第一步，根据memcached的启动参数做校验参数、配置。

main函数中，几乎600行代码都是这些参数校验。吐槽一个。

第二步，初始化。

　　2.1：初始化主线程的libevent。

main_base = event_init();

　　2.2：初始化memcached的stats信息。

　　在文本协议的memcached中，我们nc/telent后输入stats命令，会很快地输出一些当前memcached的信息的。这些就是stats信息。并不是输入stats的时候才遍历统计出来的。而是已经保存好了这份信息。代码调用在main函数中的：

stats_init();

　　具体的统计信息，可以在memcached.h这个文件中找到：

/** * Global stats. */struct stats {    pthread_mutex_t mutex;    unsigned int  curr_items;    unsigned int  total_items;    uint64_t      curr_bytes;    unsigned int  curr_conns;    unsigned int  total_conns;    uint64_t      rejected_conns;    uint64_t      malloc_fails;    unsigned int  reserved_fds;    unsigned int  conn_structs;    uint64_t      get_cmds;    uint64_t      set_cmds;    uint64_t      touch_cmds;    uint64_t      get_hits;    uint64_t      get_misses;    uint64_t      touch_hits;    uint64_t      touch_misses;    uint64_t      evictions;    uint64_t      reclaimed;    time_t        started;          /* when the process was started */    bool          accepting_conns;  /* whether we are currently accepting */    uint64_t      listen_disabled_num;    unsigned int  hash_power_level; /* Better hope it's not over 9000 */    uint64_t      hash_bytes;       /* size used for hash tables */    bool          hash_is_expanding; /* If the hash table is being expanded */    uint64_t      expired_unfetched; /* items reclaimed but never touched */    uint64_t      evicted_unfetched; /* items evicted but never touched */    bool          slab_reassign_running; /* slab reassign in progress */    uint64_t      slabs_moved;       /* times slabs were moved around */    uint64_t      lru_crawler_starts; /* Number of item crawlers kicked off */    bool          lru_crawler_running; /* crawl in progress */    uint64_t      lru_maintainer_juggles; /* number of LRU bg pokes */};

　　2.3：hash桶初始化

　　代码main函数中的：

assoc_init(settings.hashpower_init);

　　在memcached中，保存着一份hash表用来存放memcached key。默认这个hash表是2^16(65536)个key。后续会根据规则动态扩容这个hash表的。如果希望启动的时候，这个hash表更大，可以-o 参数调节。

　　hash表中， memcached key作为key，value是item指针，并不是item value。

　　2.4：初始化connection。

conn_init()

　　也就是 memcached启动参数中的-c参数，默认1024。

　　为了更快地找到connection的fd（文件描述符），实际上申请的connection会比配置的更大一点。

/* * Initializes the connections array. We don't actually allocate connection * structures until they're needed, so as to avoid wasting memory when the * maximum connection count is much higher than the actual number of * connections. * * This does end up wasting a few pointers' worth of memory for FDs that are * used for things other than connections, but that's worth it in exchange for * being able to directly index the conns array by FD. */static void conn_init(void) {    /* We're unlikely to see an FD much higher than maxconns. */    int next_fd = dup(1);    int headroom = 10;      /* account for extra unexpected open FDs */    struct rlimit rl;    max_fds = settings.maxconns + headroom + next_fd;    /* But if possible, get the actual highest FD we can possibly ever see. */    if (getrlimit(RLIMIT_NOFILE, &rl) == 0) {        max_fds = rl.rlim_max;    } else {        fprintf(stderr, "Failed to query maximum file descriptor; "                        "falling back to maxconns\n");    }    close(next_fd);    if ((conns = calloc(max_fds, sizeof(conn *))) == NULL) {        fprintf(stderr, "Failed to allocate connection structures\n");        /* This is unrecoverable so bail out early. */        exit(1);    }}

　　2.5：初始化slabs。

　　在2.3的hash桶中初始化的是key。slabs初始化的是这些key对应的value。下面摘录关键代码：

while (++i < MAX_NUMBER_OF_SLAB_CLASSES-1 && size <= settings.item_size_max / factor) {        /* Make sure items are always n-byte aligned */        if (size % CHUNK_ALIGN_BYTES)            size += CHUNK_ALIGN_BYTES - (size % CHUNK_ALIGN_BYTES);        slabclass[i].size = size;        slabclass[i].perslab = settings.item_size_max / slabclass[i].size;        size *= factor;        if (settings.verbose > 1) {            fprintf(stderr, "slab class %3d: chunk size %9u perslab %7u\n",                    i, slabclass[i].size, slabclass[i].perslab);        }    }

　　在初始化slab的时候，下一个slab的size（chunk size）总是大于等于当前slab的size的。

　　2.6：初始化worker线程。

memcached_thread_init(settings.num_threads, main_base);

　　worker线程和main线程，组成了libevent的reactor模式。

　　2.7：定时器

clock_handler(0, 0, 0);

　　用于对比对象是否过期。

第三步、libevent主线程监听事件。

/* enter the event loop */    if (event_base_loop(main_base, 0) != 0) {        retval = EXIT_FAILURE;    }

　　主线程启动堆栈：

server_sockets——>server_socket——>conn_new——>event_handler——>drive_machine——>try_read_command（这里会判定，是文本协议还是二进制协议）

第四步、关闭hash桶线程。

　　在2.3的初始化步骤中，有线程操作。这里明确关闭这个线程。

void stop_assoc_maintenance_thread() {    mutex_lock(&maintenance_lock);    do_run_maintenance_thread = 0;    pthread_cond_signal(&maintenance_cond);    mutex_unlock(&maintenance_lock);    /* Wait for the maintenance thread to stop */    pthread_join(maintenance_tid, NULL);}

　　memcached启动的主要流程就是这些了。

　　最后来个图片，描述一下启动后的memcached结构。

　　源码github上有，见：文件有点大，可能浏览器卡顿一下。这里肯定就不贴出来了^_^

转载于:https://www.cnblogs.com/ELMND/p/4581729.html

你可能感兴趣的文章

Swoole入门到实战（二）：进程，内存和协程、Swoole完美支持ThinkPHP5、分发Task异步任务机制实现...