OpenResty(Nginx Lua)获取Nginx Worker CPU使用率

在上文我们介绍了三种获取进程cpu使用率的方法，本文介绍使用openresty来获取所有nginx worker的cpu使用率，然后提供一个接口来输出cpu使用率。由于收集cpu使用率需要获取两次，两次之间需要等待一些时间，为了保证此接口的性能，决定不采用接口实时统计，采用后台定时统计，然后接口查询其数据就行。
所有步骤思路为：

1. 在init_worker阶段获取所有的worker pid
2. 在init_worker阶段启动一个定时器来统计所有nginx worker cpu使用率并存储到共享字典
3. 接口查询共享字典中的结果返回给客户端

int_worker定时统计cpu使用率

http {
    [...]
    lua_shared_dict dict 10m;
    init_worker_by_lua_block {
        -- 获取所有worker pid到字典
        local worker_pid = ngx.worker.pid()
        local worker_id = ngx.worker.id()
        ngx.shared.dict:set(worker_id,worker_pid)

        -- 统计cpu使用率函数
        local function count_cpu_usage(premature)
            -- 首次获取cpu时间
            local worker_cpu_total1 = 0
            local cpu_total1 = 0

            local worker_count = ngx.worker.count()

            for i=0, worker_count - 1 do    
                local worker_pid = ngx.shared.dict:get(i)
                local fp = io.open("/proc/"..worker_pid.."/stat","r")
                local data = fp:read("*all")
                fp:close()
                local res, err = ngx.re.match(data, "(.*? ){13}(.*?) (.*?) ", "jio")
                worker_cpu = res[2] + res[3]
                worker_cpu_total1 = worker_cpu_total1 + worker_cpu
            end

            local fp = io.open("/proc/stat","r")
            local cpu_line = fp:read()
            fp:close()
            local iterator, err = ngx.re.gmatch(cpu_line,"(\\d+)")
            while true do
                local m, err = iterator()
                if not m then
                    break
                end

                cpu_total1 = cpu_total1 + m[0]
            end

            -- 第二次获取cpu时间
            ngx.sleep(0.5)
            local worker_cpu_total2 = 0
            local cpu_total2 = 0

            for i=0, worker_count -1 do    
                local worker_pid = ngx.shared.dict:get(i)
                local fp = io.open("/proc/"..worker_pid.."/stat","r")
                local data = fp:read("*all")
                fp:close()
                local res, err = ngx.re.match(data, "(.*? ){13}(.*?) (.*?) ", "jio")
                worker_cpu = res[2] + res[3]
                worker_cpu_total2 = worker_cpu_total2 + worker_cpu
            end

            local fp = io.open("/proc/stat","r")
            local cpu_line = fp:read()
            fp:close()
            local iterator, err = ngx.re.gmatch(cpu_line,"(\\d+)")
            while true do
                local m, err = iterator()
                if not m then
                    break
                end

                cpu_total2 = cpu_total2 + m[0]
            end

            -- 获取cpu核心数
            local cpu_core = 0
            local fp = io.open("/proc/cpuinfo")
            local data = fp:read("*all")
            fp:close()
            local iterator, err = ngx.re.gmatch(data, "processor","jio")
            while true do
                local m, err = iterator()
                if not m then
                    break
                end
                cpu_core = cpu_core + 1
            end

            -- 计算出cpu时间
            local nginx_workers_cpu_time = ((worker_cpu_total2 - worker_cpu_total1) / (cpu_total2 - cpu_total1)) * 100*cpu_core
            nginx_workers_cpu_time = string.format("%d", nginx_workers_cpu_time)
            ngx.shared.dict:set("nginx_workers_cpu_time",nginx_workers_cpu_time)
        end


        -- 定时任务
        local function count_cpu_usage_timed_job()
            -- 定义间隔执行时间
            local delay = 2
            local count
            count = function(premature)
                if not premature then
                    local ok, err = pcall(count_cpu_usage, premature)
                    if not ok then
                        log(ERR, "count cpu usage error:",err)
                    end    
                    local ok, err = ngx.timer.at(delay, count)
                    if not ok then
                        return
                    end
                end
            end
            local ok, err = ngx.timer.at(delay, count)
            if not ok then
                return
            end
        end

        -- 执行定时任务
        count_cpu_usage_timed_job()
    }
    [...]
}

定义获取cpu使用率的接口

location /cpu {
    content_by_lua_block {
        local nginx_workers_cpu_time = ngx.shared.dict:get(nginx_workers_cpu_time)
        ngx.header.content_type = 'text/plain'
        ngx.say("nginx_workers_cpu_time")
    }
}