代码改变世界

深入剖析 redis 主从复制

2014-05-12 21:44  捣乱小子  阅读(17297)  评论(5编辑  收藏  举报

主从概述

redis 支持 master-slave(主从)模式,redis server 可以设置为另一个 redis server 的主机(从机),从机定期从主机拿数据。特殊的,一个 从机同样可以设置为一个 redis server 的主机,这样一来 master-slave 的分布看起来就是一个有向无环图 DAG,如此形成 redis server 集群,无论是主机还是从机都是 redis server,都可以提供服务)。

master_slave

在配置后,主机可负责读写服务,从机只负责读。redis 提高这种配置方式,为的是让其支持数据的弱一致性,即最终一致性。在业务中,选择强一致性还是若已执行,应该取决于具体的业务需求,像微博,完全可以使用弱一致性模型;像淘宝,可以选用强一致性模型。

redis 主从复制的实现主要在 replication.c 中。

这篇文章涉及较多的代码,但我已经尽量删繁就简,达到能说明问题本质。为了保留代码的原生性并让读者能够阅读原生代码的注释,剖析 redis 的几篇文章都没有删除代码中的英文注释,并已加注释。

积压空间

在《深入剖析 redis AOF 持久化策略》中,介绍了更新缓存的概念,举一个例子:客户端发来命令:set name Jhon,这一数据更新被记录为:*3\r\n$3\r\nSET\r\n$4\r\nname\r\n$3\r\nJhon\r\n,并存储在更新缓存中。

同样,在主从连接中,也有更新缓存的概念。只是两者的用途不一样,前者被写入本地,后者被写入从机,这里我们把它成为积压空间。

更新缓存存储在 server.repl_backlog,redis 将其作为一个环形空间来处理,这样做节省了空间,避免内存再分配的情况。

struct redisServer {
    /* Replication (master) */
    // 最近一次使用(访问)的数据集
    int slaveseldb;                 /* Last SELECTed DB in replication output */

    // 全局的数据同步偏移量
    long long master_repl_offset;   /* Global replication offset */

    // 主从连接心跳频率
    int repl_ping_slave_period;     /* Master pings the slave every N seconds */

    // 积压空间指针
    char *repl_backlog;             /* Replication backlog for partial syncs */

    // 积压空间大小
    long long repl_backlog_size;    /* Backlog circular buffer size */

    // 积压空间中写入的新数据的大小
    long long repl_backlog_histlen; /* Backlog actual data length */

    // 下一次向积压空间写入数据的起始位置
    long long repl_backlog_idx;     /* Backlog circular buffer current offset */

    // 积压数据的起始位置,是一个宏观值
    long long repl_backlog_off;     /* Replication offset of first byte in the
                                       backlog buffer. */

    // 积压空间有效时间
    time_t repl_backlog_time_limit; /* Time without slaves after the backlog
                                       gets released. */
}

积压空间中的数据变更记录是什么时候被写入的?在执行一个 redis 命令的时候,如果存在数据的修改(写),那么就会把变更记录传播。redis 源码中是这么实现的:call()->propagate()->replicationFeedSlaves()

注释:命令真正执行的地方在 call() 中,call() 如果发现数据被修改(dirty),则传播 propagrate(),replicationFeedSlaves() 将修改记录写入积压空间和所有已连接的从机。

这里可能会有疑问:为什么把数据添加入积压空间,又把数据分发给所有的从机?为什么不仅仅将数据分发给所有从机呢?

因为有一些从机会因特殊情况(???)与主机断开连接,注意从机断开前有暂存主机的状态信息,因此这些断开的从机就没有及时收到更新的数据。redis 为了让断开的从机在下次连接后能够获取更新数据,将更新数据加入了积压空间。从 replicationFeedSlaves() 实现来看,在线的 slave 能马上收到数据更新记录;因某些原因暂时断开连接的 slave,需要从积压空间中找回断开期间的数据更新记录。如果断开的时间足够长,master 会拒绝 slave 的部分同步请求,从而 slave 只能进行全同步。

下面是源码注释:

// call() 函数是执行命令的核心函数,真正执行命令的地方
/* Call() is the core of Redis execution of a command */
void call(redisClient *c, int flags) {
    ......
    /* Call the command. */
    c->flags &= ~(REDIS_FORCE_AOF|REDIS_FORCE_REPL);
    redisOpArrayInit(&server.also_propagate);

    // 脏数据标记,数据是否被修改
    dirty = server.dirty;

    // 执行命令对应的函数
    c->cmd->proc(c);

    dirty = server.dirty-dirty;
    duration = ustime()-start;

    ......

    // 将客户端请求的数据修改记录传播给 AOF 和从机
    /* Propagate the command into the AOF and replication link */
    if (flags & REDIS_CALL_PROPAGATE) {
        int flags = REDIS_PROPAGATE_NONE;

        // 强制主从复制
        if (c->flags & REDIS_FORCE_REPL) flags |= REDIS_PROPAGATE_REPL;

        // 强制 AOF 持久化
        if (c->flags & REDIS_FORCE_AOF) flags |= REDIS_PROPAGATE_AOF;

        // 数据被修改
        if (dirty)
            flags |= (REDIS_PROPAGATE_REPL | REDIS_PROPAGATE_AOF);

        // 传播数据修改记录
        if (flags != REDIS_PROPAGATE_NONE)
            propagate(c->cmd,c->db->id,c->argv,c->argc,flags);
    }
    ......
}

// 向 AOF 和从机发布数据更新
/* Propagate the specified command (in the context of the specified database id)
 * to AOF and Slaves.
 *
 * flags are an xor between:
 * + REDIS_PROPAGATE_NONE (no propagation of command at all)
 * + REDIS_PROPAGATE_AOF (propagate into the AOF file if is enabled)
 * + REDIS_PROPAGATE_REPL (propagate into the replication link)
 */
void propagate(struct redisCommand *cmd, int dbid, robj **argv, int argc,
               int flags)
{
    // AOF 策略需要打开,且设置 AOF 传播标记,将更新发布给本地文件
    if (server.aof_state != REDIS_AOF_OFF && flags & REDIS_PROPAGATE_AOF)
        feedAppendOnlyFile(cmd,dbid,argv,argc);

    // 设置了从机传播标记,将更新发布给从机
    if (flags & REDIS_PROPAGATE_REPL)
        replicationFeedSlaves(server.slaves,dbid,argv,argc);
}

// 向积压空间和从机发送数据
void replicationFeedSlaves(list *slaves, int dictid, robj **argv, int argc) {
    listNode *ln;
    listIter li;
    int j, len;
    char llstr[REDIS_LONGSTR_SIZE];

    // 没有积压数据且没有从机,直接退出
    /* If there aren't slaves, and there is no backlog buffer to populate,
     * we can return ASAP. */
    if (server.repl_backlog == NULL && listLength(slaves) == 0) return;

    /* We can't have slaves attached and no backlog. */
    redisAssert(!(listLength(slaves) != 0 && server.repl_backlog == NULL));

    /* Send SELECT command to every slave if needed. */
    if (server.slaveseldb != dictid) {
        robj *selectcmd;

        // 小于等于 10 的可以用共享对象
        /* For a few DBs we have pre-computed SELECT command. */
        if (dictid >= 0 && dictid < REDIS_SHARED_SELECT_CMDS) {
            selectcmd = shared.select[dictid];
        } else {
        // 不能使用共享对象,生成 SELECT 命令对应的 redis 对象
            int dictid_len;

            dictid_len = ll2string(llstr,sizeof(llstr),dictid);
            selectcmd = createObject(REDIS_STRING,
                sdscatprintf(sdsempty(),
                "*2\r\n$6\r\nSELECT\r\n$%d\r\n%s\r\n",
                dictid_len, llstr));
        }

        // 这里可能会有疑问:为什么把数据添加入积压空间,又把数据分发给所有的从机?
        // 为什么不仅仅将数据分发给所有从机呢?
        // 因为有一些从机会因特殊情况(???)与主机断开连接,注意从机断开前有暂存
        // 主机的状态信息,因此这些断开的从机就没有及时收到更新的数据。redis 为了让
        // 断开的从机在下次连接后能够获取更新数据,将更新数据加入了积压空间。

        // 将 SELECT 命令对应的 redis 对象数据添加到积压空间
        /* Add the SELECT command into the backlog. */
        if (server.repl_backlog) feedReplicationBacklogWithObject(selectcmd);

        // 将数据分发所有的从机
        /* Send it to slaves. */
        listRewind(slaves,&li);
        while((ln = listNext(&li))) {
            redisClient *slave = ln->value;
            addReply(slave,selectcmd);
        }

        // 销毁对象
        if (dictid < 0 || dictid >= REDIS_SHARED_SELECT_CMDS)
            decrRefCount(selectcmd);
    }

    // 更新最近一次使用(访问)的数据集
    server.slaveseldb = dictid;

    // 将命令写入积压空间
    /* Write the command to the replication backlog if any. */
    if (server.repl_backlog) {
        char aux[REDIS_LONGSTR_SIZE+3];

        // 命令个数
        /* Add the multi bulk reply length. */
        aux[0] = '*';
        len = ll2string(aux+1,sizeof(aux)-1,argc);
        aux[len+1] = '\r';
        aux[len+2] = '\n';
        feedReplicationBacklog(aux,len+3);

        // 逐个命令写入
        for (j = 0; j < argc; j++) {
            long objlen = stringObjectLen(argv[j]);

            /* We need to feed the buffer with the object as a bulk reply
             * not just as a plain string, so create the $..CRLF payload len
             * ad add the final CRLF */
            aux[0] = '$';
            len = ll2string(aux+1,sizeof(aux)-1,objlen);
            aux[len+1] = '\r';
            aux[len+2] = '\n';

            /* 每个命令格式如下:
            $3
            *3
            SET
            *4
            NAME
            *4
            Jhon*/

            // 命令长度
            feedReplicationBacklog(aux,len+3);
            // 命令
            feedReplicationBacklogWithObject(argv[j]);
            // 换行
            feedReplicationBacklog(aux+len+1,2);
        }
    }

    // 立即给每一个从机发送命令
    /* Write the command to every slave. */
    listRewind(slaves,&li);
    while((ln = listNext(&li))) {
        redisClient *slave = ln->value;

        // 如果从机要求全同步,则不对此从机发送数据
        /* Don't feed slaves that are still waiting for BGSAVE to start */
        if (slave->replstate == REDIS_REPL_WAIT_BGSAVE_START) continue;

        /* Feed slaves that are waiting for the initial SYNC (so these commands
         * are queued in the output buffer until the initial SYNC completes),
         * or are already in sync with the master. */

        // 向从机命令的长度
        /* Add the multi bulk length. */
        addReplyMultiBulkLen(slave,argc);

        // 向从机发送命令
        /* Finally any additional argument that was not stored inside the
         * static buffer if any (from j to argc). */
        for (j = 0; j < argc; j++)
            addReplyBulk(slave,argv[j]);
    }
}

主从数据同步机制概述

redis 主从同步有两种方式(或者所两个阶段):全同步和部分同步。

主从刚刚连接的时候,进行全同步;全同步结束后,进行部分同步。当然,如果有需要,slave 在任何时候都可以发起全同步。redis 策略是,无论如何,首先会尝试进行部分同步,如不成功,要求从机进行全同步,并启动 BGSAVE……BGSAVE 结束后,传输 RDB 文件;如果成功,允许从机进行部分同步,并传输积压空间中的数据。

下面这幅图,总结了主从同步的机制:

how_redis_replication_sync_works

如需设置 slave,master 需要向 slave 发送 SLAVEOF hostname port,从机接收到后会自动连接主机,注册相应读写事件(syncWithMaster())。

// 修改主机
void slaveofCommand(redisClient *c) {
    if (!strcasecmp(c->argv[1]->ptr,"no") &&
        !strcasecmp(c->argv[2]->ptr,"one")) {
        // slaveof no one 断开主机连接
        if (server.masterhost) {
            replicationUnsetMaster();
            redisLog(REDIS_NOTICE,"MASTER MODE enabled (user request)");
        }
    } else {
        long port;

        if ((getLongFromObjectOrReply(c, c->argv[2], &port, NULL) != REDIS_OK))
            return;

        // 可能已经连接需要连接的主机
        /* Check if we are already attached to the specified slave */
        if (server.masterhost && !strcasecmp(server.masterhost,c->argv[1]->ptr)
            && server.masterport == port) {
            redisLog(REDIS_NOTICE,"SLAVE OF would result into synchronization with the master we are already connected with. No operation performed.");
            addReplySds(c,sdsnew("+OK Already connected to specified master\r\n"));
            return;
        }

        // 断开之前连接主机的连接,连接新的。 replicationSetMaster() 并不会真正连接主机,只是修改 struct server 中关于主机的设置。真正的主机连接在 replicationCron() 中完成
        /* There was no previous master or the user specified a different one,
         * we can continue. */
        replicationSetMaster(c->argv[1]->ptr, port);
        redisLog(REDIS_NOTICE,"SLAVE OF %s:%d enabled (user request)",
            server.masterhost, server.masterport);
    }
    addReply(c,shared.ok);
}

// 设置新主机
/* Set replication to the specified master address and port. */
void replicationSetMaster(char *ip, int port) {
    sdsfree(server.masterhost);
    server.masterhost = sdsdup(ip);
    server.masterport = port;

    // 断开之前主机的连接
    if (server.master) freeClient(server.master);
    disconnectSlaves(); /* Force our slaves to resync with us as well. */

    // 取消缓存主机
    replicationDiscardCachedMaster(); /* Don't try a PSYNC. */

    // 释放积压空间
    freeReplicationBacklog(); /* Don't allow our chained slaves to PSYNC. */

    // cancelReplicationHandshake() 尝试断开数据传输和主机连接
    cancelReplicationHandshake();
    server.repl_state = REDIS_REPL_CONNECT;
    server.master_repl_offset = 0;
}

// 管理主从连接的定时程序定时程序,每秒执行一次
// 在 serverCorn() 中调用
/* --------------------------- REPLICATION CRON  ----------------------------- */

/* Replication cron funciton, called 1 time per second. */
void replicationCron(void) {
    ......
    // 如果需要( REDIS_REPL_CONNECT),尝试连接主机,真正连接主机的操作在这里
    /* Check if we should connect to a MASTER */
    if (server.repl_state == REDIS_REPL_CONNECT) {
        redisLog(REDIS_NOTICE,"Connecting to MASTER %s:%d",
            server.masterhost, server.masterport);
        if (connectWithMaster() == REDIS_OK) {
            redisLog(REDIS_NOTICE,"MASTER <-> SLAVE sync started");
        }
    }
    ......
}

全同步

接着自动发起 PSYNC 请求 master 进行全同步。无论如何,redis 首先会尝试部分同步,如果失败才尝试全同步。而刚刚建立连接的 master-slave 需要全同步。

从机连接主机后,会主动发起 PSYNC 命令,从机会提供 master_runid 和 offset,主机验证 master_runid 和 offset 是否有效?master_runid 相当于主机身份验证码,用来验证从机上一次连接的主机,offset 是全局积压空间数据的偏移量。
验证未通过则,则进行全同步:主机返回 +FULLRESYNC master_runid offset(从机接收并记录 master_runid 和 offset,并准备接收 RDB 文件)接着启动 BGSAVE 生成 RDB 文件,BGSAVE 结束后,向从机传输,从而完成全同步。

// 连接主机 connectWithMaster() 的时候,会被注册为回调函数
void syncWithMaster(aeEventLoop *el, int fd, void *privdata, int mask) {
    char tmpfile[256], *err;
    int dfd, maxtries = 5;
    int sockerr = 0, psync_result;
    socklen_t errlen = sizeof(sockerr);

    ......

    // 这里尝试向主机请求部分同步,主机会回复以拒绝或接受请求。如果拒绝部分同步,会返回 +FULLRESYNC master_runid offset
    // 从机接收后准备进行全同步    psync_result = slaveTryPartialResynchronization(fd);
    if (psync_result == PSYNC_CONTINUE) {
        redisLog(REDIS_NOTICE, "MASTER <-> SLAVE sync: Master accepted a Partial Resynchronization.");
        return;
    }

    // 执行全同步
    /* Fall back to SYNC if needed. Otherwise psync_result == PSYNC_FULLRESYNC
     * and the server.repl_master_runid and repl_master_initial_offset are
     * already populated. */

    // 未知结果,进行出错处理
    if (psync_result == PSYNC_NOT_SUPPORTED) {
        redisLog(REDIS_NOTICE,"Retrying with SYNC...");
        if (syncWrite(fd,"SYNC\r\n",6,server.repl_syncio_timeout*1000) == -1) {
            redisLog(REDIS_WARNING,"I/O error writing to MASTER: %s",
                strerror(errno));
            goto error;
        }
    }

    // 为什么要尝试 5次???
    /* Prepare a suitable temp file for bulk transfer */
    while(maxtries--) {
        snprintf(tmpfile,256,
            "temp-%d.%ld.rdb",(int)server.unixtime,(long int)getpid());
        dfd = open(tmpfile,O_CREAT|O_WRONLY|O_EXCL,0644);
        if (dfd != -1) break;
        sleep(1);
    }
    if (dfd == -1) {
        redisLog(REDIS_WARNING,"Opening the temp file needed for MASTER <-> SLAVE synchronization: %s",strerror(errno));
        goto error;
    }

    // 注册读事件,回调函数 readSyncBulkPayload(), 准备读 RDB 文件
    /* Setup the non blocking download of the bulk file. */
    if (aeCreateFileEvent(server.el,fd, AE_READABLE,readSyncBulkPayload,NULL)
            == AE_ERR)
    {
        redisLog(REDIS_WARNING,
            "Can't create readable event for SYNC: %s (fd=%d)",
            strerror(errno),fd);
        goto error;
    }

    // 设置传输 RDB 文件数据的选项
    // 状态
    server.repl_state = REDIS_REPL_TRANSFER;
    // RDB 文件大小
    server.repl_transfer_size = -1;
    // 已经传输的大小
    server.repl_transfer_read = 0;
    // 上一次同步的偏移,为的是定时写入磁盘
    server.repl_transfer_last_fsync_off = 0;
    // 本地 RDB 文件套接字
    server.repl_transfer_fd = dfd;
    // 上一次同步 IO 时间
    server.repl_transfer_lastio = server.unixtime;
    // 临时文件名
    server.repl_transfer_tmpfile = zstrdup(tmpfile);
    return;

error:
    close(fd);
    server.repl_transfer_s = -1;
    server.repl_state = REDIS_REPL_CONNECT;
    return;
}

全同步请求的数据是 RDB 数据文件和积压空间中的数据。关于 RDB 数据文件,请参看《深入剖析 redis RDB 持久化策略》。如果没有后台持久化 BGSAVE 进程,那么 BGSVAE 会被触发,否则所有请求全同步的 slave 都会被标记为等待 BGSAVE 结束。BGSAVE 结束后,master 会马上向所有的从机发送 RDB 文件。

// 主机 SYNC 和 PSYNC 命令处理函数,会尝试进行部分同步和全同步
/* SYNC ad PSYNC command implemenation. */
void syncCommand(redisClient *c) {
    ......
    // 主机尝试部分同步,失败的话向从机发送 +FULLRESYNC master_runid offset,接着启动 BGSAVE

    // 执行全同步:
    /* Full resynchronization. */
    server.stat_sync_full++;

    /* Here we need to check if there is a background saving operation
     * in progress, or if it is required to start one */
    if (server.rdb_child_pid != -1) {
    /*  存在 BGSAVE 后台进程。
        1.如果 master 现有所连接的所有从机 slaves 当中有存在 REDIS_REPL_WAIT_BGSAVE_END 的从机,那么将从机 c 设置为 REDIS_REPL_WAIT_BGSAVE_END;
        2.否则,设置为 REDIS_REPL_WAIT_BGSAVE_START*/

        /* Ok a background save is in progress. Let's check if it is a good
         * one for replication, i.e. if there is another slave that is
         * registering differences since the server forked to save */
        redisClient *slave;
        listNode *ln;
        listIter li;

        // 检测是否已经有从机申请全同步
        listRewind(server.slaves,&li);
        while((ln = listNext(&li))) {
            slave = ln->value;
            if (slave->replstate == REDIS_REPL_WAIT_BGSAVE_END) break;
        }

        if (ln) {
        // 存在状态为 REDIS_REPL_WAIT_BGSAVE_END 的从机 slave,
        // 就将此从机 c 状态设置为 REDIS_REPL_WAIT_BGSAVE_END,
        // 从而在 BGSAVE 进程结束后,可以发送 RDB 文件,
        // 同时将从机 slave 中的更新复制到此从机 c。

            /* Perfect, the server is already registering differences for
             * another slave. Set the right state, and copy the buffer. */

            // 将其他从机上的待回复的缓存复制到从机 c
            copyClientOutputBuffer(c,slave);

            // 修改从机 c 状态为「等待 BGSAVE 进程结束」
            c->replstate = REDIS_REPL_WAIT_BGSAVE_END;
            redisLog(REDIS_NOTICE,"Waiting for end of BGSAVE for SYNC");
        } else {
        // 不存在状态为 REDIS_REPL_WAIT_BGSAVE_END 的从机,就将此从机 c 状态设置为 REDIS_REPL_WAIT_BGSAVE_START,即等待新的 BGSAVE 进程的开启。

            // 修改状态为「等待 BGSAVE 进程开始」
            /* No way, we need to wait for the next BGSAVE in order to
             * register differences */
            c->replstate = REDIS_REPL_WAIT_BGSAVE_START;
            redisLog(REDIS_NOTICE,"Waiting for next BGSAVE for SYNC");
        }
    } else {
    // 不存在 BGSAVE 后台进程,启动一个新的 BGSAVE 进程

        /* Ok we don't have a BGSAVE in progress, let's start one */
        redisLog(REDIS_NOTICE,"Starting BGSAVE for SYNC");
        if (rdbSaveBackground(server.rdb_filename) != REDIS_OK) {
            redisLog(REDIS_NOTICE,"Replication failed, can't BGSAVE");
            addReplyError(c,"Unable to perform background save");
            return;
        }

        // 将此从机 c 状态设置为 REDIS_REPL_WAIT_BGSAVE_END,从而在 BGSAVE 进程结束后,可以发送 RDB 文件,同时将从机 slave 中的更新复制到此从机 c。
        c->replstate = REDIS_REPL_WAIT_BGSAVE_END;

        // 清理脚本缓存???
        /* Flush the script cache for the new slave. */
        replicationScriptCacheFlush();
    }

    if (server.repl_disable_tcp_nodelay)
        anetDisableTcpNoDelay(NULL, c->fd); /* Non critical if it fails. */
    c->repldbfd = -1;
    c->flags |= REDIS_SLAVE;
    server.slaveseldb = -1; /* Force to re-emit the SELECT command. */
    listAddNodeTail(server.slaves,c);
    if (listLength(server.slaves) == 1 && server.repl_backlog == NULL)
        createReplicationBacklog();
    return;
}

// BGSAVE 结束后,会调用
/* A background saving child (BGSAVE) terminated its work. Handle this. */
void backgroundSaveDoneHandler(int exitcode, int bysignal) {
    // 其他操作
    ......
    // 可能从机正在等待 BGSAVE 进程的终止
    /* Possibly there are slaves waiting for a BGSAVE in order to be served
     * (the first stage of SYNC is a bulk transfer of dump.rdb) */
    updateSlavesWaitingBgsave(exitcode == 0 ? REDIS_OK : REDIS_ERR);
}

// 当 RDB 持久化(backgroundSaveDoneHandler())结束后,会调用此函数
// RDB 文件就绪,给所有的从机发送 RDB 文件
/* This function is called at the end of every background saving.
* The argument bgsaveerr is REDIS_OK if the background saving succeeded
* otherwise REDIS_ERR is passed to the function.
*
* The goal of this function is to handle slaves waiting for a successful
* background saving in order to perform non-blocking synchronization. */
void updateSlavesWaitingBgsave(int bgsaveerr) {
    listNode *ln;
    int startbgsave = 0;
    listIter li;

    listRewind(server.slaves,&li);
    while((ln = listNext(&li))) {
        redisClient *slave = ln->value;

        // 等待 BGSAVE 开始。调整状态为等待下一次 BGSAVE 进程的结束
        if (slave->replstate == REDIS_REPL_WAIT_BGSAVE_START) {
            startbgsave = 1;

            slave->replstate = REDIS_REPL_WAIT_BGSAVE_END;

        // 等待 BGSAVE 结束。准备向 slave 发送 RDB 文件
        } else if (slave->replstate == REDIS_REPL_WAIT_BGSAVE_END) {
            struct redis_stat buf;

            // 如果 RDB 持久化失败, bgsaveerr 会被设置为 REDIS_ERR
            if (bgsaveerr != REDIS_OK) {
                freeClient(slave);
                redisLog(REDIS_WARNING,"SYNC failed. BGSAVE child returned an error");
                continue;
            }

            // 打开 RDB 文件
            if ((slave->repldbfd = open(server.rdb_filename,O_RDONLY)) == -1 ||
                redis_fstat(slave->repldbfd,&buf) == -1) {
                freeClient(slave);
                redisLog(REDIS_WARNING,"SYNC failed. Can't open/stat DB after BGSAVE: %s", strerror(errno));
                continue;
            }

            slave->repldboff = 0;
            slave->repldbsize = buf.st_size;
            slave->replstate = REDIS_REPL_SEND_BULK;

            // 如果之前有注册写事件,取消
            aeDeleteFileEvent(server.el,slave->fd,AE_WRITABLE);

            // 注册新的写事件,sendBulkToSlave() 传输 RDB 文件
            if (aeCreateFileEvent(server.el, slave->fd, AE_WRITABLE, sendBulkToSlave, slave) == AE_ERR) {
                freeClient(slave);
                continue;
            }
        }
    }

    // startbgsave == REDIS_ERR 表示 BGSAVE 失败,再一次进行 BGSAVE 尝试
    if (startbgsave) {
        /* Since we are starting a new background save for one or more slaves,
         * we flush the Replication Script Cache to use EVAL to propagate every
         * new EVALSHA for the first time, since all the new slaves don't know
         * about previous scripts. */
        replicationScriptCacheFlush();

        if (rdbSaveBackground(server.rdb_filename) != REDIS_OK) {
        /*BGSAVE 可能 fork 失败,所有等待 BGSAVE 的从机都将结束连接。这是 redis 自我保护的措施,fork 失败很可能是内存紧张*/

            listIter li;

            listRewind(server.slaves,&li);
            redisLog(REDIS_WARNING,"SYNC failed. BGSAVE failed");
            while((ln = listNext(&li))) {
                redisClient *slave = ln->value;

                if (slave->replstate == REDIS_REPL_WAIT_BGSAVE_START)
                    freeClient(slave);
            }
        }
    }
}

部分同步

如上所说,无论如何,redis 首先会尝试部分同步。部分同步即把积压空间缓存的数据,即更新记录发送给从机。

从机连接主机后,会主动发起 PSYNC 命令,从机会提供 master_runid 和 offset,主机验证 master_runid 和 offset 是否有效?
验证通过则,进行部分同步:主机返回 +CONTINUE(从机接收后会注册积压数据接收事件),接着发送积压空间数据。

// 连接主机 connectWithMaster() 的时候,会被注册为回调函数
void syncWithMaster(aeEventLoop *el, int fd, void *privdata, int mask) {
    char tmpfile[256], *err;
    int dfd, maxtries = 5;
    int sockerr = 0, psync_result;
    socklen_t errlen = sizeof(sockerr);

    ......

    // 尝试部分同步,主机允许进行部分同步会返回 +CONTINUE,从机接收后注册相应的事件

    /* Try a partial resynchonization. If we don't have a cached master
     * slaveTryPartialResynchronization() will at least try to use PSYNC
     * to start a full resynchronization so that we get the master run id
     * and the global offset, to try a partial resync at the next
     * reconnection attempt. */

    // 函数返回三种状态:
    // PSYNC_CONTINUE:表示会进行部分同步,在 slaveTryPartialResynchronization()
                     // 中已经设置回调函数 readQueryFromClient()
    // PSYNC_FULLRESYNC:全同步,会下载 RDB 文件
    // PSYNC_NOT_SUPPORTED:未知
    psync_result = slaveTryPartialResynchronization(fd);
    if (psync_result == PSYNC_CONTINUE) {
        redisLog(REDIS_NOTICE, "MASTER <-> SLAVE sync: Master accepted a Partial Resynchronization.");
        return;
    }

    // 执行全同步
    ......
}

// 函数返回三种状态:
// PSYNC_CONTINUE:表示会进行部分同步,已经设置回调函数
// PSYNC_FULLRESYNC:全同步,会下载 RDB 文件
// PSYNC_NOT_SUPPORTED:未知
#define PSYNC_CONTINUE 0
#define PSYNC_FULLRESYNC 1
#define PSYNC_NOT_SUPPORTED 2
int slaveTryPartialResynchronization(int fd) {
    char *psync_runid;
    char psync_offset[32];
    sds reply;

    /* Initially set repl_master_initial_offset to -1 to mark the current
     * master run_id and offset as not valid. Later if we'll be able to do
     * a FULL resync using the PSYNC command we'll set the offset at the
     * right value, so that this information will be propagated to the
     * client structure representing the master into server.master. */
    server.repl_master_initial_offset = -1;

    if (server.cached_master) {
    // 缓存了上一次与主机连接的信息,可以尝试进行部分同步,减少数据传输
        psync_runid = server.cached_master->replrunid;
        snprintf(psync_offset,sizeof(psync_offset),"%lld", server.cached_master->reploff+1);
        redisLog(REDIS_NOTICE,"Trying a partial resynchronization (request %s:%s).", psync_runid, psync_offset);
    } else {
    // 未缓存上一次与主机连接的信息,进行全同步
    // psync ? -1 可以获取主机的 master_runid
        redisLog(REDIS_NOTICE,"Partial resynchronization not possible (no cached master)");
        psync_runid = "?";
        memcpy(psync_offset,"-1",3);
    }

    // 向主机发送命令,并接收回复
    /* Issue the PSYNC command */
    reply = sendSynchronousCommand(fd,"PSYNC",psync_runid,psync_offset,NULL);

    // 全同步
    if (!strncmp(reply,"+FULLRESYNC",11)) {
        char *runid = NULL, *offset = NULL;

        /* FULL RESYNC, parse the reply in order to extract the run id
         * and the replication offset. */
        runid = strchr(reply,' ');
        if (runid) {
            runid++;
            offset = strchr(runid,' ');
            if (offset) offset++;
        }
        if (!runid || !offset || (offset-runid-1) != REDIS_RUN_ID_SIZE) {
            redisLog(REDIS_WARNING,
                "Master replied with wrong +FULLRESYNC syntax.");
            /* This is an unexpected condition, actually the +FULLRESYNC
             * reply means that the master supports PSYNC, but the reply
             * format seems wrong. To stay safe we blank the master
             * runid to make sure next PSYNCs will fail. */
            memset(server.repl_master_runid,0,REDIS_RUN_ID_SIZE+1);
        } else {
            // 拷贝 runid
            memcpy(server.repl_master_runid, runid, offset-runid-1);
            server.repl_master_runid[REDIS_RUN_ID_SIZE] = '\0';
            server.repl_master_initial_offset = strtoll(offset,NULL,10);
            redisLog(REDIS_NOTICE,"Full resync from master: %s:%lld",
                server.repl_master_runid,
                server.repl_master_initial_offset);
        }
        /* We are going to full resync, discard the cached master structure. */
        replicationDiscardCachedMaster();
        sdsfree(reply);
        return PSYNC_FULLRESYNC;
    }

    // 部分同步
    if (!strncmp(reply,"+CONTINUE",9)) {
        /* Partial resync was accepted, set the replication state accordingly */
        redisLog(REDIS_NOTICE,
            "Successful partial resynchronization with master.");
        sdsfree(reply);

        // 缓存主机替代现有主机,且为 PSYNC(部分同步) 做好准备c
        replicationResurrectCachedMaster(fd);

        return PSYNC_CONTINUE;
    }

    /* If we reach this point we receied either an error since the master does
     * not understand PSYNC, or an unexpected reply from the master.
     * Reply with PSYNC_NOT_SUPPORTED in both cases. */

    // 接收到主机发出的错误信息
    if (strncmp(reply,"-ERR",4)) {
        /* If it's not an error, log the unexpected event. */
        redisLog(REDIS_WARNING,
            "Unexpected reply to PSYNC from master: %s", reply);
    } else {
        redisLog(REDIS_NOTICE,
            "Master does not support PSYNC or is in "
            "error state (reply: %s)", reply);
    }
    sdsfree(reply);
    replicationDiscardCachedMaster();
    return PSYNC_NOT_SUPPORTED;
}

// 主机 SYNC 和 PSYNC 命令处理函数,会尝试进行部分同步和全同步
/* SYNC ad PSYNC command implemenation. */
void syncCommand(redisClient *c) {
    ......

    // 主机尝试部分同步,允许则进行部分同步,会返回 +CONTINUE,接着发送积压空间

    /* Try a partial resynchronization if this is a PSYNC command.
     * If it fails, we continue with usual full resynchronization, however
     * when this happens masterTryPartialResynchronization() already
     * replied with:
     *
     * +FULLRESYNC <runid> <offset>
     *
     * So the slave knows the new runid and offset to try a PSYNC later
     * if the connection with the master is lost. */
    if (!strcasecmp(c->argv[0]->ptr,"psync")) {
        // 部分同步
        if (masterTryPartialResynchronization(c) == REDIS_OK) {
            server.stat_sync_partial_ok++;
            return; /* No full resync needed, return. */
        } else {
        // 部分同步失败,会进行全同步,这时会收到来自客户端的 runid
            char *master_runid = c->argv[1]->ptr;

            /* Increment stats for failed PSYNCs, but only if the
             * runid is not "?", as this is used by slaves to force a full
             * resync on purpose when they are not albe to partially
             * resync. */
            if (master_runid[0] != '?') server.stat_sync_partial_err++;
        }
    } else {
        /* If a slave uses SYNC, we are dealing with an old implementation
         * of the replication protocol (like redis-cli --slave). Flag the client
         * so that we don't expect to receive REPLCONF ACK feedbacks. */
        c->flags |= REDIS_PRE_PSYNC_SLAVE;
    }

    // 执行全同步:
    ......
}

// 主机尝试是否能进行部分同步
/* This function handles the PSYNC command from the point of view of a
* master receiving a request for partial resynchronization.
*
* On success return REDIS_OK, otherwise REDIS_ERR is returned and we proceed
* with the usual full resync. */
int masterTryPartialResynchronization(redisClient *c) {
    long long psync_offset, psync_len;
    char *master_runid = c->argv[1]->ptr;
    char buf[128];
    int buflen;

    /* Is the runid of this master the same advertised by the wannabe slave
     * via PSYNC? If runid changed this master is a different instance and
     * there is no way to continue. */
    if (strcasecmp(master_runid, server.runid)) {
    // 当因为异常需要与主机断开连接的时候,从机会暂存主机的状态信息,以便
    // 下一次的部分同步。
    // 1)master_runid 是从机提供一个因缓存主机的 runid,
    // 2)server.runid 是本机(主机)的 runid。
    // 匹配失败,说明是本机(主机)不是从机缓存的主机,这时候不能进行部分同步,
    // 只能进行全同步

        // "?" 表示从机要求全同步
        // 什么时候从机会要求全同步???
        /* Run id "?" is used by slaves that want to force a full resync. */
        if (master_runid[0] != '?') {
            redisLog(REDIS_NOTICE,"Partial resynchronization not accepted: "
                "Runid mismatch (Client asked for '%s', I'm '%s')",
                master_runid, server.runid);
        } else {
            redisLog(REDIS_NOTICE,"Full resync requested by slave.");
        }
        goto need_full_resync;
    }

    // 从参数中解析整数,整数是从机指定的偏移量
    /* We still have the data our slave is asking for? */
    if (getLongLongFromObjectOrReply(c,c->argv[2],&psync_offset,NULL) !=
       REDIS_OK) goto need_full_resync;

    // 部分同步失败的情况
    if (!server.repl_backlog || /*不存在积压空间*/
        psync_offset < server.repl_backlog_off ||  /*psync_offset 太过小,
                                                    即从机错过太多更新记录,
                                                    安全起见,实行全同步*/
                                                    /*psync_offset 越界*/
        psync_offset > (server.repl_backlog_off + server.repl_backlog_histlen))
    // 经检测,不满足部分同步的条件,转而进行全同步
    {
        redisLog(REDIS_NOTICE,
            "Unable to partial resync with the slave for lack of backlog (Slave request was: %lld).", psync_offset);
        if (psync_offset > server.master_repl_offset) {
            redisLog(REDIS_WARNING,
                "Warning: slave tried to PSYNC with an offset that is greater than the master replication offset.");
        }
        goto need_full_resync;
    }

    // 执行部分同步:
    // 1)标记客户端为从机
    // 2)通知从机准备接收数据。从机收到 +CONTINUE 会做好准备
    // 3)开发发送数据
    /* If we reached this point, we are able to perform a partial resync:
     * 1) Set client state to make it a slave.
     * 2) Inform the client we can continue with +CONTINUE
     * 3) Send the backlog data (from the offset to the end) to the slave. */

    // 将连接的客户端标记为从机
    c->flags |= REDIS_SLAVE;

    // 表示进行部分同步
    // #define REDIS_REPL_ONLINE 9 /* RDB file transmitted, sending just
    // updates. */
    c->replstate = REDIS_REPL_ONLINE;

    // 更新 ack 的时间
    c->repl_ack_time = server.unixtime;

    // 添加入从机链表
    listAddNodeTail(server.slaves,c);

    // 告诉从机可以进行部分同步,从机收到后会做相关的准备(注册回调函数)
    /* We can't use the connection buffers since they are used to accumulate
     * new commands at this stage. But we are sure the socket send buffer is
     * emtpy so this write will never fail actually. */
    buflen = snprintf(buf,sizeof(buf),"+CONTINUE\r\n");
    if (write(c->fd,buf,buflen) != buflen) {
        freeClientAsync(c);
        return REDIS_OK;
    }

    // 向从机写积压空间中的数据,积压空间存储有「更新缓存」
    psync_len = addReplyReplicationBacklog(c,psync_offset);

    redisLog(REDIS_NOTICE,
        "Partial resynchronization request accepted. Sending %lld bytes of backlog starting from offset %lld.", psync_len, psync_offset);
    /* Note that we don't need to set the selected DB at server.slaveseldb
     * to -1 to force the master to emit SELECT, since the slave already
     * has this state from the previous connection with the master. */

    refreshGoodSlavesCount();
    return REDIS_OK; /* The caller can return, no full resync needed. */

need_full_resync:
    ......
    // 向从机发送 +FULLRESYNC runid repl_offset
}

暂缓主机

从机因为某些原因,譬如网络延迟(PING 超时,ACK 超时等),可能会断开与主机的连接。这时候,从机会尝试保存与主机连接的信息,譬如全局积压空间数据偏移量等,以便下一次的部分同步,并且从机会再一次尝试连接主机。注意一点,如果断开的时间足够长, 部分同步肯定会失败的。

void freeClient(redisClient *c) {
    listNode *ln;

    /* If this is marked as current client unset it */
    if (server.current_client == c) server.current_client = NULL;

    // 如果此机为从机,已经连接主机,可能需要保存主机状态信息,以便进行 PSYNC
    /* If it is our master that's beging disconnected we should make sure
     * to cache the state to try a partial resynchronization later.
     *
     * Note that before doing this we make sure that the client is not in
     * some unexpected state, by checking its flags. */
    if (server.master && c->flags & REDIS_MASTER) {
        redisLog(REDIS_WARNING,"Connection with master lost.");
        if (!(c->flags & (REDIS_CLOSE_AFTER_REPLY|
                          REDIS_CLOSE_ASAP|
                          REDIS_BLOCKED|
                          REDIS_UNBLOCKED)))
        {
            replicationCacheMaster(c);
            return;
        }
    }
    ......
}

// 为了实现部分同步,从机会保存主机的状态信息后才会断开主机的连接,主机状态信息
// 保存在 server.cached_master
// 会在 freeClient() 中调用,保存与主机连接的状态信息,以便进行 PSYNC
void replicationCacheMaster(redisClient *c) {
    listNode *ln;

    redisAssert(server.master != NULL && server.cached_master == NULL);
    redisLog(REDIS_NOTICE,"Caching the disconnected master state.");

    // 从客户端列表删除主机的信息
    /* Remove from the list of clients, we don't want this client to be
     * listed by CLIENT LIST or processed in any way by batch operations. */
    ln = listSearchKey(server.clients,c);
    redisAssert(ln != NULL);
    listDelNode(server.clients,ln);

    // 保存主机的状态信息
    /* Save the master. Server.master will be set to null later by
     * replicationHandleMasterDisconnection(). */
    server.cached_master = server.master;

    // 注销事件,关闭连接
    /* Remove the event handlers and close the socket. We'll later reuse
     * the socket of the new connection with the master during PSYNC. */
    aeDeleteFileEvent(server.el,c->fd,AE_READABLE);
    aeDeleteFileEvent(server.el,c->fd,AE_WRITABLE);
    close(c->fd);

    /* Set fd to -1 so that we can safely call freeClient(c) later. */
    c->fd = -1;

    // 修改连接的状态,设置 server.master = NULL
    /* Caching the master happens instead of the actual freeClient() call,
     * so make sure to adjust the replication state. This function will
     * also set server.master to NULL. */
    replicationHandleMasterDisconnection();
}

总结

简单来说,主从同步就是 RDB 文件的上传下载;主机有小部分的数据修改,就把修改记录传播给每个从机。这篇文章详述了 redis 主从复制的内部协议和机制。接下来的几篇关于 redis 的文章,主要是其内部数据结构。

捣乱 2014-4-22

http://daoluan.net