memcached源代码阅读笔记(6)memcached1.4.10 在linux平台多线程模型研究

为了方便gdb调试,按照如下重新编译
./configure --prefix=/usr/local/memcached --with-libevent=/usr/local/libevent CPPFLAGS='-ggdb3'

make && make install

gdb memcached
gdb>set args -p 11211 -m 64 -uroot -t2 -vv

然后

gdb>b main

然后s,n,u,disp,printf跟踪。

这里仅研究TCP的情况,udp和unix域基本差不多。

main()

thread_init(settings.num_threads, main_base);

这是初始化线程

if (start_assoc_maintenance_thread() == -1) {
        exit(EXIT_FAILURE);
 }这个是开始一个hash批量维护的线程,用于扩展hash表。

 if (settings.port && server_sockets(settings.port, tcp_transport,
                                           portnumber_file))

void thread_init(int nthreads, struct event_base *main_base) {
    int         i;
    int         power;

    pthread_mutex_init(&cache_lock, NULL);
    pthread_mutex_init(&stats_lock, NULL);

    pthread_mutex_init(&init_lock, NULL);
    pthread_cond_init(&init_cond, NULL);

    pthread_mutex_init(&cqi_freelist_lock, NULL);
    cqi_freelist = NULL;

    /* Want a wide lock table, but don't waste memory */
    if (nthreads < 3) {
        power = 10;
    } else if (nthreads < 4) {
        power = 11;
    } else if (nthreads < 5) {
        power = 12;
    } else {
        /* 8192 buckets, and central locks don't scale much past 5 threads */
        power = 13;
    }

    item_lock_count = ((unsigned long int)1 << (power));
    item_lock_mask  = item_lock_count - 1;

    item_locks = calloc(item_lock_count, sizeof(pthread_mutex_t));
    if (! item_locks) {
        perror("Can't allocate item locks");
        exit(1);
    }
    for (i = 0; i < item_lock_count; i++) {
        pthread_mutex_init(&item_locks[i], NULL);
    }

    threads = calloc(nthreads, sizeof(LIBEVENT_THREAD));
    if (! threads) {
        perror("Can't allocate thread descriptors");
        exit(1);
    }

    dispatcher_thread.base = main_base;
    dispatcher_thread.thread_id = pthread_self();

    for (i = 0; i < nthreads; i++) {
        int fds[2];
        if (pipe(fds)) {
            perror("Can't create notify pipe");
            exit(1);
        }

        threads[i].notify_receive_fd = fds[0];
        threads[i].notify_send_fd = fds[1];

        setup_thread(&threads[i]);
        /* Reserve three fds for the libevent base, and two for the pipe */
        stats.reserved_fds += 5;
    }

    /* Create threads after we've done all the libevent setup. */
    for (i = 0; i < nthreads; i++) {
        create_worker(worker_libevent, &threads[i]);
    }

    /* Wait for all the threads to set themselves up before returning. */
    pthread_mutex_lock(&init_lock);
    while (init_count < nthreads) {
        pthread_cond_wait(&init_cond, &init_lock);
    }
    pthread_mutex_unlock(&init_lock);
}


    dispatcher_thread.base = main_base;
dispatcher_thread.thread_id = pthread_self();
这说明dispatcher_thread是主线程

if (pipe(fds)) {
            perror("Can't create notify pipe");
            exit(1);
        }

threads[i].notify_receive_fd = fds[0];
threads[i].notify_send_fd = fds[1];
为每个工作线程建立通知管道
 setup_thread(&threads[i]);
初始化工作线程
create_worker(worker_libevent, &threads[i]);
创建工作线程
pthread_mutex_lock(&init_lock);
    while (init_count < nthreads) {
        pthread_cond_wait(&init_cond, &init_lock);
    }
    pthread_mutex_unlock(&init_lock);
等待所有工作线程创建完毕。


static void setup_thread(LIBEVENT_THREAD *me) {
    me->base = event_init();
    if (! me->base) {
        fprintf(stderr, "Can't allocate event base\n");
        exit(1);thread_libevent_process
    }

    /* Listen for notifications from other threads */
    event_set(&me->notify_event, me->notify_receive_fd,
              EV_READ | EV_PERSIST, thread_libevent_process, me);

为通知管道设置响应函数
    event_base_set(me->base, &me->notify_event);

    if (event_add(&me->notify_event, 0) == -1) {
        fprintf(stderr, "Can't monitor libevent notify pipe\n");
        exit(1);
    }

    me->new_conn_queue = malloc(sizeof(struct conn_queue));
    if (me->new_conn_queue == NULL) {
        perror("Failed to allocate memory for connection queue");
        exit(EXIT_FAILURE);
    }
    cq_init(me->new_conn_queue);
初始化工作队列
    if (pthread_mutex_init(&me->stats.mutex, NULL) != 0) {
        perror("Failed to initialize mutex");
        exit(EXIT_FAILURE);
    }

    me->suffix_cache = cache_create("suffix", SUFFIX_SIZE, sizeof(char*),
                                    NULL, NULL);
    if (me->suffix_cache == NULL) {
        fprintf(stderr, "Failed to create suffix cache\n");
        exit(EXIT_FAILURE);
    }


工作线程收到通知后的响应函数
static void thread_libevent_process(int fd, short which, void *arg) {
    LIBEVENT_THREAD *me = arg;
    CQ_ITEM *item;
    char buf[1];

    if (read(fd, buf, 1) != 1)//查看是否收到正确的管道通知
        if (settings.verbose > 0)
            fprintf(stderr, "Can't read from libevent pipe\n");

    item = cq_pop(me->new_conn_queue);//弹出连接队列

    if (NULL != item) {
        conn *c = conn_new(item->sfd, item->init_state, item->event_flags,
                           item->read_buffer_size, item->transport, me->base);创建一个新的连接,当然这个新的连接的状态不再是Listening
        if (c == NULL) {
            if (IS_UDP(item->transport)) {
                fprintf(stderr, "Can't listen for events on UDP socket\n");
                exit(1);
            } else {
                if (settings.verbose > 0) {
                    fprintf(stderr, "Can't listen for events on fd %d\n",
                        item->sfd);
                }
                close(item->sfd);
            }
        } else {
            c->thread = me;
        }
        cqi_free(item);
    }
}

static int server_socket(const char *interface,
                         int port,
                         enum network_transport transport,
                         FILE *portnumber_file) {
    int sfd;
    struct linger ling = {0, 0};
    struct addrinfo *ai;
    struct addrinfo *next;
    struct addrinfo hints = { .ai_flags = AI_PASSIVE,
                              .ai_family = AF_UNSPEC };
    char port_buf[NI_MAXSERV];
    int error;
    int success = 0;
    int flags =1;

    hints.ai_socktype = IS_UDP(transport) ? SOCK_DGRAM : SOCK_STREAM;

    if (port == -1) {
        port = 0;
    }
    snprintf(port_buf, sizeof(port_buf), "%d", port);
    error= getaddrinfo(interface, port_buf, &hints, &ai);
    if (error != 0) {
        if (error != EAI_SYSTEM)
          fprintf(stderr, "getaddrinfo(): %s\n", gai_strerror(error));
        else
          perror("getaddrinfo()");
        return 1;
    }

    for (next= ai; next; next= next->ai_next) {
        conn *listen_conn_add;
        if ((sfd = new_socket(next)) == -1) {
            /* getaddrinfo can return "junk" addresses,
             * we make sure at least one works before erroring.
             */
            if (errno == EMFILE) {
                /* ...unless we're out of fds */
                perror("server_socket");
                exit(EX_OSERR);
            }
            continue;
        }

#ifdef IPV6_V6ONLY
        if (next->ai_family == AF_INET6) {
            error = setsockopt(sfd, IPPROTO_IPV6, IPV6_V6ONLY, (char *) &flags, sizeof(flags));
            if (error != 0) {
                perror("setsockopt");
                close(sfd);
                continue;
            }
        }
#endif

        setsockopt(sfd, SOL_SOCKET, SO_REUSEADDR, (void *)&flags, sizeof(flags));
        if (IS_UDP(transport)) {
            maximize_sndbuf(sfd);
        } else {
            error = setsockopt(sfd, SOL_SOCKET, SO_KEEPALIVE, (void *)&flags, sizeof(flags));
            if (error != 0)
                perror("setsockopt");

            error = setsockopt(sfd, SOL_SOCKET, SO_LINGER, (void *)&ling, sizeof(ling));
            if (error != 0)
                perror("setsockopt");

            error = setsockopt(sfd, IPPROTO_TCP, TCP_NODELAY, (void *)&flags, sizeof(flags));
            if (error != 0)
                perror("setsockopt");
        }

        if (bind(sfd, next->ai_addr, next->ai_addrlen) == -1) {
            if (errno != EADDRINUSE) {
                perror("bind()");
                close(sfd);
                freeaddrinfo(ai);
                return 1;
            }
            close(sfd);
            continue;
        } else {
            success++;
            if (!IS_UDP(transport) && listen(sfd, settings.backlog) == -1) {
                perror("listen()");
                close(sfd);
                freeaddrinfo(ai);
                return 1;
            }
            if (portnumber_file != NULL &&
                (next->ai_addr->sa_family == AF_INET ||
                 next->ai_addr->sa_family == AF_INET6)) {
                union {
                    struct sockaddr_in in;
                    struct sockaddr_in6 in6;
                } my_sockaddr;
                socklen_t len = sizeof(my_sockaddr);
                if (getsockname(sfd, (struct sockaddr*)&my_sockaddr, &len)==0) {
                    if (next->ai_addr->sa_family == AF_INET) {
                        fprintf(portnumber_file, "%s INET: %u\n",
                                IS_UDP(transport) ? "UDP" : "TCP",
                                ntohs(my_sockaddr.in.sin_port));
                    } else {
                        fprintf(portnumber_file, "%s INET6: %u\n",
                                IS_UDP(transport) ? "UDP" : "TCP",
                                ntohs(my_sockaddr.in6.sin6_port));
                    }
                }
            }
        }

        if (IS_UDP(transport)) {
            int c;

            for (c = 0; c < settings.num_threads_per_udp; c++) {
                /* this is guaranteed to hit all threads because we round-robin */
                dispatch_conn_new(sfd, conn_read, EV_READ | EV_PERSIST,
                                  UDP_READ_BUFFER_SIZE, transport);
            }
        } else {
            if (!(listen_conn_add = conn_new(sfd, conn_listening,
                                             EV_READ | EV_PERSIST, 1,
                                             transport, main_base)))
//这里是设置listenging状态 {
                fprintf(stderr, "failed to create listening connection\n");
                exit(EXIT_FAILURE);
            }
            listen_conn_add->next = listen_conn;
            listen_conn = listen_conn_add;
        }
    }

    freeaddrinfo(ai);

    /* Return zero iff we detected no errors in starting up connections */
    return success == 0;
}

conn *conn_new(const int sfd, enum conn_states init_state,
                const int event_flags,
                const int read_buffer_size, enum network_transport transport,
                struct event_base *base) {
    conn *c = conn_from_freelist();

    if (NULL == c) {
        if (!(c = (conn *)calloc(1, sizeof(conn)))) {
            fprintf(stderr, "calloc()\n");
            return NULL;
        }
        MEMCACHED_CONN_CREATE(c);

        c->rbuf = c->wbuf = 0;
        c->ilist = 0;
        c->suffixlist = 0;
        c->iov = 0;
        c->msglist = 0;
        c->hdrbuf = 0;

        c->rsize = read_buffer_size;
        c->wsize = DATA_BUFFER_SIZE;
        c->isize = ITEM_LIST_INITIAL;
        c->suffixsize = SUFFIX_LIST_INITIAL;
        c->iovsize = IOV_LIST_INITIAL;
        c->msgsize = MSG_LIST_INITIAL;
        c->hdrsize = 0;

        c->rbuf = (char *)malloc((size_t)c->rsize);
        c->wbuf = (char *)malloc((size_t)c->wsize);
        c->ilist = (item **)malloc(sizeof(item *) * c->isize);
        c->suffixlist = (char **)malloc(sizeof(char *) * c->suffixsize);
        c->iov = (struct iovec *)malloc(sizeof(struct iovec) * c->iovsize);
        c->msglist = (struct msghdr *)malloc(sizeof(struct msghdr) * c->msgsize);

        if (c->rbuf == 0 || c->wbuf == 0 || c->ilist == 0 || c->iov == 0 ||
                c->msglist == 0 || c->suffixlist == 0) {
            conn_free(c);
            fprintf(stderr, "malloc()\n");
            return NULL;
        }

        STATS_LOCK();
        stats.conn_structs++;
        STATS_UNLOCK();
    }

    c->transport = transport;
    c->protocol = settings.binding_protocol;

    /* unix socket mode doesn't need this, so zeroed out.  but why
     * is this done for every command?  presumably for UDP
     * mode.  */
    if (!settings.socketpath) {
        c->request_addr_size = sizeof(c->request_addr);
    } else {
        c->request_addr_size = 0;
    }

    if (settings.verbose > 1) {
        if (init_state == conn_listening) {
            fprintf(stderr, "<%d server listening (%s)\n", sfd,
                prot_text(c->protocol));
        } else if (IS_UDP(transport)) {
            fprintf(stderr, "<%d server listening (udp)\n", sfd);
        } else if (c->protocol == negotiating_prot) {
            fprintf(stderr, "<%d new auto-negotiating client connection\n",
                    sfd);
        } else if (c->protocol == ascii_prot) {
            fprintf(stderr, "<%d new ascii client connection.\n", sfd);
        } else if (c->protocol == binary_prot) {
            fprintf(stderr, "<%d new binary client connection.\n", sfd);
        } else {
            fprintf(stderr, "<%d new unknown (%d) client connection\n",
                sfd, c->protocol);
            assert(false);
        }
    }

    c->sfd = sfd;
    c->state = init_state;
    c->rlbytes = 0;
    c->cmd = -1;
    c->rbytes = c->wbytes = 0;
    c->wcurr = c->wbuf;
    c->rcurr = c->rbuf;
    c->ritem = 0;
    c->icurr = c->ilist;
    c->suffixcurr = c->suffixlist;
    c->ileft = 0;
    c->suffixleft = 0;
    c->iovused = 0;
    c->msgcurr = 0;
    c->msgused = 0;

    c->write_and_go = init_state;
    c->write_and_free = 0;
    c->item = 0;

    c->noreply = false;

    event_set(&c->event, sfd, event_flags, event_handler, (void *)c);

    event_base_set(base, &c->event);
这里设置处理函数是eventhandler


    c->ev_flags = event_flags;

    if (event_add(&c->event, 0) == -1) {
        if (conn_add_to_freelist(c)) {
            conn_free(c);
        }
        perror("event_add");
        return NULL;
    }

    STATS_LOCK();
    stats.curr_conns++;
    stats.total_conns++;
    STATS_UNLOCK();

    MEMCACHED_CONN_ALLOCATE(c->sfd);

    return c;
}

eventhandler函数很简单
void event_handler(const int fd, const short which, void *arg) {
    conn *c;

    c = (conn *)arg;
    assert(c != NULL);

    c->which = which;

    /* sanity */
    if (fd != c->sfd) {
        if (settings.verbose > 0)
            fprintf(stderr, "Catastrophic: event fd doesn't match conn fd!\n");
        conn_close(c);
        return;
    }

    drive_machine(c);

    /* wait for next event */
    return;
}
都调用了drive_machine(c);
static void drive_machine(conn *c) {
    bool stop = false;
    int sfd, flags = 1;
    socklen_t addrlen;
    struct sockaddr_storage addr;
    int nreqs = settings.reqs_per_event;
    int res;
    const char *str;

    assert(c != NULL);

    static void drive_machine(conn *c) {
    bool stop = false;
    int sfd, flags = 1;
    socklen_t addrlen;
    struct sockaddr_storage addr;
    int nreqs = settings.reqs_per_event;
    int res;
    const char *str;

    assert(c != NULL);

    while (!stop) {

        switch(c->state) {
        case conn_listening:
            addrlen = sizeof(addr);
            if ((sfd = accept(c->sfd, (struct sockaddr *)&addr, &addrlen)) == -1) {
                if (errno == EAGAIN || errno == EWOULDBLOCK) {
                    /* these are transient, so don't log anything */
                    stop = true;
                } else if (errno == EMFILE) {
                    if (settings.verbose > 0)
                        fprintf(stderr, "Too many open connections\n");
                    accept_new_conns(false);
                    stop = true;
                } else {
                    perror("accept()");
                    stop = true;
                }
                break;
            }
            if ((flags = fcntl(sfd, F_GETFL, 0)) < 0 ||
                fcntl(sfd, F_SETFL, flags | O_NONBLOCK) < 0) {
                perror("setting O_NONBLOCK");
                close(sfd);
                break;
            }

            if (settings.maxconns_fast &&
                stats.curr_conns + stats.reserved_fds >= settings.maxconns - 1) {
                str = "ERROR Too many open connections\r\n";
                res = write(sfd, str, strlen(str));
                close(sfd);
                STATS_LOCK();
                stats.rejected_conns++;
                STATS_UNLOCK();
            } else {
                dispatch_conn_new(sfd, conn_new_cmd, EV_READ | EV_PERSIST,
                                     DATA_BUFFER_SIZE, tcp_transport);

这个函数开始向工作线程派发一个链接


void dispatch_conn_new(int sfd, enum conn_states init_state, int event_flags,
                       int read_buffer_size, enum network_transport transport) {
    CQ_ITEM *item = cqi_new();
    int tid = (last_thread + 1) % settings.num_threads;
//robin round模式轮询线程,这是最快的处理方式,也可以根据各个线程的queue number来分配,不过没太大必要,所以采用最简单的方式
    LIBEVENT_THREAD *thread = threads + tid;

    last_thread = tid;

    item->sfd = sfd;
    item->init_state = init_state;
    item->event_flags = event_flags;
    item->read_buffer_size = read_buffer_size;
    item->transport = transport;

    cq_push(thread->new_conn_queue, item);//向工作线程插入工作向
    MEMCACHED_CONN_DISPATCH(sfd, thread->thread_id);
    if (write(thread->notify_send_fd, "", 1) != 1) {//向通知管道写入数据。
        perror("Writing to thread notify pipe");
    }
}
            }

前面工作线程函数接收管道通知后就开始建立连接,连接的状态设置为EV_READ | EV_PERSIST
由此可知始终只有一个listening线程,其他都是处理工作线程。

原文地址:https://www.cnblogs.com/yanzhenan/p/2288023.html