[redis] maxmemory 数据淘汰策略

[redis] maxmemory 数据淘汰策略

redis是内存数据库，通过redis.conf配置最大内存暂用（当进行内存分配时，会统计已使用内存，在zmalloc接口中有实现）。当redis使用的内存超过限制时则会触发数据淘汰机制，直到当前内存使用量小于阈值。

maxmemory <bytes>

当redis使用的内存超过限制时，redis有两种方案释放内存：

1. 数据库中的所有kv对
2. 将要到期的kv对

淘汰的策略包括：

1. 随机淘汰
2. 先淘汰到期的或快要到期的数据
3. 近似LRU
4. 近似LFU

debug

修改redis.conf的maxmemory，在server.c line3755处打断点。

if (server.maxmemory && !isInsideYieldingLongCommand()) {
	// ...
}

调用栈为：

performEvictions()
processCommand(client *c)
...

淘汰数据的逻辑是:

redis淘汰策略

源码位置evict.c

采样淘汰

	if (server.maxmemory_policy & (MAXMEMORY_FLAG_LRU|MAXMEMORY_FLAG_LFU) ||
            server.maxmemory_policy == MAXMEMORY_VOLATILE_TTL)
        {
            struct evictionPoolEntry *pool = EvictionPoolLRU;

            while(bestkey == NULL) {
                unsigned long total_keys = 0, keys;

                /* We don't want to make local-db choices when expiring keys,
                 * so to start populate the eviction pool sampling keys from
                 * every DB. */
                for (i = 0; i < server.dbnum; i++) {
                    db = server.db+i;
                    dict = (server.maxmemory_policy & MAXMEMORY_FLAG_ALLKEYS) ?
                            db->dict : db->expires;
                    if ((keys = dictSize(dict)) != 0) {
                        evictionPoolPopulate(i, dict, db->dict, pool);
                        total_keys += keys;
                    }
                }
                if (!total_keys) break; /* No keys to evict. */

                /* Go backward from best to worst element to evict. */
                for (k = EVPOOL_SIZE-1; k >= 0; k--) {
                    if (pool[k].key == NULL) continue;
                    bestdbid = pool[k].dbid;

                    if (server.maxmemory_policy & MAXMEMORY_FLAG_ALLKEYS) {
                        de = dictFind(server.db[bestdbid].dict,
                            pool[k].key);
                    } else {
                        de = dictFind(server.db[bestdbid].expires,
                            pool[k].key);
                    }

                    /* Remove the entry from the pool. */
                    if (pool[k].key != pool[k].cached)
                        sdsfree(pool[k].key);
                    pool[k].key = NULL;
                    pool[k].idle = 0;

                    /* If the key exists, is our pick. Otherwise it is
                     * a ghost and we need to try the next element. */
                    if (de) {
                        bestkey = dictGetKey(de);
                        break;
                    } else {
                        /* Ghost... Iterate again. */
                    }
                }
            }
        }

我们从dict或者expire字典里采样key，然后插入到EvictionPoolLRU这个global池子中，然后遍历EvictionPoolLRU池选择idle最大的element淘汰（同步or异步淘汰）。
这个操作会在以下条件停止：
while (mem_freed < (long long)mem_tofree)
循环时间较长，为了避免因为一直淘汰元素占用cpu资源，在event loop中注册startEvictionTimeProc。

一致性

在我们淘汰元素时，如果是主从架构，需要将删除动作传播到slave服务器。本质上就是写入aof文件

/* Propagate expires into slaves and the AOF file.
 * When a key expires in the master, a DEL operation for this key is sent
 * to all the slaves and the AOF file if enabled.
 *
 * This way the key expiry is centralized in one place, and since both
 * AOF and the master->slave link guarantee operation ordering, everything
 * will be consistent even if we allow write operations against expiring
 * keys.
 *
 * This function may be called from:
 * 1. Within call(): Example: Lazy-expire on key access.
 *    In this case the caller doesn't have to do anything
 *    because call() handles server.also_propagate(); or
 * 2. Outside of call(): Example: Active-expire, eviction.
 *    In this the caller must remember to call
 *    propagatePendingCommands, preferably at the end of
 *    the deletion batch, so that DELs will be wrapped
 *    in MULTI/EXEC */
void propagateDeletion(redisDb *db, robj *key, int lazy) {
    robj *argv[2];

    argv[0] = lazy ? shared.unlink : shared.del;
    argv[1] = key;
    incrRefCount(argv[0]);
    incrRefCount(argv[1]);

    /* If the master decided to expire a key we must propagate it to replicas no matter what..
     * Even if module executed a command without asking for propagation. */
    int prev_replication_allowed = server.replication_allowed;
    server.replication_allowed = 1;
    alsoPropagate(db->id,argv,2,PROPAGATE_AOF|PROPAGATE_REPL);
    server.replication_allowed = prev_replication_allowed;

    decrRefCount(argv[0]);
    decrRefCount(argv[1]);
}


/* Used inside commands to schedule the propagation of additional commands
 * after the current command is propagated to AOF / Replication.
 *
 * dbid is the database ID the command should be propagated into.
 * Arguments of the command to propagate are passed as an array of redis
 * objects pointers of len 'argc', using the 'argv' vector.
 *
 * The function does not take a reference to the passed 'argv' vector,
 * so it is up to the caller to release the passed argv (but it is usually
 * stack allocated).  The function automatically increments ref count of
 * passed objects, so the caller does not need to. */
void alsoPropagate(int dbid, robj **argv, int argc, int target) {
    robj **argvcopy;
    int j;

    if (!shouldPropagate(target))
        return;

    argvcopy = zmalloc(sizeof(robj*)*argc);
    for (j = 0; j < argc; j++) {
        argvcopy[j] = argv[j];
        incrRefCount(argv[j]);
    }
    redisOpArrayAppend(&server.also_propagate,dbid,argvcopy,argc,target);
}


int redisOpArrayAppend(redisOpArray *oa, int dbid, robj **argv, int argc, int target) {
    redisOp *op;
    int prev_capacity = oa->capacity;

    if (oa->numops == 0) {
        oa->capacity = 16;
    } else if (oa->numops >= oa->capacity) {
        oa->capacity *= 2;
    }

    if (prev_capacity != oa->capacity)
        oa->ops = zrealloc(oa->ops,sizeof(redisOp)*oa->capacity);
    op = oa->ops+oa->numops;
    op->dbid = dbid;
    op->argv = argv;
    op->argc = argc;
    op->target = target;
    oa->numops++;
    return oa->numops;
}

这里的删除是通过传递两个robj对象来做的，其实就是将各个动作(del，unlink等)抽象陈robj对象。

struct sharedObjectsStruct {
    robj *crlf, *ok, *err, *emptybulk, *czero, *cone, *pong, *space,
    *queued, *null[4], *nullarray[4], *emptymap[4], *emptyset[4],
    *emptyarray, *wrongtypeerr, *nokeyerr, *syntaxerr, *sameobjecterr,
    *outofrangeerr, *noscripterr, *loadingerr,
    *slowevalerr, *slowscripterr, *slowmoduleerr, *bgsaveerr,
    *masterdownerr, *roslaveerr, *execaborterr, *noautherr, *noreplicaserr,
    *busykeyerr, *oomerr, *plus, *messagebulk, *pmessagebulk, *subscribebulk,
    *unsubscribebulk, *psubscribebulk, *punsubscribebulk, *del, *unlink,
    *rpop, *lpop, *lpush, *rpoplpush, *lmove, *blmove, *zpopmin, *zpopmax,
    *emptyscan, *multi, *exec, *left, *right, *hset, *srem, *xgroup, *xclaim,  
    *script, *replconf, *eval, *persist, *set, *pexpireat, *pexpire, 
    *time, *pxat, *absttl, *retrycount, *force, *justid, *entriesread,
    *lastid, *ping, *setid, *keepttl, *load, *createconsumer,
    *getack, *special_asterick, *special_equals, *default_username, *redacted,
    *ssubscribebulk,*sunsubscribebulk, *smessagebulk,
    *select[PROTO_SHARED_SELECT_CMDS],
    *integers[OBJ_SHARED_INTEGERS],
    *mbulkhdr[OBJ_SHARED_BULKHDR_LEN], /* "*<value>\r\n" */
    *bulkhdr[OBJ_SHARED_BULKHDR_LEN],  /* "$<value>\r\n" */
    *maphdr[OBJ_SHARED_BULKHDR_LEN],   /* "%<value>\r\n" */
    *sethdr[OBJ_SHARED_BULKHDR_LEN];   /* "~<value>\r\n" */
    sds minstring, maxstring;
};

reference

• Using Redis as an LRU cache， Random notes on improving the Redis LRU algorithm
• maxmemory 数据淘汰策略