Redis 源码简洁剖析 11 - 主 IO 线程及 Redis 6.0 多 IO 线程

Posted by Yano on February 8, 2022

Redis 到底是不是单线程的程序?

Redis 只有在处理「客户端请求」时,是单线程的;整个 Redis server 不是单线程的,还有后台线程在辅助处理任务。

Redis 选择单线程处理请求,是因为 Redis 操作的是「内存」,加上设计了「高效」的数据结构,所以操作速度极快,利用 IO 多路复用机制,单线程依旧可以有非常高的性能。

Redis 不让主线程执行一些耗时操作,比如同步写、删除等,而是交给后台线程异步完成,从而避免了对主线程的阻塞。

在 2020 年 5 月推出的 Redis 6.0 版本中,还会使用多线程来处理 IO 任务,能够充分利用服务器的多核特性,使用多核运行多线程,让多线程帮助加速数据读取命令解析数据写回的速度,提升 Redis 的整体性能。

多 IO 线程的初始化

在 main 函数中,会调用 InitServerLast 函数,Redis 6.0 源码:

void InitServerLast() {
    bioInit();
    // 初始化 IO 线程
    initThreadedIO();
    set_jemalloc_bg_thread(server.jemalloc_bg_thread);
    server.initial_memory_usage = zmalloc_used_memory();
}

在调用了 bioInit 函数后,又调用了 initThreadedIO 函数初始化多 IO 线程。initThreadedIO 函数在 networking.c 文件中。

void initThreadedIO(void) {
    // IO 线程激活标志:设置为「未激活」
    server.io_threads_active = 0;

    // 只有 1 个 io 线程,直接返回,直接在主线程处理 IO
    if (server.io_threads_num == 1) return;

    if (server.io_threads_num > IO_THREADS_MAX_NUM) {
        serverLog(LL_WARNING,"Fatal: too many I/O threads configured. "
                             "The maximum number is %d.", IO_THREADS_MAX_NUM);
        exit(1);
    }

    /* Spawn and initialize the I/O threads. */
    for (int i = 0; i < server.io_threads_num; i++) {
        io_threads_list[i] = listCreate();
        // Thread 0 是主线程
        if (i == 0) continue;

        /* Things we do only for the additional threads. */
        pthread_t tid;
        // 初始化 io_threads_mutex
        pthread_mutex_init(&io_threads_mutex[i],NULL);
        setIOPendingCount(i, 0);
        pthread_mutex_lock(&io_threads_mutex[i]); /* Thread will be stopped. */
        // pthread_create 创建 IO 线程,线程运行函数是 IOThreadMain
        if (pthread_create(&tid,NULL,IOThreadMain,(void*)(long)i) != 0) {
            serverLog(LL_WARNING,"Fatal: Can't initialize IO thread.");
            exit(1);
        }
        // 初始化 io_threads 数组,设置值为线程标识
        io_threads[i] = tid;
    }
}

代码中首先判断 io_threads_num:

  • io_threads_num = 1,表示直接在主线程处理,直接返回
  • io_threads_num > IO_THREADS_MAX_NUM,表示 IO 线程数量>宏定义的值(默认值 128),直接退出程序

initThreadedIO 函数就会给以下四个数组进行初始化操作:

  • io_threads_list 数组:保存了每个 IO 线程要处理的客户端,将数组每个元素初始化为一个 List 类型的列表
  • io_threads_pending 数组:保存等待每个 IO 线程处理的客户端个数
  • io_threads_mutex 数组:保存线程互斥锁
  • io_threads 数组:保存每个 IO 线程的描述符

这四个数组的定义都在 networking.c 文件中:


pthread_t io_threads[IO_THREADS_MAX_NUM];   //记录线程描述符的数组
pthread_mutex_t io_threads_mutex[IO_THREADS_MAX_NUM];  //记录线程互斥锁的数组
_Atomic unsigned long io_threads_pending[IO_THREADS_MAX_NUM];  //记录线程待处理的客户端个数
list *io_threads_list[IO_THREADS_MAX_NUM];  //记录线程对应处理的客户端

initThreadedIO 函数在 for 循环中,调用 pthread_create 函数创建线程。pthread_create 详细语法见:pthread_create(3) — Linux manual page

创建的线程要运行的函数是 IOThreadMain,*arg 参数就是当前创建线程的编号(从 1 开始,0 是主 IO 线程)。

/* Spawn and initialize the I/O threads. */
for (int i = 0; i < server.io_threads_num; i++) {
    io_threads_list[i] = listCreate();
    // Thread 0 是主线程
    if (i == 0) continue;

    /* Things we do only for the additional threads. */
    pthread_t tid;
    // 初始化 io_threads_mutex
    pthread_mutex_init(&io_threads_mutex[i],NULL);
    setIOPendingCount(i, 0);
    pthread_mutex_lock(&io_threads_mutex[i]);
    // pthread_create 创建 IO 线程,线程运行函数是 IOThreadMain
    if (pthread_create(&tid,NULL,IOThreadMain,(void*)(long)i) != 0) {
        serverLog(LL_WARNING,"Fatal: Can't initialize IO thread.");
        exit(1);
    }
    // 初始化 io_threads 数组,设置值为线程标识
    io_threads[i] = tid;
}

IO 线程运行函数 IOThreadMain

主要逻辑是一个 while(1) 的循环,会把 io_threads_list 在这个线程对应的元素取出来,判断并处理。

void *IOThreadMain(void *myid) {
    ……

    while(1) {
        /* Wait for start */
        for (int j = 0; j < 1000000; j++) {
            if (getIOPendingCount(id) != 0) break;
        }

        ……
        // 获取 IO 线程要处理的客户端列表
        listRewind(io_threads_list[id],&li);
        while((ln = listNext(&li))) {
            // 从客户端列表中获取一个客户端
            client *c = listNodeValue(ln);
            // 线程是「写操作」,调用 writeToClient 将数据写回客户端
            if (io_threads_op == IO_THREADS_OP_WRITE) {
                writeToClient(c,0);
            // 如果是『读操作』,调用 readQueryFromClient 从客户端读数据
            } else if (io_threads_op == IO_THREADS_OP_READ) {
                readQueryFromClient(c->conn);
            } 
            ……
        }
        // 处理完所有客户端,清空该线程的客户端列表
        listEmpty(io_threads_list[id]);
        // 将该线程的待处理任务数量设为 0
        setIOPendingCount(id, 0);
    }
}

注:上面代码中 io_threads_op 变量是在 handleClientsWithPendingWritesUsingThreads 函数和 handleClientsWithPendingReadsUsingThreads 函数中设置的。

问题:IO 线程要处理的客户端是如何添加到 io_threads_list 数组中的呢?

是在 redisServer 全局变量里,有两个 List 类型的成员变量:

  • clients_pending_write:待写回数据的客户端
  • clients_pending_read:待读取数据的客户端

struct redisServer {
    ...
    // 待写回数据的客户端
    list *clients_pending_write;  
    // 待读取数据的客户端
    list *clients_pending_read;  
    ...
}

Redis server 在接收到客户端请求、返回给客户端数据的过程中,会根据一定条件,推迟客户端的读写操作,并分别把待读写的客户端保存到这两个列表中。之后 Redis server 每次进入事件循环前,都会把列表中的客户端添加到 io_threads_list 数组中,交给 IO 线程处理。

如何推迟客户端「读」操作?

处理可读事件的回调函数是 readQueryFromClient。

void readQueryFromClient(connection *conn) {
    // 从 connection 结构中获取客户端
    client *c = connGetPrivateData(conn);
    ……

    // 是否推迟从客户端读取数据(使用多线程 IO 时)
    if (postponeClientRead(c)) return;

    ……
}

主要看下 postponeClientRead 函数。

int postponeClientRead(client *c) {
    if (server.io_threads_active &&
        server.io_threads_do_reads &&
        !ProcessingEventsWhileBlocked &&
        !(c->flags & (CLIENT_MASTER|CLIENT_SLAVE|CLIENT_PENDING_READ|CLIENT_BLOCKED))) 
    {
        // 客户端 flag 添加 CLIENT_PENDING_READ 标记,推迟客户端的读操作
        c->flags |= CLIENT_PENDING_READ;
        // 将客户端添加到 server 的 clients_pending_read 列表中
        listAddNodeHead(server.clients_pending_read,c);
        return 1;
    } else {
        return 0;
    }
}

if 的判断条件:是否可以推迟当前客户端的读操作;if 块里的执行逻辑:将客户端添加到 clients_pending_read 列表中。下面主要看下判断条件:

  1. server.io_threads_active = 1:多 IO 线程已激活。
  2. server.io_threads_do_reads = 1:多 IO 线程可用于处理延迟执行的客户端读操作,是在 Redis 配置文件 redis.conf 中,通过配置项 。io-threads-do-reads 设置的,默认值为 no。
  3. ProcessingEventsWhileBlocked = 0:ProcessingEventsWhileBlocked 函数没有在执行,当 Redis 在读取 RDB 文件或 AOF 文件时,会调用这个函数,用来处理事件驱动框架捕获到的事件,避免因读取 RDB 或 AOF 文件造成 Redis 阻塞。
  4. 客户端现有标识不能有 CLIENT_MASTERCLIENT_SLAVECLIENT_PENDING_READ
    • CLIENT_MASTER:客户端用于主从复制
    • CLIENT_SLAVE:客户端用于主从复制
    • CLIENT_PENDING_READ:客户端本来就被设置为推迟读操作

如何推迟客户端「写」操作?

Redis 在执行了客户端命令,要给客户端返回结果时,会调用 addReply 函数将待返回的结果写入输出缓冲区。addReply 函数开始就会调用 prepareClientToWrite 函数。

/* -----------------------------------------------------------------------------
 * Higher level functions to queue data on the client output buffer.
 * The following functions are the ones that commands implementations will call.
 * -------------------------------------------------------------------------- */

/* Add the object 'obj' string representation to the client output buffer. */
void addReply(client *c, robj *obj) {
    if (prepareClientToWrite(c) != C_OK) return;

    ……
}

prepareClientToWrite 函数的注释如下:

/* This function is called every time we are going to transmit new data
 * to the client. The behavior is the following:
 *
 * If the client should receive new data (normal clients will) the function
 * returns C_OK, and make sure to install the write handler in our event
 * loop so that when the socket is writable new data gets written.
 *
 * If the client should not receive new data, because it is a fake client
 * (used to load AOF in memory), a master or because the setup of the write
 * handler failed, the function returns C_ERR.
 *
 * The function may return C_OK without actually installing the write
 * event handler in the following cases:
 *
 * 1) The event handler should already be installed since the output buffer
 *    already contains something.
 * 2) The client is a slave but not yet online, so we want to just accumulate
 *    writes in the buffer but not actually sending them yet.
 *
 * Typically gets called every time a reply is built, before adding more
 * data to the clients output buffers. If the function returns C_ERR no
 * data should be appended to the output buffers. */
int prepareClientToWrite(client *c) {
    ……
    // 当前客户端没有待写回数据 && flag 不包含 CLIENT_PENDING_READ
    if (!clientHasPendingReplies(c) && !(c->flags & CLIENT_PENDING_READ))
            clientInstallWriteHandler(c);

    return C_OK;
}

clientInstallWriteHandler 如下,if 判断条件就不赘述了。

void clientInstallWriteHandler(client *c) {

    if (!(c->flags & CLIENT_PENDING_WRITE) &&
        (c->replstate == REPL_STATE_NONE ||
         (c->replstate == SLAVE_STATE_ONLINE && !c->repl_put_online_on_ack)))
    {
        // 将客户端的标识设置为 CLIENT_PENDING_WRITE(待写回)
        c->flags |= CLIENT_PENDING_WRITE;
        // 将 client 加入 server 的 clients_pending_write 列表
        listAddNodeHead(server.clients_pending_write,c);
    }
}

上面介绍如如何推迟客户端的读操作、写操作,那 Redis 是如何将推迟读写操作的客户端,分配给多 IO 线程执行的呢?是通过:

  • handleClientsWithPendingReadsUsingThreads 函数:将 clients_pending_read 列表中的客户端分配给 IO 线程
  • handleClientsWithPendingWritesUsingThreads 函数:将 clients_pending_write 列表中的客户端分配给 IO 线程

如何把待「读」客户端分配给 IO 线程执行?

beforeSleep 函数中调用了 handleClientsWithPendingReadsUsingThreads 函数:

/* We should handle pending reads clients ASAP after event loop. */
handleClientsWithPendingReadsUsingThreads();

handleClientsWithPendingReadsUsingThreads 函数如下,逻辑都在注释中:

/* When threaded I/O is also enabled for the reading + parsing side, the
 * readable handler will just put normal clients into a queue of clients to
 * process (instead of serving them synchronously). This function runs
 * the queue using the I/O threads, and process them in order to accumulate
 * the reads in the buffers, and also parse the first command available
 * rendering it in the client structures. */
int handleClientsWithPendingReadsUsingThreads(void) {
    // 判断 io_threads_active 是否被激活,io_threads_do_reads 是否可以用 IO 线程处理待读客户端
    if (!server.io_threads_active || !server.io_threads_do_reads) return 0;

    // 判断 clients_pending_read 长度
    int processed = listLength(server.clients_pending_read);
    if (processed == 0) return 0;

    /* Distribute the clients across N different lists. */
    listIter li;
    listNode *ln;
    // 获取 clients_pending_read 的客户端列表
    listRewind(server.clients_pending_read,&li);
    // 轮询方式,将客户端分配给 IO 线程
    int item_id = 0;
    while((ln = listNext(&li))) {
        client *c = listNodeValue(ln);
        int target_id = item_id % server.io_threads_num;
        listAddNodeTail(io_threads_list[target_id],c);
        item_id++;
    }

    // 将 IO 线程的操作标识设置为「读操作」
    io_threads_op = IO_THREADS_OP_READ;
    for (int j = 1; j < server.io_threads_num; j++) {
        // 每个线程等待处理的客户端数量 → io_threads_pending 数组
        int count = listLength(io_threads_list[j]);
        setIOPendingCount(j, count);
    }

    // 处理 0 号线程(主线程)的待读客户端
    listRewind(io_threads_list[0],&li);
    while((ln = listNext(&li))) {
        client *c = listNodeValue(ln);
        readQueryFromClient(c->conn);
    }
    // 清空 0 号列表
    listEmpty(io_threads_list[0]);

    // 循环,等待其他所有 IO 线程的待读客户端都处理完
    while(1) {
        unsigned long pending = 0;
        for (int j = 1; j < server.io_threads_num; j++)
            pending += getIOPendingCount(j);
        if (pending == 0) break;
    }

    /* Run the list of clients again to process the new buffers. */
    // 取出 clients_pending_read 列表
    while(listLength(server.clients_pending_read)) {
        ln = listFirst(server.clients_pending_read);
        client *c = listNodeValue(ln);
        // 判断客户端标识符是否有 CLIENT_PENDING_READ,有则表示被 IO 线程解析过
        c->flags &= ~CLIENT_PENDING_READ;
        // 将客户端从 clients_pending_read 列表中删掉
        listDelNode(server.clients_pending_read,ln);

        serverAssert(!(c->flags & CLIENT_BLOCKED));
        if (processPendingCommandsAndResetClient(c) == C_ERR) {
            /* If the client is no longer valid, we avoid
             * processing the client later. So we just go
             * to the next. */
            continue;
        }

        // 解析并执行客户端的所有命令
        processInputBuffer(c);

        /* We may have pending replies if a thread readQueryFromClient() produced
         * replies and did not install a write handler (it can't).
         */
        if (!(c->flags & CLIENT_PENDING_WRITE) && clientHasPendingReplies(c))
            clientInstallWriteHandler(c);
    }

    /* Update processed count on server */
    server.stat_io_reads_processed += processed;

    return processed;
}

如何把待「写」客户端分配给 IO 线程执行?

待写客户端的分配处理是由 handleClientsWithPendingWritesUsingThreads 函数完成的,该函数也是在 beforeSleep 函数中调用的。逻辑和 handleClientsWithPendingReadsUsingThreads 函数很像。

int handleClientsWithPendingWritesUsingThreads(void) {

    // 判断 clients_pending_write 列表的数量
    int processed = listLength(server.clients_pending_write);
    if (processed == 0) return 0;

    // 只有主 IO 线程 || 不使用 IO 线程
    if (server.io_threads_num == 1 || stopThreadedIOIfNeeded()) {
        return handleClientsWithPendingWrites();
    }

    /* Start threads if needed. */
    if (!server.io_threads_active) startThreadedIO();

    /* Distribute the clients across N different lists. */
    listIter li;
    listNode *ln;
    listRewind(server.clients_pending_write,&li);
    int item_id = 0;
    // 把待写客户端,按照轮询方式分配给 IO 线程
    while((ln = listNext(&li))) {
        client *c = listNodeValue(ln);
        c->flags &= ~CLIENT_PENDING_WRITE;

        if (c->flags & CLIENT_CLOSE_ASAP) {
            listDelNode(server.clients_pending_write, ln);
            continue;
        }

        int target_id = item_id % server.io_threads_num;
        listAddNodeTail(io_threads_list[target_id],c);
        item_id++;
    }

    // 将 IO 线程的操作标识设置为「写操作」
    io_threads_op = IO_THREADS_OP_WRITE;
    for (int j = 1; j < server.io_threads_num; j++) {
        // 每个线程等待处理的客户端数量 → io_threads_pending 数组
        int count = listLength(io_threads_list[j]);
        setIOPendingCount(j, count);
    }

    /* Also use the main thread to process a slice of clients. */
    listRewind(io_threads_list[0],&li);
    while((ln = listNext(&li))) {
        client *c = listNodeValue(ln);
        writeToClient(c,0);
    }
    listEmpty(io_threads_list[0]);

    // 循环,等待其他所有 IO 线程的待写客户端都处理完
    while(1) {
        unsigned long pending = 0;
        for (int j = 1; j < server.io_threads_num; j++)
            pending += getIOPendingCount(j);
        if (pending == 0) break;
    }

    /* Run the list of clients again to install the write handler where
     * needed. */
    listRewind(server.clients_pending_write,&li);
    while((ln = listNext(&li))) {
        client *c = listNodeValue(ln);

        // 再次检查是否有待写客户端
        if (clientHasPendingReplies(c) &&
                connSetWriteHandler(c->conn, sendReplyToClient) == AE_ERR)
        {
            freeClientAsync(c);
        }
    }
    listEmpty(server.clients_pending_write);

    /* Update processed count on server */
    server.stat_io_writes_processed += processed;

    return processed;
}

需要注意的是,stopThreadedIOIfNeeded 函数中会判断待写入的客户端数量如果 < IO 线程数 * 2,则也会直接返回,直接使用主 IO 线程处理待写客户端。这是因为待写客户端不多时,使用多线程效率反而会下降。

int stopThreadedIOIfNeeded(void) {
    int pending = listLength(server.clients_pending_write);

    /* Return ASAP if IO threads are disabled (single threaded mode). */
    if (server.io_threads_num == 1) return 1;

    if (pending < (server.io_threads_num*2)) {
        if (server.io_threads_active) stopThreadedIO();
        return 1;
    } else {
        return 0;
    }
}

总结

Redis 6.0 实现的多 IO 线程机制,主要是使用多个 IO 线程,并发处理客户端读取数据解析命令写回数据,充分利用服务器的多核特性,提高 IO 效率。

Redis server 会根据 readQueryFromClient 函数调用 postponeClientRead 函数决定是否要推迟客户端操作;会根据 addReply 函数中的 prepareClientToWrite 函数,决定是否推迟客户端的写操作。待读客户端加入到 clients_pending_read 列表,待写客户端加入 clients_pending_write 列表。

IO 线程创建之后,会一直检测 io_threads_list 列表,如果有待读写的客户端,IO 线程就会调用 readQueryFromClient 或 writeToClient 函数进行处理。

但是多 IO 线程并不会执行命令,执行命令仍然在主 IO 线程

参考链接

  • [极客时间:12 Redis 真的是单线程吗?](https://time.geekbang.org/column/article/409927)
  • [极客时间:13 Redis 6.0 多 IO 线程的效率提高了吗?](https://time.geekbang.org/column/article/410666)
  • pthread_create(3) — Linux manual page

Redis 源码简洁剖析系列

Java 编程思想-最全思维导图-GitHub 下载链接,需要的小伙伴可以自取~

原创不易,希望大家转载时请先联系我,并标注原文链接。

我的公众号

coding 笔记、读书笔记、点滴记录,以后的文章也会同步到公众号(Coding Insight)中,希望大家关注^_^