redis多线程模型
redis为什么引入I/O多线程
Redis 的性能瓶颈在网络 IO 的处理上。Redis 是网络 IO 密集型,需要同时处理多条并发请求,读写 IO 的问题(请求大量数据,写日志业务等)。多线程处理网络 IO,单线程执行命令。
Redis 线程池作用读写 IO 阶段,即 read, decode 和 encode, send 阶段。主线程处理业务逻辑,之所以用单线程执行命令,是因为 Redis 采用高效的数据结构,其业务逻辑处理较快。
I/O多线程模型
主线程拥有两个全局队列clients_pending_read
和clients_pending_write
,每个 io 线程(主线程同时也是 io 线程)拥有一个专属队列 io_threads_list[id]
。主线程既作为生产者,产生任务;又作为消费者,获取任务执行。
首先,主线程将一次循环的所有就绪的读事件收集到自己的全局任务队列clients_pending_read
中,再把每个事件负载均衡地分配到每个 io 线程的专属任务队列中。一次事件循环中不会出现同名 fd,不同的 fd 分配到每个 io 线程各自的队列中,避免了多个 io 线程同时从全局队列中取数据,因此,不需要加锁操作。
接下来,io 线程从自己的专属队列中取出任务,(除主线程外)并发执行 read 和 decode 操作。主线程将解析后的任务做 compute 操作。最后,io 线程(包括主线程)并发执行 encode 和 send 操作。
redis的单线程是指,命令执行(logic)都是在单线程中运行的
接受数据read和发送数据write都是可以在io多线程(线程池)中去运行
在Redis中,生产者也可以作为消费者,反之亦然,没有明确界限。
源码解析
测试设置
redis 线程池默认作用在 encode, send 阶段,这是因为客户端从 redis 获取大量数据需要并发处理。若想作用在 read, decode 阶段,需要手动开启。在 redis.conf 文件中,可以设置:
# 开启io线程的数量 io-threads 4 # 优化:read deconde 过程。默认优化,encode send从 redis 获取大量数据 io-threads-do-reads yes
开启 io 多线程的前提是有多个并发连接。如何在单个连接的情况下,开启 io 多线程调试,需要修改 redis 源码:
// networking.c int stopThreadedIOIfNeeded(void) { // 单个连接的情况下,开启多线程调试,永远不关闭 io 多线程 return 0; ... }
连接建立
主线程处理连接建立,listenfd
- 连接到达,触发读事件回调:acceptTcpHandler
- 接收连接:acceptTcpHandler
- 初始化新连接:createClient
// server.c void initServer(void) { ... // 1、连接到来,触发读事件回调 if (aeCreateFileEvent(server.el, server.ipfd[j], AE_READABLE, acceptTcpHandler,NULL) == AE_ERR) ... } // networking.c void acceptTcpHandler(aeEventLoop *el, int fd, void *privdata, int mask) { ... while(max--) { // 2、接收连接:内部封装 accept cfd = anetTcpAccept(server.neterr, fd, cip, sizeof(cip), &cport); ... // 为 cfd 初始化新连接,内部调用 createClient acceptCommonHandler(connCreateAcceptedSocket(cfd),0,cip); } } static void acceptCommonHandler(connection *conn, int flags, char *ip) { ... /* Create connection and client */ // 3、创建新的连接 if ((c = createClient(conn)) == NULL) { ... } ... } client *createClient(connection *conn) { client *c = zmalloc(sizeof(client)); /* passing NULL as conn it is possible to create a non connected client. * This is useful since all the commands needs to be executed * in the context of a client. When commands are executed in other * contexts (for instance a Lua script) we need a non connected client. */ if (conn) { connNonBlock(conn); connEnableTcpNoDelay(conn); if (server.tcpkeepalive) connKeepAlive(conn,server.tcpkeepalive); // 4.接收数据的读事件触发,回调readQueryFromClient函数 connSetReadHandler(conn, readQueryFromClient); connSetPrivateData(conn, c); } }
数据传输
clientfd
- 读事件回调:readQueryFromClient
- 分割并处理数据包 processInputBuffer
- 分割数据包:processInlineBuffer 和 processMultibulkBuffer
- 处理数据包:processCommandAndResetClient
- 数据写到 buffer:addReply
- 数据写到 socket:writeToClient
- 写事件回调:sendReplyToClient
当读事件触发时,执行读事件回调函数。主线程收集读事件就绪的连接放入全局任务队列``clients_pending_read,并设置连接状态为
CLIENT_PENDING_READ`。子线程从该全局队列中获取任务后,也调用该读事件回调函数,进行 read 和 decode 的业务逻辑处理。
// networking.c void readQueryFromClient(connection *conn) { ... /* Check if we want to read from the client later when exiting from * the event loop. This is the case if threaded I/O is enabled. */ // 开启 io 线程后,延迟处理客户端的读,将任务丢到全局队列,再分配给 io 线程 // 主线程返回 1,不执行业务逻辑处理; // 子线程返回 0,继续往下,执行业务逻辑处理 if (postponeClientRead(c)) return; // 1、read 阶段,(io 线程)将任务读到缓冲区 nread = connRead(c->conn, c->querybuf+qblen, readlen); // 2、decode 阶段,(io 线程)解析数据包 processInputBuffer(c); } int postponeClientRead(client *c) { if (server.io_threads_active && server.io_threads_do_reads && !clientsArePaused() && !ProcessingEventsWhileBlocked && !(c->flags & (CLIENT_MASTER|CLIENT_SLAVE|CLIENT_PENDING_READ))) { // 主线程,返回 1 // 将连接状态设置为 CLIENT_PENDING_READ c->flags |= CLIENT_PENDING_READ; // 收集任务,把客户端连接放到全局队列中,后续会分配到 io 线程 listAddNodeHead(server.clients_pending_read,c); return 1; } else { // 子线程,即 io 线程,返回 0 return 0; } }
子线程(IO 线程)从专属任务队列 io_threads_pending
获取任务,执行 read decode 和 encode write 业务逻辑处理。
// networking.c // 线程池入口函数:子线程 void *IOThreadMain(void *myid) { ... while(1) { /* Wait for start */ // 等待获取专属任务队列中的任务 for (int j = 0; j < 1000000; j++) { if (io_threads_pending[id] != 0) break; } ... /* Process: note that the main thread will never touch our list * before we drop the pending count to 0. */ listIter li; listNode *ln; // 从专属任务队列中取出任务 listRewind(io_threads_list[id],&li); while((ln = listNext(&li))) { client *c = listNodeValue(ln); if (io_threads_op == IO_THREADS_OP_WRITE) { // encode 和 write writeToClient(c,0); } else if (io_threads_op == IO_THREADS_OP_READ) { // read 和 decode,读事件回调函数 readQueryFromClient(c->conn); } else { serverPanic("io_threads_op value is unknown"); } } listEmpty(io_threads_list[id]); io_threads_pending[id] = 0; ... } }
子线程 decode 结束后,设置连接状态 CLIENT_PENDING_COMMAND
,交给主线程来 compute,退出读事件回调函数。主线程负责 compute ,解析 redis 命令。
// networking.c // readQueryFromClient 函数中 decode 阶段调用 void processInputBuffer(client *c) { /* Keep processing while there is something in the input buffer */ while(c->qb_pos < sdslen(c->querybuf)) { ... if (c->reqtype == PROTO_REQ_INLINE) { // 分割数据包。并判断是否完整 if (processInlineBuffer(c) != C_OK) break; ... } else if (c->reqtype == PROTO_REQ_MULTIBULK) { // 分割 pipline 的数据包,并判断是否完整 if (processMultibulkBuffer(c) != C_OK) break; } ... else { /* If we are in the context of an I/O thread, we can't really * execute the command here. All we can do is to flag the client * as one that needs to process the command. */ // io 线程设置任务状态,交给主线程compute,退出读事件回调函数 if (c->flags & CLIENT_PENDING_READ) { c->flags |= CLIENT_PENDING_COMMAND; break; } /* We are finally ready to execute the command. */ // 3、compute,主线程解析命令 if (processCommandAndResetClient(c) == C_ERR) { /* If the client is no longer valid, we avoid exiting this * loop and trimming the client buffer later. So we return * ASAP in that case. */ return; } } } ... }
主线程 compute 结束后,调用 addReply 函数,将处理完的连接放到全局任务队列clients_pending_write
,并将待发送的数据写到缓冲区。
// networking.c int processCommandAndResetClient(client *c) { ... // 处理命令 if (processCommand(c) == C_OK) { commandProcessed(c); } ... } // server.c int processCommand(client *c) { ... /* Exec the command */ // 开启 io 多线程,且不是事务命令 if (c->flags & CLIENT_MULTI && c->cmd->proc != execCommand && c->cmd->proc != discardCommand && c->cmd->proc != multiCommand && c->cmd->proc != watchCommand) { // 把数据写到缓冲区 addReply(c,shared.queued); } else { // 执行 redis 命令 call(c,CMD_CALL_FULL); ... } ... } // networking.c // 数据写到发送缓冲区 void addReply(client *c, robj *obj) { if (prepareClientToWrite(c) != C_OK) return; ... } int prepareClientToWrite(client *c) { ... if (!clientHasPendingReplies(c) && !(c->flags & CLIENT_PENDING_READ)) clientInstallWriteHandler(c); // 任务写到全局队列中 ... }
接下来,子线程和主线程都可以从自己的专属任务队列中获得该任务,执行 encode 和 send 的业务逻辑处理 writeToClient。若数据未发送完,则注册写事件回调,等待再次发送。
// 子线程:线程池入口函数 void *IOThreadMain(void *myid) { ... if (io_threads_op == IO_THREADS_OP_WRITE) { // encode 和 write writeToClient(c,0); // 数据写到 socket } else if (io_threads_op == IO_THREADS_OP_READ) { // read 和 decode readQueryFromClient(c->conn); // 读事件回调函数 ... } // 主线程 int handleClientsWithPendingWritesUsingThreads(void) { int processed = listLength(server.clients_pending_write); if (processed == 0) return 0; /* Return ASAP if there are no clients. */ /* If I/O threads are disabled or we have few clients to serve, don't * use I/O threads, but the boring synchronous code. */ if (server.io_threads_num == 1 || stopThreadedIOIfNeeded()) {//判断是否有必要开启IO多线程 return handleClientsWithPendingWrites(); } /* Start threads if needed. */ if (!server.io_threads_active) startThreadedIO();//开启io多线程 /* Distribute the clients across N different lists. */ listIter li; listNode *ln; listRewind(server.clients_pending_write,&li);//创建一个迭代器li,用于遍历任务队列clients_pending_write int item_id = 0;//默认是0,先分配给主线程去做(生产者也可能是消费者),如果设置成1,则先让io线程1去做 //io_threads_list[0] 主线程 //io_threads_list[1] io线程 //io_threads_list[2] io线程 //io_threads_list[3] io线程 //io_threads_list[4] io线程 while((ln = listNext(&li))) { client *c = listNodeValue(ln);//取出一个任务 c->flags &= ~CLIENT_PENDING_WRITE; /* Remove clients from the list of pending writes since * they are going to be closed ASAP. */ if (c->flags & CLIENT_CLOSE_ASAP) {//表示该客户端的输出缓冲区超过了服务器允许范围,将在下一次循环进行一个关闭,也不返回任何信息给客户端,删除待读客户端 listDelNode(server.clients_pending_write, ln); continue; } /* Since all replicas and replication backlog use global replication * buffer, to guarantee data accessing thread safe, we must put all * replicas client into io_threads_list[0] i.e. main thread handles * sending the output buffer of all replicas. */ if (getClientType(c) == CLIENT_TYPE_SLAVE) { listAddNodeTail(io_threads_list[0],c); continue; } //负载均衡:将任务队列中的任务 添加 到不同的线程消费队列中去,每个线程就可以从当前线程的消费队列中取任务就行了 //这样做的好处是,避免加锁。当前是在主线程中,进行分配任务 //通过取余操作,将任务均分给不同io线程 int target_id = item_id % server.io_threads_num; listAddNodeTail(io_threads_list[target_id],c); item_id++; } /* Give the start condition to the waiting threads, by setting the * start condition atomic var. */ io_threads_op = IO_THREADS_OP_WRITE; for (int j = 1; j < server.io_threads_num; j++) { int count = listLength(io_threads_list[j]); setIOPendingCount(j, count);//设置io线程启动条件,启动io线程 } /* Also use the main thread to process a slice of clients. */ // 让主线程去处理一部分任务 listRewind(io_threads_list[0],&li); while((ln = listNext(&li))) { client *c = listNodeValue(ln); writeToClient(c,0); } listEmpty(io_threads_list[0]); /* Wait for all the other threads to end their work. */ while(1) {//剩下的任务io_threads_list[1],io_threads_list[2].....给io线程去做,等待io线程完成任务 unsigned long pending = 0; for (int j = 1; j < server.io_threads_num; j++) pending += getIOPendingCount(j);//等待io线程结束,并返回处理的数量 if (pending == 0) break; } io_threads_op = IO_THREADS_OP_IDLE; /* 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); /* Install the write handler if there are pending writes in some * of the clients. */ // 数据没写完,注册写事件回调 if (clientHasPendingReplies(c) && connSetWriteHandler(c->conn, sendReplyToClient) == AE_ERR) { freeClientAsync(c); } } listEmpty(server.clients_pending_write); ... }
负载均衡:将任务队列中的任务 添加 到不同的线程消费队列中去,每个线程就可以从当前线程的消费队列中取任务就行了。这样做的好处是,避免加锁。当前是在主线程中,进行分配任务通过取余操作,将任务均分给不同的io线程。
线程调度
开启io线程startThreadedIO
每个io线程都有一把锁,如果主线程把锁还回去了,那么io线程就会启动,不再阻塞
并设置io线程标识为活跃状态io_threads_active=1
void startThreadedIO(void) { serverAssert(server.io_threads_active == 0); for (int j = 1; j < server.io_threads_num; j++) pthread_mutex_unlock(&io_threads_mutex[j]); server.io_threads_active = 1; }
关闭io线程stopThreadedIO
每个io线程都有一把锁,如果主线程拿了,那么io线程就会阻塞等待,也就是停止了IO线程
并设置io线程标识为非活跃状态io_threads_active=0
void stopThreadedIO(void) { /* We may have still clients with pending reads when this function * is called: handle them before stopping the threads. */ handleClientsWithPendingReadsUsingThreads(); serverAssert(server.io_threads_active == 1); for (int j = 1; j < server.io_threads_num; j++) pthread_mutex_lock(&io_threads_mutex[j]);// server.io_threads_active = 0; }