skynet服务之launcher
本文着重讨论skynet框架中，第一个服务launcher的启动流程，其他服务也是类似的；
launcher.lua代码如下：
local skynet = require "skynet"
local core = require "skynet.core"
require "skynet.manager" -- import manager apis
local string = string
local services = {}
local command = {}
local instance = {} -- for confirm (function command.LAUNCH / command.ERROR / command.LAUNCHOK)
local function handle_to_address(handle)
return tonumber("0x" .. string.sub(handle , 2))
end
local NORET = {}
function command.LIST()
local list = {}
for k,v in pairs(services) do
list[skynet.address(k)] = v
end
return list
end
function command.STAT()
local list = {}
for k,v in pairs(services) do
local ok, stat = pcall(skynet.call,k,"debug","STAT")
if not ok then
stat = string.format("ERROR (%s)",v)
end
list[skynet.address(k)] = stat
end
return list
end
function command.KILL(_, handle)
handle = handle_to_address(handle)
skynet.kill(handle)
local ret = { [skynet.address(handle)] = tostring(services[handle]) }
services[handle] = nil
return ret
end
function command.MEM()
local list = {}
for k,v in pairs(services) do
local ok, kb, bytes = pcall(skynet.call,k,"debug","MEM")
if not ok then
list[skynet.address(k)] = string.format("ERROR (%s)",v)
else
list[skynet.address(k)] = string.format("%.2f Kb (%s)",kb,v)
end
end
return list
end
function command.GC()
for k,v in pairs(services) do
skynet.send(k,"debug","GC")
end
return command.MEM()
end
function command.REMOVE(_, handle, kill)
services[handle] = nil
local response = instance[handle]
if response then
-- instance is dead
response(not kill) -- return nil to caller of newservice, when kill == false
instance[handle] = nil
end
-- don't return (skynet.ret) because the handle may exit
return NORET
end
local function launch_service(service, ...)
local param = table.concat({...}, " ")
local inst = skynet.launch(service, param)
local response = skynet.response()
if inst then
services[inst] = service .. " " .. param
instance[inst] = response
else
response(false)
return
end
return inst
end
function command.LAUNCH(_, service, ...)
launch_service(service, ...)
return NORET
end
function command.LOGLAUNCH(_, service, ...)
local inst = launch_service(service, ...)
if inst then
core.command("LOGON", skynet.address(inst))
end
return NORET
end
function command.ERROR(address)
-- see serivce-src/service_lua.c
-- init failed
local response = instance[address]
if response then
response(false)
instance[address] = nil
end
services[address] = nil
return NORET
end
function command.LAUNCHOK(address)
-- init notice
local response = instance[address]
if response then
response(true, address)
instance[address] = nil
end
return NORET
end
-- for historical reasons, launcher support text command (for C service)
skynet.register_protocol {
name = "text",
id = skynet.PTYPE_TEXT,
unpack = skynet.tostring,
dispatch = function(session, address , cmd)
if cmd == "" then
command.LAUNCHOK(address)
elseif cmd == "ERROR" then
command.ERROR(address)
else
error ("Invalid text command " .. cmd)
end
end,
}
skynet.dispatch("lua", function(session, address, cmd , ...)
cmd = string.upper(cmd)
local f = command[cmd]
if f then
local ret = f(address, ...)
if ret ~= NORET then
skynet.ret(skynet.pack(ret))
end
else
skynet.ret(skynet.pack {"Unknown command"} )
end
end)
skynet.start(function() end)
这个服务的启动比较特殊，我们看看bootstrap.lua中的启动代码：
local launcher = assert(skynet.launch("snlua","launcher"))
skynet.name(".launcher", launcher)
该服务是通过调用skynet.launch("snlua","launcher")函数来启动的，该函数的实现在文件./lualib/skynet/ manager.lua中，代码如下：
function skynet.launch(...)
local addr = c.command("LAUNCH",
table.concat({...}," "))
if addr then
return
tonumber("0x"
..
string.sub(addr ,
2))
end
end
而这里直接调用的C库函数command()函数，我们来看看这个command函数的实现：
static
int
lcommand(lua_State *L) {
struct
skynet_context * context = lua_touserdata(L, lua_upvalueindex(1));
const
char * cmd = luaL_checkstring(L,1);
const
char * result;
const
char * parm = NULL;
if (lua_gettop(L) == 2) {
parm = luaL_checkstring(L,2);
}
result = skynet_command(context, cmd, parm);
if (result) {
lua_pushstring(L, result);
return 1;
}
return 0;
}
实际上是调用了skynet_command(context, cmd, parm)函数，再来看看这个函数的实现：
static
struct
command_func
cmd_funcs[] = {
{ "TIMEOUT", cmd_timeout },
{ "REG", cmd_reg },
{ "QUERY", cmd_query },
{ "NAME", cmd_name },
{ "EXIT", cmd_exit },
{ "KILL", cmd_kill },
{ "LAUNCH", cmd_launch },
{ "GETENV", cmd_getenv },
{ "SETENV", cmd_setenv },
{ "STARTTIME", cmd_starttime },
{ "ABORT", cmd_abort },
{ "MONITOR", cmd_monitor },
{ "STAT", cmd_stat },
{ "LOGON", cmd_logon },
{ "LOGOFF", cmd_logoff },
{ "SIGNAL", cmd_signal },
{ NULL, NULL },
};
const
char *
skynet_command(struct
skynet_context * context, const
char * cmd , const
char * param)
{
struct
command_func * method = &cmd_funcs[0];
while(method->name)
{
if (strcmp(cmd, method->name) == 0)
{
return
method->func(context, param);
}
++method;
}
return
NULL;
}
该函数是通过传入的命令字符串，到事先维护好的一个二维表寻找对应的执行函数，我们传入的LAUNCH，那么对应的执行函数为cmd_launch，实现如下：
static
const
char *
cmd_launch(struct
skynet_context * context, const
char * param) {
size_t
sz = strlen(param);
char
tmp[sz+1];
strcpy(tmp,param);
char * args = tmp;
char * mod = strsep(&args, " \t\r\n");
args = strsep(&args, "\r\n");
LOG("mod=%s, args=%s", mod, args);
struct
skynet_context * inst = skynet_context_new(mod,args);
if (inst == NULL) {
return
NULL;
} else {
id_to_hex(context->result, inst->handle);
return
context->result;
}
}
代码比较简单，该函数接下来执行了skynet_context_new(mod,args)，这个函数的实现代码：
struct skynet_context *
skynet_context_new(const char * name, const char *param) {
struct skynet_module * mod = skynet_module_query(name);//snlua
if (mod == NULL)
return NULL;
void *inst = skynet_module_instance_create(mod);
if (inst == NULL)
return NULL;
struct skynet_context * ctx = skynet_malloc(sizeof(*ctx));
CHECKCALLING_INIT(ctx)
ctx->mod = mod;
ctx->instance = inst;
ctx->ref = 2;
ctx->cb = NULL;
ctx->cb_ud = NULL;
ctx->session_id = 0;
ctx->logfile = NULL;
ctx->init = false;
ctx->endless = false;
ctx->cpu_cost = 0;
ctx->cpu_start = 0;
ctx->message_count = 0;
ctx->profile = G_NODE.profile;
// Should set to 0 first to avoid skynet_handle_retireall get an uninitialized handle
ctx->handle = 0;
ctx->handle = skynet_handle_register(ctx);//将当前服务挂载到全局服务列表，并分配服务地址
struct message_queue * queue = ctx->queue = skynet_mq_create(ctx->handle);
// init function maybe use ctx->handle, so it must init at last
context_inc();
CHECKCALLING_BEGIN(ctx)
int r = skynet_module_instance_init(mod, inst, ctx, param);
CHECKCALLING_END(ctx)
if (r == 0) {
struct skynet_context * ret = skynet_context_release(ctx);
if (ret) {
ctx->init = true;
}
skynet_globalmq_push(queue);
if (ret) {
skynet_error(ret, "LAUNCH %s %s", name, param ? param : "");
}
return ret;
} else {
skynet_error(ctx, "FAILED launch %s", name);
uint32_t handle = ctx->handle;
skynet_context_release(ctx);
skynet_handle_retire(handle);
struct drop_t d = { handle };
skynet_mq_release(queue, drop_message, &d);
return NULL;
}
}
该函数首先会去加载一个模块，模块就是C动态库，并且具有一致的函数接口（xxx_create()，xxx_init()，xxx_release()），如果模块之前已经加载则不会加载直接返回模块的指针，
如果从未加载，先从模块路径加载，再返回，所有lua服务都是通过snlua模块来启动的，这里以snlua模块为例来说明：
skynet_module_instance_create(mod)函数会执行snlua_create(…)函数，该函数实现代码如下
struct
snlua *
snlua_create(void) {
struct
snlua * l = skynet_malloc(sizeof(*l));
memset(l,0,sizeof(*l));
l->mem_report = MEMORY_WARNING_REPORT;
l->mem_limit = 0;
l->L = lua_newstate(lalloc, l);
return
l;
}
该函数主要执行一些内存分配的初始化操作，对于snlua_create()，还会创建一个lua虚拟机，并使用自定义内存分配策略来为该虚拟机分配内存；
skynet_module_instance_init(…)对新context执行了初始化，这里调用的是snlua_init(…)函数，实现如下：
int
snlua_init(struct
snlua *l, struct
skynet_context *ctx, const
char * args)
{
int
sz = strlen(args);
char * tmp = skynet_malloc(sz);
memcpy(tmp, args, sz);
skynet_callback(ctx, l , launch_cb);
const
char * self = skynet_command(ctx, "REG", NULL);
uint32_t
handle_id = strtoul(self+1, NULL, 16);
// it must be first message
skynet_send(ctx, 0, handle_id, PTYPE_TAG_DONTCOPY,0, tmp, sz);
return 0;
}
这里有一个比较关键的函数skynet_callback(…)，该函数将会为新创建的context设置回调函数，函数实现如下：
void
skynet_callback(struct skynet_context * context, void *ud, skynet_cb cb) {
context->cb = cb;
context->cb_ud = ud;
}
接下来调用了skynet_command（…）函数，传递的命令为"REG"，通过上面的分析，可以得到会调用cmd_reg(…)，实现代码如下：
static const char *
cmd_reg(struct skynet_context * context, const char * param)
{
if (param == NULL || param[0] == '\0') {
sprintf(context->result, ":%x", context->handle);
return context->result;
} else if (param[0] == '.') {
return skynet_handle_namehandle(context->handle, param + 1);
} else {
skynet_error(context, "Can't register global name %s in C", param);
return NULL;
}
}
我们知道前面传递的param=NULL，所以直接执行下面的语句：
sprintf(context->result, ":%x", context->handle);
return context->result;
以上代码将handle转为16进制字符串并返回，返回之后往这个地址发送了一条消息，这里为什么需要这么做呢？
接着看如下代码片段：
CHECKCALLING_BEGIN(ctx)
int r = skynet_module_instance_init(mod, inst, ctx, param);//这里调用snlua_init(...)函数，初始化当前ctx
CHECKCALLING_END(ctx)
if (r == 0) {
struct skynet_context * ret = skynet_context_release(ctx);
if (ret) {
ctx->init = true;
}
skynet_globalmq_push(queue);
if (ret) {
skynet_error(ret, "LAUNCH %s %s", name, param ? param : "");
}
return ret;
}
初始化函数执行成功之后，调用了skynet_context_release(…)函数，代码如下：
struct skynet_context *
skynet_context_release(struct skynet_context *ctx) {
if (ATOM_DEC(&ctx->ref) == 0) {
delete_context(ctx);
return NULL;
}
return ctx;
}
注意一个细节，我们再为新的context分配内存块之后，有如下赋值：
ctx->ref = 2;
那么这里执行ATOM_DEC(&ctx->ref)后的结果为1，如果为0，就是异常，释放当前ctx内存空间；
正常情况下返回的ctx->ref=1，至此，ctx的初始化基本完成；
接下来就调用skynet_globalmq_push(queue)，将先前创建的当前ctx的消息队列丢入全局消息队列；
以上分析就创建了一个完整的launcher服务，接下来就是服务接收消息并处理消息；
上面还遗留了一个问题，在执行skynet_module_instance_init(…)函数的时候，最后给自己发了一条消息，这条消息时怎么流转的呢？
我们来仔细分析一下下面的过程：
// it must be first message
skynet_send(
ctx,
0, 发送
handle_id,
PTYPE_TAG_DONTCOPY,
0,
tmp,
sz);
源代码注释中已经说明，这是第一条消息，我们继续看看skynet_send(…)，对入参进行处理后，调用了skynet_context_push(…)
int
skynet_send(struct skynet_context * context, uint32_t source, uint32_t destination , int type, int session, void * data, size_t sz) {
if ((sz & MESSAGE_TYPE_MASK) != sz) {
skynet_error(context, "The message to %x is too large", destination);
if (type & PTYPE_TAG_DONTCOPY) {
skynet_free(data);
}
return -1;
}
_filter_args(context, type, &session, (void **)&data, &sz);
if (source == 0) {
source = context->handle;
}
if (destination == 0) {
return session;
}
if (skynet_harbor_message_isremote(destination)) {
struct remote_message * rmsg = skynet_malloc(sizeof(*rmsg));
rmsg->destination.handle = destination;
rmsg->message = data;
rmsg->sz = sz;
skynet_harbor_send(rmsg, source, session);
} else {
struct skynet_message smsg;
smsg.source = source;
smsg.session = session;
smsg.data = data;
smsg.sz = sz;
if (skynet_context_push(destination, &smsg)) {
skynet_free(data);
return -1;
}
}
return session;
}
然后继续翻skynet_context_push(…)的代码，实现代码如下：
int
skynet_context_push(uint32_t handle, struct skynet_message *message) {
struct skynet_context * ctx = skynet_handle_grab(handle);
if (ctx == NULL) {
return -1;
}
skynet_mq_push(ctx->queue, message);
skynet_context_release(ctx);
return 0;
}
先取接收消息地址对应的context，然后将消息丢到这个ctx对应的消息队列中，接着看看skynet_mq_push(…)的代码实现：
void
skynet_mq_push(struct message_queue *q, struct skynet_message *message) {
assert(message);
SPIN_LOCK(q)
//上一条消息挂载后，已经为下一条消息确定了挂载位置
q->queue[q->tail] = *message;
q->tail++;
if(q->tail > q->cap)
{
q->tail = 0;
}
if (q->head == q->tail)
{
expand_queue(q);
}
if (q->in_global == 0) {
q->in_global = MQ_IN_GLOBAL;
skynet_globalmq_push(q);
}
SPIN_UNLOCK(q)
}
以上代码很简单，先将消息丢入服务(context)的私有消息队列，然后判断，消息队列是否在全局消息队列中，不在就挂到全局消息队列；
消息何时被处理？
框架一旦启动，工作线程也会随之启动，工作线程的任务很简单，就是监控全局消息队列中挂载的各个服务的私有消息队列是否有数据，
如果发现某个私有消息队列有消息进来了，就会执行相应的处理；照旧，继续翻代码，工作线在这里启动的：
void
skynet_start(struct skynet_config * config) {
// register SIGHUP for log file reopen
struct sigaction sa;
sa.sa_handler = &handle_hup;
sa.sa_flags = SA_RESTART;
sigfillset(&sa.sa_mask);
sigaction(SIGHUP, &sa, NULL);
if (config->daemon) {
if (daemon_init(config->daemon)) {
exit(1);
}
}
skynet_harbor_init(config->harbor);
skynet_handle_init(config->harbor);
skynet_mq_init();
skynet_module_init(config->module_path);
skynet_timer_init();
skynet_socket_init();
skynet_profile_enable(config->profile);
struct skynet_context *ctx = skynet_context_new(config->logservice, config->logger);
if (ctx == NULL) {
fprintf(stderr, "Can't launch %s service\n", config->logservice);
exit(1);
}
bootstrap(ctx, config->bootstrap);
start(config->thread);
// harbor_exit may call socket send, so it should exit before socket_free
skynet_harbor_exit();
skynet_socket_free();
if (config->daemon) {
daemon_exit(config->daemon);
}
}
上面代码是有先后顺序的，我们看到启动工作线程是在lua初始化服务都启动完成之后再启动的，看看start(…)函数的实现：
static
void
start(int
thread)
{
pthread_t
pid[thread+3];
struct
monitor *m = skynet_malloc(sizeof(*m));
memset(m, 0, sizeof(*m));
m->count = thread;
m->sleep = 0;
m->m = skynet_malloc(thread * sizeof(struct
skynet_monitor *));
int
i;
for (i=0;i<thread;i++) {
m->m[i] = skynet_monitor_new();
}
if (pthread_mutex_init(&m->mutex, NULL)) {
fprintf(stderr, "Init mutex error");
exit(1);
}
if (pthread_cond_init(&m->cond, NULL)) {
fprintf(stderr, "Init cond error");
exit(1);
}
create_thread(&pid[0], thread_monitor, m);
create_thread(&pid[1], thread_timer, m);
create_thread(&pid[2], thread_socket, m);
static
int
weight[] = {
-1, -1, -1, -1, 0, 0, 0, 0,
1, 1, 1, 1, 1, 1, 1, 1,
2, 2, 2, 2, 2, 2, 2, 2,
3, 3, 3, 3, 3, 3, 3, 3, };
struct
worker_parm
wp[thread];
for (i=0;i<thread;i++) {
wp[i].m = m;
wp[i].id = i;
if (i < sizeof(weight)/sizeof(weight[0])) {
wp[i].weight= weight[i];
} else {
wp[i].weight = 0;
}
create_thread(&pid[i+3], thread_worker, &wp[i]);
}
for (i=0;i<thread+3;i++) {
pthread_join(pid[i], NULL);
}
free_monitor(