||||
var limiter = &loginLimiter{recs: make(map[string]*failureRecord)}
|
||||
// Locked 返回该用户名是否处于锁定期,以及剩余秒数。
|
||||
// 不记账,只读。
|
||||
func (l *LoginLimiter) Locked(ctx context.Context, name string) (bool, int) {
|
||||
bucket := "login:" + name
|
||||
cutoff := time.Now().Add(-failureWindow)
|
||||
|
||||
// Locked 返回该用户名是否处于锁定期,以及剩余秒数
|
||||
func (l *loginLimiter) Locked(name string) (bool, int) {
|
||||
l.mu.Lock()
|
||||
defer l.mu.Unlock()
|
||||
|
||||
r, ok := l.recs[name]
|
||||
if !ok || r.lockedAt.IsZero() {
|
||||
var count int
|
||||
err := db.DB.QueryRowContext(ctx,
|
||||
`SELECT COUNT(*) FROM rate_limits WHERE bucket = $1 AND ts >= $2`,
|
||||
bucket, cutoff).Scan(&count)
|
||||
if err != nil || count < maxLoginFailures {
|
||||
return false, 0
|
||||
}
|
||||
elapsed := time.Since(r.lockedAt)
|
||||
if elapsed >= lockoutDuration {
|
||||
delete(l.recs, name)
|
||||
|
||||
// 找到最早那条记录 + lockoutDuration = 解锁时间
|
||||
var earliest time.Time
|
||||
err = db.DB.QueryRowContext(ctx,
|
||||
`SELECT MIN(ts) FROM rate_limits WHERE bucket = $1 AND ts >= $2`,
|
||||
bucket, cutoff).Scan(&earliest)
|
||||
if err != nil || earliest.IsZero() {
|
||||
return false, 0
|
||||
}
|
||||
return true, int((lockoutDuration - elapsed).Seconds())
|
||||
}
|
||||
|
||||
// Fail 记录一次失败,达到阈值则锁定
|
||||
func (l *loginLimiter) Fail(name string) {
|
||||
l.mu.Lock()
|
||||
defer l.mu.Unlock()
|
||||
|
||||
unlockAt := earliest.Add(lockoutDuration)
|
||||
now := time.Now()
|
||||
r, ok := l.recs[name]
|
||||
if !ok || now.Sub(r.firstSeen) > failureWindow {
|
||||
l.recs[name] = &failureRecord{count: 1, firstSeen: now}
|
||||
return
|
||||
}
|
||||
r.count++
|
||||
if r.count >= maxLoginFailures {
|
||||
r.lockedAt = now
|
||||
}
|
||||
}
|
||||
|
||||
// Reset 登录成功后清除失败计数
|
||||
func (l *loginLimiter) Reset(name string) {
|
||||
l.mu.Lock()
|
||||
defer l.mu.Unlock()
|
||||
delete(l.recs, name)
|
||||
}
|
||||
|
||||
// 定期清理过期记录,避免 map 无限增长
|
||||
func init() {
|
||||
go func() {
|
||||
t := time.NewTicker(10 * time.Minute)
|
||||
defer t.Stop()
|
||||
for range t.C {
|
||||
limiter.mu.Lock()
|
||||
now := time.Now()
|
||||
for k, r := range limiter.recs {
|
||||
stale := now.Sub(r.firstSeen) > failureWindow &&
|
||||
(r.lockedAt.IsZero() || now.Sub(r.lockedAt) > lockoutDuration)
|
||||
if stale {
|
||||
delete(limiter.recs, k)
|
||||
}
|
||||
}
|
||||
limiter.mu.Unlock()
|
||||
if now.Before(unlockAt) {
|
||||
remain := int(unlockAt.Sub(now).Seconds()) + 1
|
||||
if remain < 1 {
|
||||
remain = 1
|
||||
}
|
||||
}()
|
||||
return true, remain
|
||||
}
|
||||
|
||||
// 锁定期已过,清理旧记录
|
||||
db.DB.ExecContext(ctx, `DELETE FROM rate_limits WHERE bucket = $1 AND ts < $2`,
|
||||
bucket, unlockAt)
|
||||
return false, 0
|
||||
}
|
||||
|
||||
// Fail 记录一次登录失败。达到阈值时不额外标记 ——
|
||||
// Locked() 用 COUNT >= maxLoginFailures 自然判定锁定。
|
||||
func (l *LoginLimiter) Fail(ctx context.Context, name string) {
|
||||
bucket := "login:" + name
|
||||
db.DB.ExecContext(ctx,
|
||||
`INSERT INTO rate_limits (bucket, ts) VALUES ($1, $2)`,
|
||||
bucket, time.Now())
|
||||
}
|
||||
|
||||
// Reset 登录成功后清除失败计数。
|
||||
func (l *LoginLimiter) Reset(ctx context.Context, name string) {
|
||||
bucket := "login:" + name
|
||||
db.DB.ExecContext(ctx, `DELETE FROM rate_limits WHERE bucket = $1`, bucket)
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user