Fix err (#2608)

* refactor: refactor workflows contents.

* add tool workflows.

* update field.

* fix: remove chat error.

* Fix err.

* fix error.

* remove cn comment.

* update workflows files.

* update infra config.

* move workflows.

* feat: update bot.

* fix: solve uncorrect outdated msg get.

* update get docIDs logic.

* update

* update skip logic.

* fix

* update.

* fix: delay deleteObject func.

* remove unused content.

* update log type.

* feat: implement request batch count limit.

* update

* update

* feat: add rocksTimeout

* feat: wrap logs

* feat: add logs

* feat: listen config

* feat: enable listen TIME_WAIT port

* feat: add logs

* feat: cache batch

* chore: enable fullUserCache

* feat: push rpc num

* feat: push err

* feat: with operationID

* feat: sleep

* feat: change 1s

* feat: change log

* feat: implement Getbatch in rpcCache.

* feat: print getOnline cost

* feat: change log

* feat: change kafka and push config

* feat: del interface

* feat: fix err

* feat: change config

* feat: go mod

* feat: change config

* feat: change config

* feat: add sleep in push

* feat: warn logs

* feat: logs

* feat: logs

* feat: change port

* feat: start config

* feat: remove port reuse

* feat: prometheus config

* feat: prometheus config

* feat: prometheus config

* feat: add long time send msg to grafana

* feat: init

* feat: init

* feat: implement offline push.

* feat: batch get user online

* feat: implement batch Push spilt

* update go mod

* Revert "feat: change port"

This reverts commit 06d5e944

* feat: change port

* feat: change config

* feat: implement kafka producer and consumer.

* update format,

* add PushMQ log.

* feat: get all online users and init push

* feat: lock in online cache

* feat: config

* fix: init online status

* fix: add logs

* fix: userIDs

* fix: add logs

* feat: update Handler logic.

* update MQ logic.

* update

* update

* fix: method name

* fix: update OfflinePushConsumerHandler.

* fix: prommetrics

* fix: add logs

* fix: ctx

* fix: log

* fix: config

* feat: change port

* fix: atomic online cache status

---------

Co-authored-by: Monet Lee <monet_lee@163.com>

pkg/common/config/config.go

@@ -224,6 +224,7 @@ type Push struct {
 		BadgeCount bool   `mapstructure:"badgeCount"`
 		Production bool   `mapstructure:"production"`
 	} `mapstructure:"iosPush"`
+	FullUserCache bool `mapstructure:"fullUserCache"`
 }
 
 type Auth struct {

pkg/common/startrpc/start.go

@@ -54,15 +54,11 @@ func Start[T any](ctx context.Context, discovery *config.Discovery, prometheusCo
 	log.CInfo(ctx, "RPC server is initializing", "rpcRegisterName", rpcRegisterName, "rpcPort", rpcPort,
 		"prometheusPorts", prometheusConfig.Ports)
 	rpcTcpAddr := net.JoinHostPort(network.GetListenIP(listenIP), strconv.Itoa(rpcPort))
+
 	listener, err := net.Listen(
 		"tcp",
 		rpcTcpAddr,
 	)
-	if err != nil {
-		return errs.WrapMsg(err, "listen err", "rpcTcpAddr", rpcTcpAddr)
-	}
-
-	defer listener.Close()
 	client, err := kdisc.NewDiscoveryRegister(discovery, share)
 	if err != nil {
 		return err

pkg/common/storage/cache/cachekey/online.go

@@ -1,6 +1,9 @@
 package cachekey
 
-import "time"
+import (
+	"strings"
+	"time"
+)
 
 const (
 	OnlineKey     = "ONLINE:"
@@ -11,3 +14,7 @@ const (
 func GetOnlineKey(userID string) string {
 	return OnlineKey + userID
 }
+
+func GetOnlineKeyUserID(key string) string {
+	return strings.TrimPrefix(key, OnlineKey)
+}

pkg/common/storage/cache/online.go

@@ -5,4 +5,5 @@ import "context"
 type OnlineCache interface {
 	GetOnline(ctx context.Context, userID string) ([]int32, error)
 	SetUserOnline(ctx context.Context, userID string, online, offline []int32) error
+	GetAllOnlineUsers(ctx context.Context, cursor uint64) (map[string][]int32, uint64, error)
 }

pkg/common/storage/cache/redis/batch.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"encoding/json"
 	"github.com/dtm-labs/rockscache"
+	"github.com/openimsdk/tools/errs"
 	"github.com/openimsdk/tools/log"
 	"github.com/redis/go-redis/v9"
 	"golang.org/x/sync/singleflight"
@@ -65,6 +66,7 @@ func batchGetCache2[K comparable, V any](ctx context.Context, rcClient *rockscac
 				}
 				bs, err := json.Marshal(value)
 				if err != nil {
+					log.ZError(ctx, "marshal failed", err)
 					return nil, err
 				}
 				cacheIndex[index] = string(bs)
@@ -72,7 +74,7 @@ func batchGetCache2[K comparable, V any](ctx context.Context, rcClient *rockscac
 			return cacheIndex, nil
 		})
 		if err != nil {
-			return nil, err
+			return nil, errs.WrapMsg(err, "FetchBatch2 failed")
 		}
 		for index, data := range indexCache {
 			if data == "" {
@@ -80,7 +82,7 @@ func batchGetCache2[K comparable, V any](ctx context.Context, rcClient *rockscac
 			}
 			var value V
 			if err := json.Unmarshal([]byte(data), &value); err != nil {
-				return nil, err
+				return nil, errs.WrapMsg(err, "Unmarshal failed")
 			}
 			if cb, ok := any(&value).(BatchCacheCallback[K]); ok {
 				cb.BatchCache(keyId[keys[index]])

pkg/common/storage/cache/redis/batch_handler.go

@@ -28,6 +28,10 @@ import (
 	"time"
 )
 
+const (
+	rocksCacheTimeout = 11 * time.Second
+)
+
 // BatchDeleterRedis is a concrete implementation of the BatchDeleter interface based on Redis and RocksCache.
 type BatchDeleterRedis struct {
 	redisClient    redis.UniversalClient
@@ -106,6 +110,8 @@ func (c *BatchDeleterRedis) AddKeys(keys ...string) {
 // GetRocksCacheOptions returns the default configuration options for RocksCache.
 func GetRocksCacheOptions() *rockscache.Options {
 	opts := rockscache.NewDefaultOptions()
+	opts.LockExpire = rocksCacheTimeout
+	opts.WaitReplicasTimeout = rocksCacheTimeout
 	opts.StrongConsistency = true
 	opts.RandomExpireAdjustment = 0.2
 

pkg/common/storage/cache/redis/online.go

@@ -2,8 +2,10 @@ package redis
 
 import (
 	"context"
+	"fmt"
 	"github.com/openimsdk/open-im-server/v3/pkg/common/storage/cache"
 	"github.com/openimsdk/open-im-server/v3/pkg/common/storage/cache/cachekey"
+	"github.com/openimsdk/protocol/constant"
 	"github.com/openimsdk/tools/errs"
 	"github.com/openimsdk/tools/log"
 	"github.com/redis/go-redis/v9"
@@ -49,6 +51,36 @@ func (s *userOnline) GetOnline(ctx context.Context, userID string) ([]int32, err
 	return platformIDs, nil
 }
 
+func (s *userOnline) GetAllOnlineUsers(ctx context.Context, cursor uint64) (map[string][]int32, uint64, error) {
+	result := make(map[string][]int32)
+
+	keys, nextCursor, err := s.rdb.Scan(ctx, cursor, fmt.Sprintf("%s*", cachekey.OnlineKey), constant.ParamMaxLength).Result()
+	if err != nil {
+		return nil, 0, err
+	}
+
+	for _, key := range keys {
+		userID := cachekey.GetOnlineKeyUserID(key)
+		strValues, err := s.rdb.ZRange(ctx, key, 0, -1).Result()
+		if err != nil {
+			return nil, 0, err
+		}
+
+		values := make([]int32, 0, len(strValues))
+		for _, value := range strValues {
+			intValue, err := strconv.Atoi(value)
+			if err != nil {
+				return nil, 0, errs.Wrap(err)
+			}
+			values = append(values, int32(intValue))
+		}
+
+		result[userID] = values
+	}
+
+	return result, nextCursor, nil
+}
+
 func (s *userOnline) SetUserOnline(ctx context.Context, userID string, online, offline []int32) error {
 	script := `
 	local key = KEYS[1]

pkg/localcache/lru/lru.go

@@ -20,7 +20,9 @@ type EvictCallback[K comparable, V any] simplelru.EvictCallback[K, V]
 
 type LRU[K comparable, V any] interface {
 	Get(key K, fetch func() (V, error)) (V, error)
+	Set(key K, value V)
 	SetHas(key K, value V) bool
+	GetBatch(keys []K, fetch func(keys []K) (map[K]V, error)) (map[K]V, error)
 	Del(key K) bool
 	Stop()
 }

pkg/localcache/lru/lru_expiration.go

@@ -51,6 +51,11 @@ type ExpirationLRU[K comparable, V any] struct {
 	target     Target
 }
 
+func (x *ExpirationLRU[K, V]) GetBatch(keys []K, fetch func(keys []K) (map[K]V, error)) (map[K]V, error) {
+	//TODO implement me
+	panic("implement me")
+}
+
 func (x *ExpirationLRU[K, V]) Get(key K, fetch func() (V, error)) (V, error) {
 	x.lock.Lock()
 	v, ok := x.core.Get(key)
@@ -99,5 +104,11 @@ func (x *ExpirationLRU[K, V]) SetHas(key K, value V) bool {
 	return false
 }
 
+func (x *ExpirationLRU[K, V]) Set(key K, value V) {
+	x.lock.Lock()
+	defer x.lock.Unlock()
+	x.core.Add(key, &expirationLruItem[V]{value: value})
+}
+
 func (x *ExpirationLRU[K, V]) Stop() {
 }

pkg/localcache/lru/lru_lazy.go

@@ -88,18 +88,76 @@ func (x *LayLRU[K, V]) Get(key K, fetch func() (V, error)) (V, error) {
 	return v.value, v.err
 }
 
-//func (x *LayLRU[K, V]) Set(key K, value V) {
-//	x.lock.Lock()
-//	x.core.Add(key, &layLruItem[V]{value: value, expires: time.Now().Add(x.successTTL).UnixMilli()})
-//	x.lock.Unlock()
-//}
-//
+func (x *LayLRU[K, V]) GetBatch(keys []K, fetch func(keys []K) (map[K]V, error)) (map[K]V, error) {
+	var (
+		err  error
+		once sync.Once
+	)
+
+	x.lock.Lock()
+	res := make(map[K]V)
+	queries := make([]K, 0)
+	setVs := make(map[K]*layLruItem[V])
+	for _, key := range keys {
+		v, ok := x.core.Get(key)
+		if ok {
+			x.lock.Unlock()
+			v.lock.Lock()
+			expires, value, err1 := v.expires, v.value, v.err
+			if expires != 0 && expires > time.Now().UnixMilli() {
+				v.lock.Unlock()
+				x.target.IncrGetHit()
+				res[key] = value
+				if err1 != nil {
+					once.Do(func() {
+						err = err1
+					})
+				}
+				continue
+			}
+		}
+		queries = append(queries, key)
+		x.lock.Unlock()
+	}
+	values, err1 := fetch(queries)
+	if err1 != nil {
+		once.Do(func() {
+			err = err1
+		})
+	}
+	for key, val := range values {
+		v := &layLruItem[V]{}
+		v.value = val
+
+		if err == nil {
+			v.expires = time.Now().Add(x.successTTL).UnixMilli()
+			x.target.IncrGetSuccess()
+		} else {
+			v.expires = time.Now().Add(x.failedTTL).UnixMilli()
+			x.target.IncrGetFailed()
+		}
+		setVs[key] = v
+		x.lock.Lock()
+		x.core.Add(key, v)
+		x.lock.Unlock()
+		res[key] = val
+	}
+
+	return res, err
+}
+
 //func (x *LayLRU[K, V]) Has(key K) bool {
 //	x.lock.Lock()
 //	defer x.lock.Unlock()
 //	return x.core.Contains(key)
 //}
 
+func (x *LayLRU[K, V]) Set(key K, value V) {
+	x.lock.Lock()
+	defer x.lock.Unlock()
+	x.core.Add(key, &layLruItem[V]{value: value, expires: time.Now().Add(x.successTTL).UnixMilli()})
+}
+
 func (x *LayLRU[K, V]) SetHas(key K, value V) bool {
 	x.lock.Lock()
 	defer x.lock.Unlock()

pkg/localcache/lru/lru_slot.go

@@ -32,6 +32,29 @@ type slotLRU[K comparable, V any] struct {
 	hash  func(k K) uint64
 }
 
+func (x *slotLRU[K, V]) GetBatch(keys []K, fetch func(keys []K) (map[K]V, error)) (map[K]V, error) {
+	var (
+		slotKeys = make(map[uint64][]K)
+		vs       = make(map[K]V)
+	)
+
+	for _, k := range keys {
+		index := x.getIndex(k)
+		slotKeys[index] = append(slotKeys[index], k)
+	}
+
+	for k, v := range slotKeys {
+		batches, err := x.slots[k].GetBatch(v, fetch)
+		if err != nil {
+			return nil, err
+		}
+		for key, value := range batches {
+			vs[key] = value
+		}
+	}
+	return vs, nil
+}
+
 func (x *slotLRU[K, V]) getIndex(k K) uint64 {
 	return x.hash(k) % x.n
 }
@@ -40,6 +63,10 @@ func (x *slotLRU[K, V]) Get(key K, fetch func() (V, error)) (V, error) {
 	return x.slots[x.getIndex(key)].Get(key, fetch)
 }
 
+func (x *slotLRU[K, V]) Set(key K, value V) {
+	x.slots[x.getIndex(key)].Set(key, value)
+}
+
 func (x *slotLRU[K, V]) SetHas(key K, value V) bool {
 	return x.slots[x.getIndex(key)].SetHas(key, value)
 }

pkg/rpccache/online.go

@@ -2,60 +2,197 @@ package rpccache
 
 import (
 	"context"
+	"fmt"
+	"github.com/openimsdk/protocol/constant"
+	"github.com/openimsdk/protocol/user"
+	"math/rand"
+	"strconv"
+	"sync"
+	"sync/atomic"
+	"time"
+
 	"github.com/openimsdk/open-im-server/v3/pkg/common/storage/cache/cachekey"
 	"github.com/openimsdk/open-im-server/v3/pkg/localcache"
 	"github.com/openimsdk/open-im-server/v3/pkg/localcache/lru"
 	"github.com/openimsdk/open-im-server/v3/pkg/rpcclient"
 	"github.com/openimsdk/open-im-server/v3/pkg/util/useronline"
+	"github.com/openimsdk/tools/db/cacheutil"
 	"github.com/openimsdk/tools/log"
 	"github.com/openimsdk/tools/mcontext"
 	"github.com/redis/go-redis/v9"
-	"math/rand"
-	"strconv"
-	"time"
 )
 
-func NewOnlineCache(user rpcclient.UserRpcClient, group *GroupLocalCache, rdb redis.UniversalClient, fn func(ctx context.Context, userID string, platformIDs []int32)) *OnlineCache {
+func NewOnlineCache(user rpcclient.UserRpcClient, group *GroupLocalCache, rdb redis.UniversalClient, fullUserCache bool, fn func(ctx context.Context, userID string, platformIDs []int32)) (*OnlineCache, error) {
+	l := &sync.Mutex{}
 	x := &OnlineCache{
-		user:  user,
-		group: group,
-		local: lru.NewSlotLRU(1024, localcache.LRUStringHash, func() lru.LRU[string, []int32] {
-			return lru.NewLayLRU[string, []int32](2048, cachekey.OnlineExpire/2, time.Second*3, localcache.EmptyTarget{}, func(key string, value []int32) {})
-		}),
+		user:          user,
+		group:         group,
+		fullUserCache: fullUserCache,
+		Lock:          l,
+		Cond:          sync.NewCond(l),
 	}
+
+	ctx := mcontext.SetOperationID(context.TODO(), strconv.FormatInt(time.Now().UnixNano()+int64(rand.Uint32()), 10))
+
+	switch x.fullUserCache {
+	case true:
+		log.ZDebug(ctx, "fullUserCache is true")
+		x.mapCache = cacheutil.NewCache[string, []int32]()
+		go func() {
+			if err := x.initUsersOnlineStatus(ctx); err != nil {
+				log.ZError(ctx, "initUsersOnlineStatus failed", err)
+			}
+		}()
+	case false:
+		log.ZDebug(ctx, "fullUserCache is false")
+		x.lruCache = lru.NewSlotLRU(1024, localcache.LRUStringHash, func() lru.LRU[string, []int32] {
+			return lru.NewLayLRU[string, []int32](2048, cachekey.OnlineExpire/2, time.Second*3, localcache.EmptyTarget{}, func(key string, value []int32) {})
+		})
+		x.CurrentPhase.Store(DoSubscribeOver)
+		x.Cond.Broadcast()
+	}
+
 	go func() {
-		ctx := mcontext.SetOperationID(context.Background(), cachekey.OnlineChannel+strconv.FormatUint(rand.Uint64(), 10))
-		for message := range rdb.Subscribe(ctx, cachekey.OnlineChannel).Channel() {
-			userID, platformIDs, err := useronline.ParseUserOnlineStatus(message.Payload)
-			if err != nil {
-				log.ZError(ctx, "OnlineCache setUserOnline redis subscribe parseUserOnlineStatus", err, "payload", message.Payload, "channel", message.Channel)
-				continue
-			}
-			storageCache := x.setUserOnline(userID, platformIDs)
-			log.ZDebug(ctx, "OnlineCache setUserOnline", "userID", userID, "platformIDs", platformIDs, "payload", message.Payload, "storageCache", storageCache)
-			if fn != nil {
-				fn(ctx, userID, platformIDs)
-			}
-		}
+		x.doSubscribe(ctx, rdb, fn)
 	}()
-	return x
+	return x, nil
 }
 
+const (
+	Begin uint32 = iota
+	DoOnlineStatusOver
+	DoSubscribeOver
+)
+
 type OnlineCache struct {
 	user  rpcclient.UserRpcClient
 	group *GroupLocalCache
-	local lru.LRU[string, []int32]
+
+	// fullUserCache if enabled, caches the online status of all users using mapCache;
+	// otherwise, only a portion of users' online statuses (regardless of whether they are online) will be cached using lruCache.
+	fullUserCache bool
+
+	lruCache lru.LRU[string, []int32]
+	mapCache *cacheutil.Cache[string, []int32]
+
+	Lock         *sync.Mutex
+	Cond         *sync.Cond
+	CurrentPhase atomic.Uint32
+}
+
+func (o *OnlineCache) initUsersOnlineStatus(ctx context.Context) (err error) {
+	log.ZDebug(ctx, "init users online status begin")
+
+	var (
+		totalSet      atomic.Int64
+		maxTries      = 5
+		retryInterval = time.Second * 5
+
+		resp *user.GetAllOnlineUsersResp
+	)
+
+	defer func(t time.Time) {
+		log.ZInfo(ctx, "init users online status end", "cost", time.Since(t), "totalSet", totalSet.Load())
+		o.CurrentPhase.Store(DoOnlineStatusOver)
+		o.Cond.Broadcast()
+	}(time.Now())
+
+	retryOperation := func(operation func() error, operationName string) error {
+		for i := 0; i < maxTries; i++ {
+			if err = operation(); err != nil {
+				log.ZWarn(ctx, fmt.Sprintf("initUsersOnlineStatus: %s failed", operationName), err)
+				time.Sleep(retryInterval)
+			} else {
+				return nil
+			}
+		}
+		return err
+	}
+
+	cursor := uint64(0)
+	for resp == nil || resp.NextCursor != 0 {
+		if err = retryOperation(func() error {
+			resp, err = o.user.GetAllOnlineUsers(ctx, cursor)
+			if err != nil {
+				return err
+			}
+
+			for _, u := range resp.StatusList {
+				if u.Status == constant.Online {
+					o.setUserOnline(u.UserID, u.PlatformIDs)
+				}
+				totalSet.Add(1)
+			}
+			cursor = resp.NextCursor
+			return nil
+		}, "getAllOnlineUsers"); err != nil {
+			return err
+		}
+	}
+
+	return nil
+}
+
+func (o *OnlineCache) doSubscribe(ctx context.Context, rdb redis.UniversalClient, fn func(ctx context.Context, userID string, platformIDs []int32)) {
+	o.Lock.Lock()
+	ch := rdb.Subscribe(ctx, cachekey.OnlineChannel).Channel()
+	for o.CurrentPhase.Load() < DoOnlineStatusOver {
+		o.Cond.Wait()
+	}
+	o.Lock.Unlock()
+	log.ZInfo(ctx, "begin doSubscribe")
+
+	doMessage := func(message *redis.Message) {
+		userID, platformIDs, err := useronline.ParseUserOnlineStatus(message.Payload)
+		if err != nil {
+			log.ZError(ctx, "OnlineCache setHasUserOnline redis subscribe parseUserOnlineStatus", err, "payload", message.Payload, "channel", message.Channel)
+			return
+		}
+		log.ZDebug(ctx, fmt.Sprintf("get subscribe %s message", cachekey.OnlineChannel), "useID", userID, "platformIDs", platformIDs)
+		switch o.fullUserCache {
+		case true:
+			if len(platformIDs) == 0 {
+				// offline
+				o.mapCache.Delete(userID)
+			} else {
+				o.mapCache.Store(userID, platformIDs)
+			}
+		case false:
+			storageCache := o.setHasUserOnline(userID, platformIDs)
+			log.ZDebug(ctx, "OnlineCache setHasUserOnline", "userID", userID, "platformIDs", platformIDs, "payload", message.Payload, "storageCache", storageCache)
+			if fn != nil {
+				fn(ctx, userID, platformIDs)
+			}
+		}
+	}
+
+	if o.CurrentPhase.Load() == DoOnlineStatusOver {
+		for done := false; !done; {
+			select {
+			case message := <-ch:
+				doMessage(message)
+			default:
+				o.CurrentPhase.Store(DoSubscribeOver)
+				o.Cond.Broadcast()
+				done = true
+			}
+		}
+	}
+
+	for message := range ch {
+		doMessage(message)
+	}
 }
 
 func (o *OnlineCache) getUserOnlinePlatform(ctx context.Context, userID string) ([]int32, error) {
-	platformIDs, err := o.local.Get(userID, func() ([]int32, error) {
+	platformIDs, err := o.lruCache.Get(userID, func() ([]int32, error) {
 		return o.user.GetUserOnlinePlatform(ctx, userID)
 	})
 	if err != nil {
 		log.ZError(ctx, "OnlineCache GetUserOnlinePlatform", err, "userID", userID)
 		return nil, err
 	}
-	log.ZDebug(ctx, "OnlineCache GetUserOnlinePlatform", "userID", userID, "platformIDs", platformIDs)
+	//log.ZDebug(ctx, "OnlineCache GetUserOnlinePlatform", "userID", userID, "platformIDs", platformIDs)
 	return platformIDs, nil
 }
 
@@ -69,6 +206,16 @@ func (o *OnlineCache) GetUserOnlinePlatform(ctx context.Context, userID string)
 	return platformIDs, nil
 }
 
+// func (o *OnlineCache) GetUserOnlinePlatformBatch(ctx context.Context, userIDs []string) (map[string]int32, error) {
+// 	platformIDs, err := o.getUserOnlinePlatform(ctx, userIDs)
+// 	if err != nil {
+// 		return nil, err
+// 	}
+// 	tmp := make([]int32, len(platformIDs))
+// 	copy(tmp, platformIDs)
+// 	return platformIDs, nil
+// }
+
 func (o *OnlineCache) GetUserOnline(ctx context.Context, userID string) (bool, error) {
 	platformIDs, err := o.getUserOnlinePlatform(ctx, userID)
 	if err != nil {
@@ -77,10 +224,68 @@ func (o *OnlineCache) GetUserOnline(ctx context.Context, userID string) (bool, e
 	return len(platformIDs) > 0, nil
 }
 
+func (o *OnlineCache) getUserOnlinePlatformBatch(ctx context.Context, userIDs []string) (map[string][]int32, error) {
+	platformIDsMap, err := o.lruCache.GetBatch(userIDs, func(missingUsers []string) (map[string][]int32, error) {
+		platformIDsMap := make(map[string][]int32)
+
+		usersStatus, err := o.user.GetUsersOnlinePlatform(ctx, missingUsers)
+		if err != nil {
+			return nil, err
+		}
+
+		for _, u := range usersStatus {
+			platformIDsMap[u.UserID] = u.PlatformIDs
+		}
+
+		return platformIDsMap, nil
+	})
+	if err != nil {
+		log.ZError(ctx, "OnlineCache GetUserOnlinePlatform", err, "userID", userIDs)
+		return nil, err
+	}
+	return platformIDsMap, nil
+}
+
+func (o *OnlineCache) GetUsersOnline(ctx context.Context, userIDs []string) ([]string, []string, error) {
+	t := time.Now()
+
+	var (
+		onlineUserIDs  = make([]string, 0, len(userIDs))
+		offlineUserIDs = make([]string, 0, len(userIDs))
+	)
+
+	switch o.fullUserCache {
+	case true:
+		for _, userID := range userIDs {
+			if _, ok := o.mapCache.Load(userID); ok {
+				onlineUserIDs = append(onlineUserIDs, userID)
+			} else {
+				offlineUserIDs = append(offlineUserIDs, userID)
+			}
+		}
+	case false:
+		userOnlineMap, err := o.getUserOnlinePlatformBatch(ctx, userIDs)
+		if err != nil {
+			return nil, nil, err
+		}
+
+		for key, value := range userOnlineMap {
+			if len(value) > 0 {
+				onlineUserIDs = append(onlineUserIDs, key)
+			} else {
+				offlineUserIDs = append(offlineUserIDs, key)
+			}
+		}
+	}
+
+	log.ZInfo(ctx, "get users online", "online users length", len(userIDs), "offline users length", len(offlineUserIDs), "cost", time.Since(t))
+	return userIDs, offlineUserIDs, nil
+}
+
 //func (o *OnlineCache) GetUsersOnline(ctx context.Context, userIDs []string) ([]string, error) {
 //	onlineUserIDs := make([]string, 0, len(userIDs))
 //	for _, userID := range userIDs {
-//		online, err := o.GetUserOnline(ctx, userID)
+// online, err := o.GetUserOnline(ctx, userID)
 //		if err != nil {
 //			return nil, err
 //		}
@@ -111,6 +316,15 @@ func (o *OnlineCache) GetUserOnline(ctx context.Context, userID string) (bool, e
 //	return onlineUserIDs, nil
 //}
 
-func (o *OnlineCache) setUserOnline(userID string, platformIDs []int32) bool {
-	return o.local.SetHas(userID, platformIDs)
+func (o *OnlineCache) setUserOnline(userID string, platformIDs []int32) {
+	switch o.fullUserCache {
+	case true:
+		o.mapCache.Store(userID, platformIDs)
+	case false:
+		o.lruCache.Set(userID, platformIDs)
+	}
+}
+
+func (o *OnlineCache) setHasUserOnline(userID string, platformIDs []int32) bool {
+	return o.lruCache.SetHas(userID, platformIDs)
 }

pkg/rpcclient/user.go

@@ -169,6 +169,15 @@ func (u *UserRpcClient) Access(ctx context.Context, ownerUserID string) error {
 	return authverify.CheckAccessV3(ctx, ownerUserID, u.imAdminUserID)
 }
 
+// GetAllUserID retrieves all user IDs with pagination options.
+func (u *UserRpcClient) GetAllUserID(ctx context.Context, pageNumber, showNumber int32) (*user.GetAllUserIDResp, error) {
+	resp, err := u.Client.GetAllUserID(ctx, &user.GetAllUserIDReq{Pagination: &sdkws.RequestPagination{PageNumber: pageNumber, ShowNumber: showNumber}})
+	if err != nil {
+		return nil, err
+	}
+	return resp, nil
+}
+
 // GetAllUserIDs retrieves all user IDs with pagination options.
 func (u *UserRpcClient) GetAllUserIDs(ctx context.Context, pageNumber, showNumber int32) ([]string, error) {
 	resp, err := u.Client.GetAllUserID(ctx, &user.GetAllUserIDReq{Pagination: &sdkws.RequestPagination{PageNumber: pageNumber, ShowNumber: showNumber}})
@@ -215,3 +224,7 @@ func (u *UserRpcClient) GetUserOnlinePlatform(ctx context.Context, userID string
 	}
 	return resp[0].PlatformIDs, nil
 }
+
+func (u *UserRpcClient) GetAllOnlineUsers(ctx context.Context, cursor uint64) (*user.GetAllOnlineUsersResp, error) {
+	return u.Client.GetAllOnlineUsers(ctx, &user.GetAllOnlineUsersReq{Cursor: cursor})
+}