redis启动过程源码解析

Posted 2021-02-13 killianxu

大部分人在用开源软件的时候,并不知道一个优秀的开源软件内部是如何启动起来的,本文通过分析redis启动过程的源码,来学习优秀软件的启动过程

  redis整个程序的入口函数在server.c中的main函数,函数调用关系如下图1,调用顺序为从上到下,从左至右。

图1 redis启动函数调用图

  main函数源码如下,1-55行根据配置文件和启动命令参数设置全局对象server ,57-59设置redis的服务器端为后台进程, initServer主要提前创建一些经常用到的对象用于节约内存,根据设置的ip地址和端口创建监听套接字用于客户端连接,并初始化时间事件,64行用于设置server->el ->beforesleep = beforesleep,aeMain函数是个循环函数,用于监听客户端连接，接收客户端命令并进行处理等。

 1 //整个程序的入口函数
 2 int main(int argc, char **argv) {
 3     //初始化服务器配置,设置全局对象server的状态
 4     initServerConfig();
 5 
 6     /* Store the executable path and arguments in a safe place in order
 7      * to be able to restart the server later. */
 8     //存储运行命令的绝对路径及运行参数
 9     server.executable = getAbsolutePath(argv[0]);
10     server.exec_argv = zmalloc(sizeof(char*)*(argc+1));
11     server.exec_argv[argc] = NULL;
12     for (j = 0; j < argc; j++) server.exec_argv[j] = zstrdup(argv[j]);
13 
14     if (argc >= 2) {
15         j = 1; /* First option to parse in argv[] */
16         sds options = sdsempty();
17         char *configfile = NULL;
18 
19         //命令指定了配置文件,对配置文件做处理,配置文件跟在程序名后第一位
20         /* First argument is the config file name? */
21         if (argv[j][0] != \'-\' || argv[j][1] != \'-\') {
22             configfile = argv[j];
23             server.configfile = getAbsolutePath(configfile);
24             /* Replace the config file in server.exec_argv with
25              * its absolute path. */
26             zfree(server.exec_argv[j]);
27             server.exec_argv[j] = zstrdup(server.configfile);
28             j++;
29         }
30 
31         /* All the other options are parsed and conceptually appended to the
32          * configuration file. For instance --port 6380 will generate the
33          * string "port 6380\\n" to be parsed after the actual file name
34          * is parsed, if any. */
35         while(j != argc) {//解析除配置文件外的其它参数
36             if (argv[j][0] == \'-\' && argv[j][1] == \'-\') {
37                 /* Option name */
38                 if (!strcmp(argv[j], "--check-rdb")) {
39                     /* Argument has no options, need to skip for parsing. */
40                     j++;
41                     continue;
42                 }
43                 if (sdslen(options)) options = sdscat(options,"\\n");
44                 options = sdscat(options,argv[j]+2);//去掉参数前面的--
45                 options = sdscat(options," ");//参数对应的值和参数名应" "分隔
46             } else {
47                 /* Option argument */
48                 options = sdscatrepr(options,argv[j],strlen(argv[j]));
49                 options = sdscat(options," ");
50             }
51             j++;
52         }
53         //从指定配置文件和命令选项设置服务器对象server参数,覆盖默认配置
54         loadServerConfig(configfile,options);
55     }
56 
57     server.supervised = redisIsSupervised(server.supervised_mode);
58     int background = server.daemonize && !server.supervised;
59     if (background) daemonize();//后台进程模式
60 
61     //初始化服务器功能,包括时间事件1ms调用serverCron,文件事件(套接字可读可写时的处理函数)，集群初始化等
62     initServer();
63     aeSetBeforeSleepProc(server.el,beforeSleep);//设置beforeSleep事件处理函数
64     aeSetAfterSleepProc(server.el,afterSleep);//设置aftersleep事件处理函数
65     aeMain(server.el);//循环,接受客户端连接,处理命令等
66     aeDeleteEventLoop(server.el);//退出循环,删除事件处理
67     return 0;
68 }

  initServer函数源码如下,主要对server中的变量进行初始化,其中listenToPort根据监听的地址和端口, 设置server.ipfd(监听套接字数组)和server.ipfd_count(监听套接字数目),85行aeCreateTimeEvent设置定时器事件,每1ms执行serverCron函数,94行aeCreateFileEvent函数设置监听套接字有客户端连接时执行的事件处理函数acceptTcpHandler。

  1 void initServer(void) {
  2     int j;
  3     server.hz = server.config_hz;
  4     server.pid = getpid();
  5     server.current_client = NULL;
  6     server.clients = listCreate();
  7     server.clients_index = raxNew();
  8     server.clients_to_close = listCreate();
  9     server.slaves = listCreate();
 10     server.monitors = listCreate();
 11     server.clients_pending_write = listCreate();
 12     server.slaveseldb = -1; /* Force to emit the first SELECT command. */
 13     server.unblocked_clients = listCreate();
 14     server.ready_keys = listCreate();
 15     server.clients_waiting_acks = listCreate();
 16     server.get_ack_from_slaves = 0;
 17     server.clients_paused = 0;
 18     server.system_memory_size = zmalloc_get_memory_size();
 19     //创建常用字符串,比如命令回复pong及命令字符串等
 20     createSharedObjects();
 21     //根据配置的最大连接客户端数和CONFIG_MIN_RESERVED_FDS(用于监听、日志文件等文件描述符)设置最大允许打开的文件描述符数
 22     adjustOpenFilesLimit();
 23     //初始化aeEventLoop对象并赋值给server中的指针变量el
 24     server.el = aeCreateEventLoop(server.maxclients+CONFIG_FDSET_INCR);
 25     server.db = zmalloc(sizeof(redisDb)*server.dbnum);
 26 
 27     /* Open the TCP listening socket for the user commands. */
 28     //用于创建监听文件描述符
 29     if (server.port != 0 &&
 30         listenToPort(server.port,server.ipfd,&server.ipfd_count) == C_ERR)
 31         exit(1);
 32 
 33     /* Create the Redis databases, and initialize other internal state. */
 34     for (j = 0; j < server.dbnum; j++) {
 35         server.db[j].dict = dictCreate(&dbDictType,NULL);
 36         server.db[j].expires = dictCreate(&keyptrDictType,NULL);
 37         server.db[j].blocking_keys = dictCreate(&keylistDictType,NULL);
 38         server.db[j].ready_keys = dictCreate(&objectKeyPointerValueDictType,NULL);
 39         server.db[j].watched_keys = dictCreate(&keylistDictType,NULL);
 40         server.db[j].id = j;
 41         server.db[j].avg_ttl = 0;
 42         server.db[j].defrag_later = listCreate();
 43     }
 44     //创建lru池,内存超过设置的值时,根据淘汰策略,清除部分key
 45     evictionPoolAlloc(); /* Initialize the LRU keys pool. */
 46     server.pubsub_channels = dictCreate(&keylistDictType,NULL);
 47     server.pubsub_patterns = listCreate();
 48     listSetFreeMethod(server.pubsub_patterns,freePubsubPattern);
 49     listSetMatchMethod(server.pubsub_patterns,listMatchPubsubPattern);
 50     server.cronloops = 0;
 51     server.rdb_child_pid = -1;
 52     server.aof_child_pid = -1;
 53     server.rdb_child_type = RDB_CHILD_TYPE_NONE;
 54     server.rdb_bgsave_scheduled = 0;
 55     server.child_info_pipe[0] = -1;
 56     server.child_info_pipe[1] = -1;
 57     server.child_info_data.magic = 0;
 58     aofRewriteBufferReset();
 59     server.aof_buf = sdsempty();
 60     server.lastsave = time(NULL); /* At startup we consider the DB saved. */
 61     server.lastbgsave_try = 0;    /* At startup we never tried to BGSAVE. */
 62     server.rdb_save_time_last = -1;
 63     server.rdb_save_time_start = -1;
 64     server.dirty = 0;
 65     resetServerStats();
 66     /* A few stats we don\'t want to reset: server startup time, and peak mem. */
 67     server.stat_starttime = time(NULL);
 68     server.stat_peak_memory = 0;
 69     server.stat_rdb_cow_bytes = 0;
 70     server.stat_aof_cow_bytes = 0;
 71     server.cron_malloc_stats.zmalloc_used = 0;
 72     server.cron_malloc_stats.process_rss = 0;
 73     server.cron_malloc_stats.allocator_allocated = 0;
 74     server.cron_malloc_stats.allocator_active = 0;
 75     server.cron_malloc_stats.allocator_resident = 0;
 76     server.lastbgsave_status = C_OK;
 77     server.aof_last_write_status = C_OK;
 78     server.aof_last_write_errno = 0;
 79     server.repl_good_slaves_count = 0;
 80 
 81     /* Create the timer callback, this is our way to process many background
 82      * operations incrementally, like clients timeout, eviction of unaccessed
 83      * expired keys and so forth. */
 84     //创建时间事件,1ms调用serverCron
 85     if (aeCreateTimeEvent(server.el, 1, serverCron, NULL, NULL) == AE_ERR) {
 86         serverPanic("Can\'t create event loop timers.");
 87         exit(1);
 88     }
 89 
 90     /* Create an event handler for accepting new connections in TCP and Unix
 91      * domain sockets. */
 92     //设定套接字可读可写状态的处理函数
 93     for (j = 0; j < server.ipfd_count; j++) {
 94         if (aeCreateFileEvent(server.el, server.ipfd[j], AE_READABLE,
 95             acceptTcpHandler,NULL) == AE_ERR)
 96             {
 97                 serverPanic(
 98                     "Unrecoverable error creating server.ipfd file event.");
 99             }
100     }
101 }

  listenToPort函数会对所有配置需要监听的ip地址和端口调用_anetTcpServer函数(anet.c), _anetTcpServer函数源码如下,1-17行建立socket套接字s,23行调用函数anetListen完成服务端socket的bind和listen。

 1 static int _anetTcpServer(char *err, int port, char *bindaddr, int af, int backlog)
 2 {
 3     int s = -1, rv;
 4     char _port[6];  /* strlen("65535") */
 5     struct addrinfo hints, *servinfo, *p;
 6     snprintf(_port,6,"%d",port);
 7     memset(&hints,0,sizeof(hints));
 8     hints.ai_family = af;
 9     hints.ai_socktype = SOCK_STREAM;
10     hints.ai_flags = AI_PASSIVE;    /* No effect if bindaddr != NULL */
11 
12     if ((rv = getaddrinfo(bindaddr,_port,&hints,&servinfo)) != 0) {
13         anetSetError(err, "%s", gai_strerror(rv));
14         return ANET_ERR;
15     }
16     for (p = servinfo; p != NULL; p = p->ai_next) {
17         if ((s = socket(p->ai_family,p->ai_socktype,p->ai_protocol)) == -1)
18             continue;
19 
20         if (af == AF_INET6 && anetV6Only(err,s) == ANET_ERR) goto error;
21         if (anetSetReuseAddr(err,s) == ANET_ERR) goto error;
22         //anetListen调用listen和bind
23         if (anetListen(err,s,p->ai_addr,p->ai_addrlen,backlog) == ANET_ERR) s = ANET_ERR;
24         goto end;
25     }
26     if (p == NULL) {
27         anetSetError(err, "unable to bind socket, errno: %d", errno);
28         goto error;
29     }
30 
31 error:
32     if (s != -1) close(s);
33     s = ANET_ERR;
34 end:
35     freeaddrinfo(servinfo);
36     return s;
37 }

  anetListen函数源码如下:

 1 static int anetListen(char *err, int s, struct sockaddr *sa, socklen_t len, int backlog) {
 2     if (bind(s,sa,len) == -1) {
 3         anetSetError(err, "bind: %s", strerror(errno));
 4         close(s);
 5         return ANET_ERR;
 6     }
 7 
 8     if (listen(s, backlog) == -1) {
 9         anetSetError(err, "listen: %s", strerror(errno));
10         close(s);
11         return ANET_ERR;
12     }
13     return ANET_OK;
14 }

  在分析aeCreateTimeEvent和aeCreateFileEvent函数之前,先给出server中的aeEventLoop类型变量el结构体定义(ae.h):

/* File event structure */
//文件事件结构体
typedef struct aeFileEvent {
    ////读或者写，也用于标识该事件结构体是否正在使用
    int mask; /* one of AE_(READABLE|WRITABLE|BARRIER) */
    //读事件的处理函数
    aeFileProc *rfileProc;
    //写事件的处理函数
    aeFileProc *wfileProc;
    //传递给上述两个函数的数据
    void *clientData;
} aeFileEvent;

/* Time event structure */
//时间事件
typedef struct aeTimeEvent {
    //时间事件标识符，用于唯一标识该时间事件，并且用于删除时间事件
    long long id; /* time event identifier. */
    long when_sec; /* seconds */
    long when_ms; /* milliseconds */
    //事件对应的处理程序
    aeTimeProc *timeProc;
    //时间事件的最后一次处理程序，若已设置，则删除时间事件时会被调用
    aeEventFinalizerProc *finalizerProc;
    void *clientData;
    struct aeTimeEvent *prev;
    struct aeTimeEvent *next;
} aeTimeEvent;

/* A fired event */
//用于保存已触发的事件
typedef struct aeFiredEvent {
    int fd;
    int mask;
} aeFiredEvent;

/* State of an event based program */
typedef struct aeEventLoop {
    //最大文件描述符
    int maxfd;   /* highest file descriptor currently registered */
    //文件描述符的最大监听数
    int setsize; /* max number of file descriptors tracked */
    //生成时间事件的唯一标识
    long long timeEventNextId;
    //检测系统时钟偏差
    time_t lastTime;     /* Used to detect system clock skew */
    //双向链表,注册文件事件
    aeFileEvent *events; /* Registered events */
    //已触发的文件事件
    aeFiredEvent *fired; /* Fired events */
    //注册的时间事件
    aeTimeEvent *timeEventHead;
    int stop;
    //处理底层特定API的数据，对于epoll来说，该结构体包含了epoll fd和epoll_event
    void *apidata; /* This is used for polling API specific data */
    aeBeforeSleepProc *beforesleep;
    aeBeforeSleepProc *aftersleep;
} aeEventLoop;

  函数aeCreateTimeEvent(ae.c)用于往server中的变量el时间事件队列添加元素,源码如下:

 1 long long aeCreateTimeEvent(aeEventLoop *eventLoop, long long milliseconds,
 2         aeTimeProc *proc, void *clientData,
 3         aeEventFinalizerProc *finalizerProc)
 4 {
 5     long long id = eventLoop->timeEventNextId++;
 6     aeTimeEvent *te;
 7 
 8     te = zmalloc(sizeof(*te));
 9     if (te == NULL) return AE_ERR;
10     te->id = id;
11     aeAddMillisecondsToNow(milliseconds,&te->when_sec,&te->when_ms);
12     te->timeProc = proc;
13     te->finalizerProc = finalizerProc;
14     te->clientData = clientData;
15     te->prev = NULL;
16     te->next = eventLoop->timeEventHead;
17     if (te->next)
18         te->next->prev = te;
19     eventLoop->timeEventHead = te;
20     return id;
21 }

  aeCreateFileEvent(ae.c)函数用于server.el增加文件事件队列添加元素,源码如下:

static int aeApiAddEvent(aeEventLoop *eventLoop, int fd, int mask) {
    aeApiState *state = eventLoop->apidata;

    if (mask & AE_READABLE) FD_SET(fd,&state->rfds);
    if (mask & AE_WRITABLE) FD_SET(fd,&state->wfds);
    return 0;
}

int aeCreateFileEvent(aeEventLoop *eventLoop, int fd, int mask,
        aeFileProc *proc, void *clientData)
{
    if (fd >= eventLoop->setsize) {
        errno = ERANGE;
        return AE_ERR;
    }
    aeFileEvent *fe = &eventLoop->events[fd];

    if (aeApiAddEvent(eventLoop, fd, mask) == -1)
        return AE_ERR;
    fe->mask |= mask;
    if (mask & AE_READABLE) fe->rfileProc = proc;
    if (mask & AE_WRITABLE) fe->wfileProc = proc;
    fe->clientData = clientData;
    if (fd > eventLoop->maxfd)
        eventLoop->maxfd = fd;
    return AE_OK;
}

  执行完initServer及aeSetBeforeSleepProc和aeSetAfterSleepProc函数设置好全局变量server后,进程调用aeMain函数进入循环,开始接收客户端连接和命令交互。aeMain(ae.c)函数源码如下:

 1 //处理客户端命令的主函数，由main函数调用,aeEventLoop定义在(ae.h),beforesleep和aftersleep均在服务器启动初始化时设定
 2 void aeMain(aeEventLoop *eventLoop) {
 3     eventLoop->stop = 0;
 4     while (!eventLoop->stop) {
 5         if (eventLoop->beforesleep != NULL)
 6             eventLoop->beforesleep(eventLoop);
 7         //处理文件时间及时间事件
 8         aeProcessEvents(eventLoop, AE_ALL_EVENTS|AE_CALL_AFTER_SLEEP);
 9     }
10 }