feat: 按运营商账户调度的分发器、卡任务执行器与模式切换

- 四个轮询处理器拆为步骤函数 Step(返回结果分类),旧 Asynq Handle 只做并发控制与原有重入队映射,行为不变;carrier 模式下旧 handler 直接结束
- 网关限流增加等待模式(ctx 标记),NSQ 消费者等名额到单步超时,超时按限流处理不计失败;记录限流等待次数
- 卡任务执行器:核对 token 并抢执行权,按配置步骤顺序串行执行到期检查(单步超时、总超时、panic 恢复、流量已触发评估时跳过套餐),执行中续期并 Touch,续期失败立即停止;收尾写回到期、连续失败、退避、延迟确认与告警
- 分发器:每秒按账户抢分发锁领取(执行上限按消费能力收紧、优先占比、按限流组剩余 QPS 估算),发布到 polling.card.{账户},发布失败释放;消费旧优先提示与手动触发;心跳与处理中标记到期计数
- NSQ 生产者与按账户动态订阅的消费者管理器,优雅退出宽限期
- 失败告警异步写 tb_polling_failure_alert(同卡同类型一条未处理,成功转已恢复),账户窗口内告警数超阈值产生账户级告警
- 模式监听:切 carrier 时旧分片调度停止出队,加锁初始化新结构后开始领取并订阅;回 legacy 时停止消费、重建旧分片队列;生命周期(API 与 Worker)在 carrier 模式写账户队列
This commit is contained in:
邹进
2026-10-25 18:44:23 +09:00
parent e7f8c4bae9
commit 2fe7c1f8ac
23 changed files with 1899 additions and 72 deletions

View File

@@ -0,0 +1,199 @@
package polling
import (
"context"
"sync"
"sync/atomic"
"time"
"go.uber.org/zap"
"github.com/break/junhong_cmp_fiber/internal/model"
"github.com/break/junhong_cmp_fiber/internal/polling/carrierqueue"
"github.com/break/junhong_cmp_fiber/internal/store/postgres"
"github.com/break/junhong_cmp_fiber/pkg/constants"
)
// CarrierScheduling 是按运营商账户调度在轮询包内的接缝:判断当前模式,并按卡计算各检查类型到期时间写入账户队列。
// 所有写入轮询结构的入口(初始化、生命周期、优先提示、手动触发)按当前模式写新结构或旧分片结构。
type CarrierScheduling struct {
store *carrierqueue.Store
settings *carrierqueue.SettingsProvider
configMgr *PollingConfigManager
logger *zap.Logger
}
// NewCarrierScheduling 创建按运营商账户调度接缝。
func NewCarrierScheduling(store *carrierqueue.Store, settings *carrierqueue.SettingsProvider, configMgr *PollingConfigManager, logger *zap.Logger) *CarrierScheduling {
if logger == nil {
logger = zap.NewNop()
}
return &CarrierScheduling{store: store, settings: settings, configMgr: configMgr, logger: logger}
}
// Active 判断当前是否为按运营商账户调度。
func (c *CarrierScheduling) Active(ctx context.Context) bool {
return c != nil && c.store != nil && c.settings != nil && c.settings.Current(ctx).CarrierMode()
}
// Store 返回账户队列存储。
func (c *CarrierScheduling) Store() *carrierqueue.Store {
if c == nil {
return nil
}
return c.store
}
// Settings 返回调度参数提供者。
func (c *CarrierScheduling) Settings() *carrierqueue.SettingsProvider {
if c == nil {
return nil
}
return c.settings
}
// DuesForCard 按卡匹配的轮询配置计算各检查类型下次到期时间。
// fromLastCheck 为 true 时按各类型上次检查时间计算(初始化);否则按当前时间加抖动(新卡、恢复轮询、状态变化)。
// 间隔为 0 或未配置的类型不返回;重置日前加密窗口规则算出的时间更早时以它为准。
func (c *CarrierScheduling) DuesForCard(ctx context.Context, card *model.IotCard, fromLastCheck bool) map[string]time.Time {
dues := map[string]time.Time{}
if c == nil || c.configMgr == nil || card == nil {
return dues
}
now := time.Now()
for taskType, info := range c.configMgr.MergedTaskIntervals(card) {
checkType := constants.CheckTypeFromTaskType(taskType)
if checkType == "" || info.Interval <= 0 {
continue
}
var due time.Time
if fromLastCheck {
due = calculateNextCheckTime(lastCheckAtByTaskType(card, taskType), info.Interval, now)
if windowDue := c.configMgr.NextCheckAt(ctx, card, taskType, info, now); windowDue.Before(due) && windowDue.After(now) {
due = windowDue
}
} else {
due = calcInitialDelay(info.Interval)
}
dues[checkType] = due
}
return dues
}
// UpsertCard 把卡写入所属账户队列;fromLastCheck 含义同 DuesForCard。
func (c *CarrierScheduling) UpsertCard(ctx context.Context, card *model.IotCard, fromLastCheck bool) {
if c == nil || c.store == nil || card == nil {
return
}
if err := c.store.Upsert(ctx, card.ID, card.CarrierID, c.DuesForCard(ctx, card, fromLastCheck)); err != nil {
c.logger.Warn("写入运营商账户轮询队列失败", zap.Uint("card_id", card.ID), zap.Uint("carrier_id", card.CarrierID), zap.Error(err))
}
}
// RemoveCard 把卡移出所属账户队列;carrierID 为 0 时按到期信息中记录的账户移出。
func (c *CarrierScheduling) RemoveCard(ctx context.Context, cardID, carrierID uint) {
if c == nil || c.store == nil {
return
}
if carrierID == 0 {
due, err := c.store.ReadDue(ctx, cardID)
if err != nil || due.CarrierID == 0 {
return
}
carrierID = due.CarrierID
}
if err := c.store.Remove(ctx, cardID, carrierID); err != nil {
c.logger.Warn("移出运营商账户轮询队列失败", zap.Uint("card_id", cardID), zap.Error(err))
}
}
// RequestTaskType 把卡的某检查类型(旧任务类型 polling:xxx)设为立即到期放入账户优先队列(优先提示、手动触发)。
// 卡不在轮询中时不处理。
func (c *CarrierScheduling) RequestTaskType(ctx context.Context, cardID, carrierID uint, taskType string) {
checkType := constants.CheckTypeFromTaskType(taskType)
if c == nil || c.store == nil || checkType == "" {
return
}
if _, err := c.store.RequestCheck(ctx, carrierqueue.CheckRequest{CardID: cardID, CarrierID: carrierID, CheckTypes: []string{checkType}}); err != nil {
c.logger.Warn("写入运营商账户优先队列失败", zap.Uint("card_id", cardID), zap.String("task_type", taskType), zap.Error(err))
}
}
// CarrierInitializer 从数据库按批初始化全部运营商账户队列与卡到期信息;进度与旧初始化器同结构,供 init-progress 展示。
type CarrierInitializer struct {
scheduling *CarrierScheduling
iotCardStore *postgres.IotCardStore
logger *zap.Logger
mu sync.RWMutex
progress InitProgress
running atomic.Bool
}
// NewCarrierInitializer 创建新结构初始化器。
func NewCarrierInitializer(scheduling *CarrierScheduling, iotCardStore *postgres.IotCardStore, logger *zap.Logger) *CarrierInitializer {
if logger == nil {
logger = zap.NewNop()
}
return &CarrierInitializer{scheduling: scheduling, iotCardStore: iotCardStore, logger: logger}
}
// Progress 返回初始化进度快照。
func (i *CarrierInitializer) Progress() InitProgress {
i.mu.RLock()
defer i.mu.RUnlock()
return i.progress
}
// Run 同步执行一次全量初始化:按批读取启用轮询的卡,计算各类型到期并写入账户队列。返回写入的卡数。
func (i *CarrierInitializer) Run(ctx context.Context) (int64, error) {
if !i.running.CompareAndSwap(false, true) {
return 0, nil
}
defer i.running.Store(false)
started := time.Now()
total, err := i.iotCardStore.CountForPolling(ctx)
if err != nil {
i.setProgress(func(p *InitProgress) { p.Status, p.ErrorMessage = "failed", err.Error() })
return 0, err
}
i.setProgress(func(p *InitProgress) {
*p = InitProgress{TotalCards: total, StartTime: started, Status: "running"}
})
const batchSize = 2000
var lastID uint
var written int64
for {
if err := ctx.Err(); err != nil {
i.setProgress(func(p *InitProgress) { p.Status, p.ErrorMessage = "failed", err.Error() })
return written, err
}
cards, fetchErr := i.iotCardStore.ListForPollingBatch(ctx, lastID, batchSize)
if fetchErr != nil {
i.setProgress(func(p *InitProgress) { p.Status, p.ErrorMessage = "failed", fetchErr.Error() })
return written, fetchErr
}
if len(cards) == 0 {
break
}
if tier := i.scheduling.configMgr.TierResolver(); tier != nil {
tier.Prefill(ctx, cards)
}
for _, card := range cards {
i.scheduling.UpsertCard(ctx, card, true)
written++
}
lastID = cards[len(cards)-1].ID
loaded := written
i.setProgress(func(p *InitProgress) { p.LoadedCards, p.LastBatchTime = loaded, time.Now() })
}
i.setProgress(func(p *InitProgress) { p.Status = "completed" })
i.logger.Info("运营商账户轮询结构初始化完成", zap.Int64("cards", written), zap.Duration("duration", time.Since(started)))
return written, nil
}
func (i *CarrierInitializer) setProgress(update func(*InitProgress)) {
i.mu.Lock()
update(&i.progress)
i.mu.Unlock()
}

View File

@@ -0,0 +1,177 @@
package carrierqueue
import (
"context"
"strconv"
"time"
"github.com/redis/go-redis/v9"
"go.uber.org/zap"
"gorm.io/gorm"
"github.com/break/junhong_cmp_fiber/pkg/constants"
)
// FailureEvent 是一次需要告警的检查失败。
type FailureEvent struct {
CardID uint
ICCID string
CarrierID uint
CheckType string
Category string
Message string
FailCount int
FirstFailed time.Time
At time.Time
}
type alertOp struct {
failure *FailureEvent
recovered bool
cardID uint
checkType string
}
// AlertWriter 异步写卡轮询失败告警:同一卡同一检查类型只保留一条未处理告警,后续失败更新计数;
// 首次成功转为已恢复;同账户窗口内新增告警超阈值时产生账户级告警。写入失败只记日志,不影响轮询。
type AlertWriter struct {
db *gorm.DB
redis *redis.Client
settings *SettingsProvider
logger *zap.Logger
ch chan alertOp
}
// NewAlertWriter 创建告警写入器;调用 Start 后开始消费。
func NewAlertWriter(db *gorm.DB, redisClient *redis.Client, settings *SettingsProvider, logger *zap.Logger) *AlertWriter {
if logger == nil {
logger = zap.NewNop()
}
return &AlertWriter{db: db, redis: redisClient, settings: settings, logger: logger, ch: make(chan alertOp, 2048)}
}
// Start 启动后台写入协程,直到 ctx 结束。
func (w *AlertWriter) Start(ctx context.Context) {
go func() {
for {
select {
case <-ctx.Done():
return
case op := <-w.ch:
w.apply(op)
}
}
}()
}
// ReportFailure 提交一条失败告警(达到阈值后才调用)。
func (w *AlertWriter) ReportFailure(event FailureEvent) {
w.offer(alertOp{failure: &event})
}
// ReportRecovered 提交一次恢复:该卡该检查类型的未处理告警转为已恢复。
func (w *AlertWriter) ReportRecovered(cardID uint, checkType string) {
w.offer(alertOp{recovered: true, cardID: cardID, checkType: checkType})
}
func (w *AlertWriter) offer(op alertOp) {
if w == nil {
return
}
select {
case w.ch <- op:
default:
w.logger.Warn("卡轮询失败告警队列已满,丢弃本次写入", zap.String("metric", "polling.failure_alert.dropped"))
}
}
func (w *AlertWriter) apply(op alertOp) {
ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
defer cancel()
if op.recovered {
err := w.db.WithContext(ctx).Exec(`UPDATE tb_polling_failure_alert SET status = ?, recovered_at = NOW(), updated_at = NOW()
WHERE alert_level = ? AND card_id = ? AND check_type = ? AND status = ?`,
constants.PollingAlertStatusRecovered, constants.PollingAlertLevelCard, op.cardID, op.checkType, constants.PollingAlertStatusOpen).Error
if err != nil {
w.logger.Warn("卡轮询告警转为已恢复失败", zap.Uint("card_id", op.cardID), zap.String("check_type", op.checkType), zap.Error(err))
}
return
}
event := op.failure
var inserted bool
err := w.db.WithContext(ctx).Raw(`INSERT INTO tb_polling_failure_alert
(alert_level, card_id, iccid, carrier_id, check_type, error_category, fail_count, first_failed_at, last_failed_at, last_error, status)
VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
ON CONFLICT (alert_level, card_id, carrier_id, check_type) WHERE status = 1
DO UPDATE SET fail_count = EXCLUDED.fail_count, error_category = EXCLUDED.error_category,
last_failed_at = EXCLUDED.last_failed_at, last_error = EXCLUDED.last_error, updated_at = NOW()
RETURNING (xmax = 0)`,
constants.PollingAlertLevelCard, event.CardID, event.ICCID, event.CarrierID, event.CheckType, event.Category, event.FailCount,
nullableTime(event.FirstFailed), event.At, truncate(event.Message, 500), constants.PollingAlertStatusOpen).Row().Scan(&inserted)
if err != nil {
w.logger.Warn("写卡轮询失败告警失败", zap.Uint("card_id", event.CardID), zap.String("check_type", event.CheckType), zap.Error(err))
return
}
if !inserted {
return
}
w.logger.Error("卡轮询持续失败告警", zap.Uint("card_id", event.CardID), zap.String("iccid", event.ICCID),
zap.Uint("carrier_id", event.CarrierID), zap.String("check_type", event.CheckType), zap.String("category", event.Category),
zap.Int("fail_count", event.FailCount), zap.String("metric", "polling.failure_alert.opened"))
w.checkAccountAlert(ctx, event.CarrierID)
}
// checkAccountAlert 同账户窗口内新增卡级告警数超过阈值时,产生(或更新)一条账户级告警。
func (w *AlertWriter) checkAccountAlert(ctx context.Context, carrierID uint) {
settings := w.settings.Current(ctx)
window, limit := settings.AccountAlertWindow, settings.AccountAlertLimit
if window <= 0 || limit <= 0 || w.redis == nil {
return
}
now := time.Now()
key := constants.RedisCarrierAccountAlertKey(carrierID)
member := strconv.FormatInt(now.UnixNano(), 36)
pipe := w.redis.TxPipeline()
pipe.ZAdd(ctx, key, redis.Z{Score: float64(now.UnixMilli()), Member: member})
pipe.ZRemRangeByScore(ctx, key, "-inf", strconv.FormatInt(now.Add(-window).UnixMilli(), 10))
count := pipe.ZCard(ctx, key)
pipe.Expire(ctx, key, window*2)
if _, err := pipe.Exec(ctx); err != nil {
return
}
if int(count.Val()) <= limit {
return
}
var inserted bool
err := w.db.WithContext(ctx).Raw(`INSERT INTO tb_polling_failure_alert
(alert_level, card_id, carrier_id, check_type, error_category, fail_count, first_failed_at, last_failed_at, last_error, status)
VALUES (?, 0, ?, ?, '', ?, ?, ?, ?, ?)
ON CONFLICT (alert_level, card_id, carrier_id, check_type) WHERE status = 1
DO UPDATE SET fail_count = EXCLUDED.fail_count, last_failed_at = EXCLUDED.last_failed_at, updated_at = NOW()
RETURNING (xmax = 0)`,
constants.PollingAlertLevelCarrier, carrierID, constants.PollingAlertCheckTypeAll, count.Val(), now, now,
"窗口内同账户新增失败告警数超过阈值,通常是账户额度、凭证或上游故障,请联系上游处理账户", constants.PollingAlertStatusOpen).Row().Scan(&inserted)
if err != nil {
w.logger.Warn("写账户级轮询告警失败", zap.Uint("carrier_id", carrierID), zap.Error(err))
return
}
if inserted {
w.logger.Error("运营商账户轮询失败告警数超过阈值", zap.Uint("carrier_id", carrierID), zap.Int64("count", count.Val()),
zap.Duration("window", window), zap.String("metric", "polling.failure_alert.carrier_opened"))
}
}
func nullableTime(t time.Time) any {
if t.IsZero() {
return nil
}
return t
}
func truncate(s string, max int) string {
runes := []rune(s)
if len(runes) <= max {
return s
}
return string(runes[:max])
}

View File

@@ -0,0 +1,326 @@
package carrierqueue
import (
"context"
"encoding/json"
stderrors "errors"
"strconv"
"sync"
"time"
"github.com/redis/go-redis/v9"
"go.uber.org/zap"
"gorm.io/gorm"
"github.com/break/junhong_cmp_fiber/internal/model"
"github.com/break/junhong_cmp_fiber/pkg/config"
"github.com/break/junhong_cmp_fiber/pkg/constants"
)
// CardMessage 是卡任务 NSQ 消息体。
type CardMessage struct {
CardID uint `json:"card_id"`
CarrierID uint `json:"carrier_id"`
Token string `json:"token"`
DispatchedAt int64 `json:"dispatched_at"`
}
// Publisher 发布 NSQ 消息。
type Publisher interface {
Publish(topic string, body []byte) error
}
// BudgetProvider 返回账户各查询操作所属限流组本秒剩余名额;limited=false 表示不按 QPS 限制领取。
type BudgetProvider interface {
Budget(ctx context.Context, carrierID uint, ops []string) (groups map[string]int, opGroup map[string]string, limited bool)
}
// CarrierParams 是一个运营商账户生效的调度参数(已合并全局默认)。
type CarrierParams struct {
InflightLimit int
CardMaxInFlight int
PriorityRatio float64
Configured bool // 是否有该账户的配置行
}
// carrierEntry 是分发器缓存的账户信息。
type carrierEntry struct {
id uint
realnameControl bool
}
// Dispatcher 每个分发间隔对每个运营商账户抢分发锁、领取到期卡并发布到账户 topic。多实例安全。
type Dispatcher struct {
db *gorm.DB
store *Store
settings *SettingsProvider
publisher Publisher
budgets BudgetProvider
cfg config.CarrierSchedulerConfig
instanceID string
logger *zap.Logger
mu sync.Mutex
carriers []carrierEntry
params map[uint]model.CarrierPollingParam
loadedAt time.Time
tightened map[uint]time.Time
}
// NewDispatcher 创建分发器。
func NewDispatcher(db *gorm.DB, store *Store, settings *SettingsProvider, publisher Publisher, budgets BudgetProvider,
cfg config.CarrierSchedulerConfig, instanceID string, logger *zap.Logger) *Dispatcher {
if logger == nil {
logger = zap.NewNop()
}
return &Dispatcher{db: db, store: store, settings: settings, publisher: publisher, budgets: budgets, cfg: cfg,
instanceID: instanceID, logger: logger, params: map[uint]model.CarrierPollingParam{}, tightened: map[uint]time.Time{}}
}
// Run 按分发间隔循环,直到 ctx 结束。
func (d *Dispatcher) Run(ctx context.Context) {
interval := d.cfg.DispatchInterval
if interval <= 0 {
interval = time.Second
}
ticker := time.NewTicker(interval)
defer ticker.Stop()
for {
select {
case <-ctx.Done():
return
case <-ticker.C:
d.tick(ctx, interval)
}
}
}
// Ready 判断新调度是否可领取:模式为 carrier 且新结构初始化已完成。
func (d *Dispatcher) Ready(ctx context.Context) bool {
if !d.settings.Current(ctx).CarrierMode() {
return false
}
state, err := d.store.redis.Get(ctx, constants.RedisCarrierSchedulerInitStateKey).Result()
return err == nil && state == InitStateDone
}
func (d *Dispatcher) tick(ctx context.Context, interval time.Duration) {
defer func() {
if r := recover(); r != nil {
d.logger.Error("分发器 panic,已恢复", zap.Any("panic", r))
}
}()
if !d.Ready(ctx) {
return
}
d.store.redis.HSet(ctx, constants.RedisCarrierSchedulerHeartbeatKey, d.instanceID, time.Now().UnixMilli())
d.drainLegacyHints(ctx)
d.refreshCarriers(ctx)
d.mu.Lock()
carriers := append([]carrierEntry(nil), d.carriers...)
d.mu.Unlock()
lockTTL := interval * 9 / 10
if lockTTL <= 0 {
lockTTL = 900 * time.Millisecond
}
for _, carrier := range carriers {
if ctx.Err() != nil {
return
}
ok, err := d.store.redis.SetNX(ctx, constants.RedisCarrierDispatchLockKey(carrier.id), d.instanceID, lockTTL).Result()
if err != nil || !ok {
continue
}
d.dispatchCarrier(ctx, carrier)
}
}
func (d *Dispatcher) dispatchCarrier(ctx context.Context, carrier carrierEntry) {
params := d.ParamsFor(carrier.id)
limit := params.InflightLimit
// 执行上限不得大于订阅进程 max_in_flight 之和,否则领出的消息在 NSQ 中排队到租约过期;不满足时收紧本轮上限并上报。
processes := d.cfg.SubscriberProcesses
if processes <= 0 {
processes = 1
}
if capacity := params.CardMaxInFlight * processes; capacity > 0 && limit > capacity {
limit = capacity
d.reportTightened(carrier.id, params.InflightLimit, capacity)
}
input := ClaimInput{
CarrierID: carrier.id, Lease: d.lease(), InflightLimit: limit, PriorityRatio: params.PriorityRatio,
TokenPrefix: d.instanceID + ":" + strconv.FormatInt(time.Now().UnixNano(), 36),
CountRealname: carrier.realnameControl,
}
if d.budgets != nil {
ops := []string{constants.GatewayOpQueryStatus, constants.GatewayOpQueryFlow, constants.GatewayOpQueryRealname}
if groups, opGroup, limited := d.budgets.Budget(ctx, carrier.id, ops); limited {
input.GroupBudget = groups
input.TypeGroup = map[string]string{
constants.CheckTypeCardStatus: opGroup[constants.GatewayOpQueryStatus],
constants.CheckTypeCarddata: opGroup[constants.GatewayOpQueryFlow],
constants.CheckTypeRealname: opGroup[constants.GatewayOpQueryRealname],
}
}
}
claimed, expired, err := d.store.Claim(ctx, input)
if err != nil {
d.logger.Warn("运营商账户领取失败", zap.Uint("carrier_id", carrier.id), zap.Error(err))
return
}
if expired > 0 {
minute := strconv.FormatInt(time.Now().Unix()/60, 10)
key := constants.RedisCarrierExpiredClaimsKey(carrier.id)
d.store.redis.HIncrBy(ctx, key, minute, expired)
d.store.redis.Expire(ctx, key, 2*time.Hour)
d.logger.Warn("运营商账户处理中标记到期(执行崩溃或超时)", zap.Uint("carrier_id", carrier.id), zap.Int64("count", expired),
zap.String("metric", "polling.carrier.claim_expired"))
}
topic := constants.NSQCardTopic(carrier.id)
for _, item := range claimed {
body, _ := json.Marshal(CardMessage{CardID: item.CardID, CarrierID: carrier.id, Token: item.Token, DispatchedAt: time.Now().UnixMilli()})
if err := d.publisher.Publish(topic, body); err != nil {
// 发布失败:用独立短超时上下文按 token 删除处理中标记并恢复 score;恢复也失败则等租约到期后自然重新领取。
releaseCtx, cancel := context.WithTimeout(context.WithoutCancel(ctx), 2*time.Second)
if releaseErr := d.store.Release(releaseCtx, item.CardID, carrier.id, item.Token); releaseErr != nil {
d.logger.Warn("发布失败后释放处理中标记失败", zap.Uint("card_id", item.CardID), zap.Error(releaseErr))
}
cancel()
d.logger.Warn("发布卡任务失败,已释放", zap.Uint("card_id", item.CardID), zap.String("topic", topic), zap.Error(err))
}
}
}
func (d *Dispatcher) lease() time.Duration {
if d.cfg.Lease > 0 {
return d.cfg.Lease
}
return 5 * time.Minute
}
func (d *Dispatcher) reportTightened(carrierID uint, configured, capacity int) {
d.mu.Lock()
last := d.tightened[carrierID]
if time.Since(last) < 10*time.Minute {
d.mu.Unlock()
return
}
d.tightened[carrierID] = time.Now()
d.mu.Unlock()
d.logger.Warn("账户执行上限大于订阅进程 max_in_flight 之和,已按消费能力收紧本轮领取上限",
zap.Uint("carrier_id", carrierID), zap.Int("inflight_limit", configured), zap.Int("capacity", capacity),
zap.String("metric", "polling.carrier.inflight_tightened"))
}
// refreshCarriers 每 10 秒重新读取运营商账户与调度参数。
func (d *Dispatcher) refreshCarriers(ctx context.Context) {
d.mu.Lock()
fresh := time.Since(d.loadedAt) < 10*time.Second
d.mu.Unlock()
if fresh {
return
}
var carriers []model.Carrier
if err := d.db.WithContext(ctx).Select("id", "realname_control_enabled").Order("id").Find(&carriers).Error; err != nil {
d.logger.Warn("读取运营商账户失败,沿用上一次列表", zap.Error(err))
return
}
var rows []model.CarrierPollingParam
if err := d.db.WithContext(ctx).Find(&rows).Error; err != nil {
d.logger.Warn("读取运营商账户调度参数失败,沿用上一次参数", zap.Error(err))
return
}
entries := make([]carrierEntry, 0, len(carriers))
for _, carrier := range carriers {
entries = append(entries, carrierEntry{id: carrier.ID, realnameControl: carrier.RealnameControlEnabled == 1})
}
params := make(map[uint]model.CarrierPollingParam, len(rows))
for _, row := range rows {
params[row.CarrierID] = row
}
d.mu.Lock()
d.carriers, d.params, d.loadedAt = entries, params, time.Now()
d.mu.Unlock()
}
// CarrierIDs 返回当前缓存的运营商账户 ID(消费者据此订阅)。
func (d *Dispatcher) CarrierIDs(ctx context.Context) []uint {
d.refreshCarriers(ctx)
d.mu.Lock()
defer d.mu.Unlock()
ids := make([]uint, 0, len(d.carriers))
for _, carrier := range d.carriers {
ids = append(ids, carrier.id)
}
return ids
}
// ParamsFor 返回账户生效调度参数(未配置项用全局默认)。
func (d *Dispatcher) ParamsFor(carrierID uint) CarrierParams {
d.mu.Lock()
row, ok := d.params[carrierID]
d.mu.Unlock()
return MergeParams(row, ok, d.cfg)
}
// MergeParams 合并账户配置行与全局默认。
func MergeParams(row model.CarrierPollingParam, configured bool, cfg config.CarrierSchedulerConfig) CarrierParams {
params := CarrierParams{InflightLimit: cfg.DefaultInflightLimit, CardMaxInFlight: cfg.DefaultCardMaxInFlight, Configured: configured}
if params.InflightLimit <= 0 {
params.InflightLimit = 20
}
if params.CardMaxInFlight <= 0 {
params.CardMaxInFlight = 20
}
if !configured {
return params
}
if row.InflightLimit != nil && *row.InflightLimit > 0 {
params.InflightLimit = *row.InflightLimit
}
if row.CardMaxInFlight != nil && *row.CardMaxInFlight > 0 {
params.CardMaxInFlight = *row.CardMaxInFlight
}
if row.PriorityMaxRatio != nil && *row.PriorityMaxRatio > 0 {
params.PriorityRatio = *row.PriorityMaxRatio
}
return params
}
// drainLegacyHints 消费旧优先提示通道与手动触发队列:优先提示放入账户优先队列,手动触发在普通队列中立即到期。
func (d *Dispatcher) drainLegacyHints(ctx context.Context) {
for _, checkType := range constants.AllCheckTypes() {
taskType := constants.TaskTypeFromCheckType(checkType)
d.drainList(ctx, constants.RedisPollingPriorityQueueKey(taskType), checkType, false)
d.drainList(ctx, constants.RedisPollingManualQueueKey(taskType), checkType, true)
}
}
func (d *Dispatcher) drainList(ctx context.Context, key, checkType string, normal bool) {
for i := 0; i < 500; i++ {
raw, err := d.store.redis.LPop(ctx, key).Result()
if err != nil {
if !stderrors.Is(err, redis.Nil) {
d.logger.Warn("读取旧提示队列失败", zap.String("key", key), zap.Error(err))
}
return
}
cardID, convErr := strconv.ParseUint(raw, 10, 64)
if convErr != nil {
continue
}
due, dueErr := d.store.ReadDue(ctx, uint(cardID))
if dueErr != nil || due.CarrierID == 0 {
continue
}
if _, err := d.store.RequestCheck(ctx, CheckRequest{CardID: uint(cardID), CarrierID: due.CarrierID, CheckTypes: []string{checkType}, Normal: normal}); err != nil {
d.logger.Warn("写入账户队列失败", zap.Uint64("card_id", cardID), zap.Error(err))
}
}
}
// 新结构初始化状态(constants.RedisCarrierSchedulerInitStateKey)。
const (
InitStateRunning = "running"
InitStateDone = "done"
)

View File

@@ -182,7 +182,8 @@ return 1
// 只作用于仍在轮询中的卡(在优先或普通队列中)。冷却期内(最近成功距今未超过冷却)的类型不请求。
//
// KEYS[1] due;KEYS[2] 优先队列;KEYS[3] 普通队列;KEYS[4] claim。
// ARGV:now_ms、card_id、cooldown_ms(0 不冷却)、确认延迟(秒,逗号分隔,可空)、期望结果(可空)、类型…。
// ARGV:now_ms、card_id、cooldown_ms(0 不冷却)、确认延迟(秒,逗号分隔,可空)、期望结果(可空)、
// 是否放入优先队列(1/0;0 用于手动触发:只把到期设为现在并在普通队列中立即到期)、类型…。
// 返回本次实际请求的类型数;卡不在轮询中返回 -1。
const requestCheckLua = `
local now = tonumber(ARGV[1])
@@ -192,7 +193,8 @@ local inPriority = redis.call('ZSCORE', KEYS[2], cardId)
local inQueue = redis.call('ZSCORE', KEYS[3], cardId)
if not inPriority and not inQueue then return -1 end
local requested = 0
for i = 6, #ARGV do
local toPriority = ARGV[6] == '1'
for i = 7, #ARGV do
local t = ARGV[i]
local skip = false
if cooldown > 0 then
@@ -209,6 +211,11 @@ for i = 6, #ARGV do
end
end
if requested == 0 then return 0 end
if not toPriority then
if redis.call('EXISTS', KEYS[4]) == 1 then return requested end
if inQueue then redis.call('ZADD', KEYS[3], now, cardId) else redis.call('ZADD', KEYS[2], now, cardId) end
return requested
end
redis.call('HSET', KEYS[1], 'prio', '1')
if redis.call('EXISTS', KEYS[4]) == 1 then return requested end
redis.call('ZREM', KEYS[3], cardId)

View File

@@ -10,7 +10,9 @@ import (
"sync"
"time"
"github.com/redis/go-redis/v9"
"go.uber.org/zap"
"gorm.io/gorm"
"github.com/break/junhong_cmp_fiber/internal/infrastructure/systemconfig"
"github.com/break/junhong_cmp_fiber/pkg/constants"
@@ -376,3 +378,15 @@ func (p *SettingsProvider) load(ctx context.Context) (Settings, error) {
}
return s, nil
}
// NewDefaultSettingsProvider 用独立注册表与读取器创建调度参数提供者(API 与 Worker 共用同一份数据库与缓存数据)。
func NewDefaultSettingsProvider(db *gorm.DB, redisClient *redis.Client, logger *zap.Logger) *SettingsProvider {
registry := systemconfig.NewRegistry()
RegisterConfigDefinitions(registry, logger)
var cache systemconfig.Cache
if redisClient != nil {
cache = systemconfig.NewRedisCache(redisClient)
}
reader := systemconfig.NewReader(db, registry, cache, systemconfig.NewLogAlertSink(logger))
return NewSettingsProvider(reader, logger)
}

View File

@@ -136,6 +136,8 @@ type CheckRequest struct {
Cooldown time.Duration
ConfirmDelays []time.Duration
Expected string // 延迟确认的期望结果(card_status 为 online / offline),满足时取消后续确认
// Normal 为 true 时不放入优先队列:只把到期设为现在、在普通队列中立即到期(手动触发走普通通道)。
Normal bool
}
// RequestCheck 把卡的指定检查类型设为立即到期放入优先队列;返回实际请求的类型数,卡不在轮询中返回 -1。
@@ -147,7 +149,11 @@ func (s *Store) RequestCheck(ctx context.Context, request CheckRequest) (int64,
for _, delay := range request.ConfirmDelays {
confirm = append(confirm, strconv.FormatInt(int64(delay/time.Second), 10))
}
args := []any{time.Now().UnixMilli(), request.CardID, request.Cooldown.Milliseconds(), strings.Join(confirm, ","), request.Expected}
toPriority := "1"
if request.Normal {
toPriority = "0"
}
args := []any{time.Now().UnixMilli(), request.CardID, request.Cooldown.Milliseconds(), strings.Join(confirm, ","), request.Expected, toPriority}
for _, checkType := range request.CheckTypes {
args = append(args, checkType)
}

View File

@@ -22,6 +22,18 @@ type PollingLifecycleService struct {
deviceBindingStore *postgres.DeviceSimBindingStore
deviceStore *postgres.DeviceStore
logger *zap.Logger
// carrier 按运营商账户调度接缝;当前模式为 carrier 时生命周期事件写入账户队列,不写旧分片队列。
carrier *CarrierScheduling
}
// SetCarrierScheduling 注入按运营商账户调度接缝。
func (s *PollingLifecycleService) SetCarrierScheduling(carrier *CarrierScheduling) {
s.carrier = carrier
}
// carrierActive 判断当前是否按运营商账户调度。
func (s *PollingLifecycleService) carrierActive(ctx context.Context) bool {
return s.carrier.Active(ctx)
}
// NewPollingLifecycleService 创建卡生命周期轮询管理服务
@@ -52,6 +64,10 @@ func (s *PollingLifecycleService) OnCardCreated(ctx context.Context, card *model
s.logger.Debug("卡禁用轮询,跳过入队", zap.Uint("card_id", card.ID))
return
}
if s.carrierActive(ctx) {
s.carrier.UpsertCard(ctx, card, false)
return
}
s.enqueueCard(ctx, card)
}
@@ -64,6 +80,11 @@ func (s *PollingLifecycleService) OnBatchCardsCreated(ctx context.Context, cards
// OnCardStatusChanged 卡状态变化后清理缓存、重新匹配配置并更新队列
func (s *PollingLifecycleService) OnCardStatusChanged(ctx context.Context, cardID uint) {
if s.carrierActive(ctx) {
s.queueMgr.InvalidateCardCache(ctx, cardID)
s.carrierRefresh(ctx, cardID)
return
}
if err := s.queueMgr.RemoveFromAllQueues(ctx, cardID); err != nil {
s.logger.Warn("卡状态变化:从队列移除失败", zap.Uint("card_id", cardID), zap.Error(err))
}
@@ -87,6 +108,13 @@ func (s *PollingLifecycleService) OnBatchCardsStatusChanged(ctx context.Context,
if len(cardIDs) == 0 {
return
}
if s.carrierActive(ctx) {
for _, cardID := range cardIDs {
s.queueMgr.InvalidateCardCache(ctx, cardID)
s.carrierRefresh(ctx, cardID)
}
return
}
// 单次 pipeline:批量从当前分片队列移除并清理缓存
if err := s.queueMgr.RemoveBatchFromCurrentShardQueues(ctx, cardIDs); err != nil {
s.logger.Warn("批量卡状态变化:从队列移除失败", zap.Int("count", len(cardIDs)), zap.Error(err))
@@ -107,6 +135,11 @@ func (s *PollingLifecycleService) OnBatchCardsStatusChanged(ctx context.Context,
// OnCardDeleted 卡删除后移除所有队列并清理缓存
func (s *PollingLifecycleService) OnCardDeleted(ctx context.Context, cardID uint) {
if s.carrierActive(ctx) {
s.carrier.RemoveCard(ctx, cardID, 0)
s.queueMgr.InvalidateCardCache(ctx, cardID)
return
}
if err := s.queueMgr.OnCardDeleted(ctx, cardID); err != nil {
s.logger.Warn("卡删除:清理队列和缓存失败", zap.Uint("card_id", cardID), zap.Error(err))
}
@@ -123,11 +156,21 @@ func (s *PollingLifecycleService) OnCardEnabled(ctx context.Context, cardID uint
if !s.ShouldEnqueue(ctx, card) {
return
}
if s.carrierActive(ctx) {
// 恢复轮询时各检查类型到期设为当前时间加抖动。
s.carrier.UpsertCard(ctx, card, false)
return
}
s.enqueueCard(ctx, card)
}
// OnCardDisabled 卡禁用轮询后移除所有队列
func (s *PollingLifecycleService) OnCardDisabled(ctx context.Context, cardID uint) {
if s.carrierActive(ctx) {
s.carrier.RemoveCard(ctx, cardID, 0)
s.queueMgr.InvalidateCardCache(ctx, cardID)
return
}
if err := s.queueMgr.RemoveFromAllQueues(ctx, cardID); err != nil {
s.logger.Warn("卡禁用:从队列移除失败", zap.Uint("card_id", cardID), zap.Error(err))
}
@@ -183,3 +226,17 @@ func calcInitialDelay(interval int) time.Time {
}
return now.Add(time.Duration(rand.Intn(jitterMax)) * time.Second)
}
// carrierRefresh 按运营商账户调度时重新匹配配置并写入账户队列:不再轮询的卡移出,其余按当前时间加抖动重排。
func (s *PollingLifecycleService) carrierRefresh(ctx context.Context, cardID uint) {
card, err := s.iotCardStore.GetByID(ctx, cardID)
if err != nil {
s.logger.Error("卡状态变化:加载卡信息失败", zap.Uint("card_id", cardID), zap.Error(err))
return
}
if !s.ShouldEnqueue(ctx, card) {
s.carrier.RemoveCard(ctx, card.ID, card.CarrierID)
return
}
s.carrier.UpsertCard(ctx, card, false)
}

View File

@@ -30,6 +30,8 @@ type Scheduler struct {
packageActivationHandler *PackageActivationHandler
dataResetHandler *DataResetHandler
initializer *PollingInitializer // 可选,nil 表示不守卫 Init 完成
// carrier 按运营商账户调度接缝:当前模式为 carrier 时旧分片调度停止出队(套餐激活检查照常执行)。
carrier *CarrierScheduling
stopChan chan struct{}
wg sync.WaitGroup
@@ -156,6 +158,10 @@ func (s *Scheduler) scheduleLoop(ctx context.Context) {
// processShardSchedule 处理手动队列和分片定时队列(每 1 秒触发)
// 使用 90% 的 tick 间隔作为超时,确保单分片 Redis 挂起时不阻塞下一个 tick
func (s *Scheduler) processShardSchedule(ctx context.Context) {
if s.carrier.Active(ctx) {
// 已切到按运营商账户调度:旧分片调度停止出队,优先提示与手动触发由新分发器消费。
return
}
// 优先提示与手动队列都不受 Init 影响(ConfigManager 已就绪即可)。
// 同一调度周期内先排空优先提示、再排空手动触发队列:优先的强度边界仅为调度入口领先,
// 提示排空不参与分片背压判断,因此分片积压不会抑制加急入队。
@@ -296,3 +302,8 @@ func (s *Scheduler) enqueueBatch(ctx context.Context, taskType string, cardIDs [
}
}
}
// SetCarrierScheduling 注入按运营商账户调度接缝。
func (s *Scheduler) SetCarrierScheduling(carrier *CarrierScheduling) {
s.carrier = carrier
}