`skynet.call`内部消息延迟定位求助
#1724
-
|
目前使用 目前出现的问题是： 也就是说, 在这个时间点, 所在机器配置为: 32 核 64G 当前时间节点前后一秒 当前时间节点前后一秒所在机器的 目前分析可能性有三点
目前为止更倾向于由 |
Beta Was this translation helpful? Give feedback.
Answered by
gjy1606
May 11, 2023
Replies: 6 comments 19 replies
-
|
C 层有个功能可以 log 单个服务的每一条消息。通过 debug console 的 logon 指令打开。 它能记录每条消息的具体内容和时间戳。 https://github.com/cloudwu/skynet/blob/master/skynet-src/skynet_log.c |
Beta Was this translation helpful? Give feedback.
8 replies
-
|
注入一段代码添加一个 debug 指令，让它运行暂停或恢复 gc 的指令。 |
Beta Was this translation helpful? Give feedback.
-
|
尴尬了, 我是在 -- skynet.lua
function skynet.timeout(ti, func)
local session = c.intcommand("TIMEOUT",ti)
assert(session)
local co = co_create_for_timeout(func, ti)
assert(session_id_coroutine[session] == nil)
session_id_coroutine[session] = co
return co -- for debug
end
function skynet.timer(ti, func)
local f
f = function()
if func(ti) then
skynet.timeout(ti, f)
end
end
skynet.timeout(ti, f)
end
-- service.lua
skynet.start(function()
skynet.timer(100*60*5, function ( ... )
collectgar.memCollectGarbage()
return true
end)
-- skynet.dispatch(xxxxx)
end) |
Beta Was this translation helpful? Give feedback.
-
|
collectgar.memCollectGarbage 把这个函数替换掉就可以了？ |
Beta Was this translation helpful? Give feedback.
-
|
使用 inject 注入代码添加 console 命令可以参考这个示例，在开发环境先测试注入的代码是否正确。 |
Beta Was this translation helpful? Give feedback.
-
|
好, 感谢, 我试试 |
Beta Was this translation helpful? Give feedback.
-
|
会不会是这个队列过长导致的 function skynet.wakeup(token)
if sleep_session[token] then
tinsert(wakeup_queue, token)
return true
end
end |
Beta Was this translation helpful? Give feedback.
4 replies
-
|
我算了一下,平均每秒 50 条消息的样子,不至于这里太长, |
Beta Was this translation helpful? Give feedback.
-
|
额外GC太频繁？ |
Beta Was this translation helpful? Give feedback.
-
这里堆积，应该在mqlen里面看不出来的 |
Beta Was this translation helpful? Give feedback.
-
5 分钟额外主动调用一次, 还好吧 |
Beta Was this translation helpful? Give feedback.
-
|
center_data高耗时调用出现的频率是怎样？是每次调用都这么高耗时 还是 偶尔一段时间出现？偶尔一段时间出现才比较可能是gc引起的。 |
Beta Was this translation helpful? Give feedback.
3 replies
-
|
我加下监控, 现在看是这样 |
Beta Was this translation helpful? Give feedback.
-
|
都还没过G，那跟gc关系不大。 |
Beta Was this translation helpful? Give feedback.
-
|
@lmess 看来不是 gc 的问题了 |
Beta Was this translation helpful? Give feedback.
-
|
昨晚把 logon 打开的日志整理好了, 把时间转化了一下, trace 的各个状态数据对上了, 关键信息如下(详细的信息放在最后): 可以看出, function center_data.CheckGateLogin(gateId, args)
local accountId = args.accountId
local gate
for _, gi in pairs(gateInfo) do
if gi.id == gateId then
gate = gi
break
end
end
if not gate then
return {result = Opr.LoginNotGate}
end
local acInfo = gate.accountList and gate.accountList[accountId]
if not acInfo then
return {result = Opr.LoginAccTimeout}
end
if acInfo.gameToken == args.gameToken then
return {result = Opr.Ok, gateInfo = gate, acInfo = acInfo}
else
return {result = Opr.LoginTokenErr}
end
end转化时间以后的 |
Beta Was this translation helpful? Give feedback.
3 replies
-
|
accountList 完全可以放在缓存系统中存储 |
Beta Was this translation helpful? Give feedback.
-
|
|
Beta Was this translation helpful? Give feedback.
-
|
accountList 中的 key 是整型还是字符串？ |
Beta Was this translation helpful? Give feedback.
-
|
开了大部分的 |
Beta Was this translation helpful? Give feedback.
0 replies
Answer selected by
gjy1606
-
|
最后是如何解决的呢？可以考虑做差异同步 https://blog.codingnow.com/2017/05/syncobj.html |
Beta Was this translation helpful? Give feedback.
1 reply
-
|
抱歉，没说清楚，这是逻辑BUG，这个 |
Beta Was this translation helpful? Give feedback.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
开了大部分的
trace，最终把问题定位在skynet.pack上了，然后找到了根源，每次返回调用地址的时候，都返回了accountList，缓存了所有登录过的账号信息，最大可能有几十万，所以pack的耗时不断的增长，只是很缓慢，review 代码的时候也一直忽略了过去。