008_falcon磁盘io计算方法
Posted itcomputer
tags:
篇首语:本文由小常识网(cha138.com)小编为大家整理,主要介绍了008_falcon磁盘io计算方法相关的知识,希望对你有一定的参考价值。
一、falcon磁盘IO告警计算方法
(1)线上告警示例 【falcon】环境: prod 时间: 2018-11-10 22:29 共1条 【#主机磁盘io过高(appid)】主机hostname磁盘dfa io过高98.76>98% (2)cat /proc/diskstats 252 0 dfa 2689642164 0 513826977162 100403006 6803204529 0 1348198360088 2893131263 0 934780608 3002409596 每项的具体含义如下: 1 252 主设备号 code:Major 2 0 次设备号 code:Minor 3 dfa 设备名称 code:Name 4 2689642164 <1>读完成次数{读磁盘的次数,成功完成读的总次数}[2689642164] code:Readios 5 0 <2>合并读完成次数 code:ReadMerges 6 513826977162 <3>读扇区的次数,成功读过的扇区总次数 5.ReadSectors 7 100403006 <4>读花费的毫秒数;这是所有读操作所花费的毫秒数{从__make_request()到end_that_request_last()的测量} code:ReadTicks 8 6803204529 <5>写完成次数;写完成的次数,成功写完成的总次数 code:WriteIOs 9 0 <6>合并写完成次数。为了效率可能会合并相邻的读和写,从而两次4K的读在它最终被处理到磁盘上之前可能会变成一次8K的读,才被计数(和排队),因此只有一次I/O操作,这个域使你知道这样的操作有多频繁 code:WriteMerges 10 1348198360088 <7>写扇区次数;写扇区的次数,成功写扇区总次数 code:WriteSectors 11 2893131263 <8>写操作花费的毫秒数;写花费的毫秒数,这是所有写操作所花费的毫秒数{是从__make_request()到end_that_request_last()的测量} code:WriteTicks 12 0 <9>正在处理的输入/输出请求数;I/O的当前进度,只有这个域应该是0.当请求被交给适当的request_queue_t时增加和请求完成时减小 code:InFlight 13 934780608 <10>输入/输出操作花费的毫秒数;花在I/O操作上的毫秒数,只要field9不为0这个域就会增长 code:IOTicks 14 3002409596 <11>输入/输出操作花费的加权毫秒数;花在I/O操作上的毫秒数,在每次I/O开始,I/O结束,I/O合并或读取自动上次更新这个域以来(第<9>列正在进行的io数量乘以花费在io上的毫秒数)时这个域都会增加.这可以给I/O完成时间和存储那些可以累积的提供一个便利的测量标准 code:TimeInQueue (3)告警计算公式 io_ticks{crt.IOTicks - last.IOTicks(上10s的值)}/100 > 98 会进行报警
二、实现代码
(1) const diskStatPath = "/proc/diskstats" type DiskStatCollector struct { lastDiskStats map[string]*linux.DiskStat } func (c *DiskStatCollector) Collect() ([]*model.Metric, error) { if c.lastDiskStats == nil { c.lastDiskStats = make(map[string]*linux.DiskStat) } disks, err := linux.ReadDiskStats(diskStatPath) if err != nil { return nil, fmt.Errorf("collect disk stat: %v", err) } var metrics []*model.Metric for _, crt := range disks { if len(crt.Name) == 3 && (strings.HasPrefix(crt.Name, "sd") || strings.HasPrefix(crt.Name, "vd") || strings.HasPrefix(crt.Name, "df")) { if last, ok := c.lastDiskStats[crt.Name]; ok { metrics = append(metrics, diskStatDiff(&crt, last)...) } tmp := crt c.lastDiskStats[crt.Name] = &tmp } } return metrics, nil } func diskStatDiff(crt *linux.DiskStat, last *linux.DiskStat) []*model.Metric { var metrics []*model.Metric if crt.IOTicks < last.IOTicks { return metrics } metrics = append(metrics, &model.Metric{ Name: "disk.io", Fields: []*model.Field{ {"read_ios", model.Gauge, crt.ReadIOs - last.ReadIOs}, {"read_merges", model.Gauge, crt.ReadMerges - last.ReadMerges}, {"read_sectors", model.Gauge, crt.ReadSectors - last.ReadSectors}, {"read_ticks", model.Gauge, crt.ReadTicks - last.ReadTicks}, {"write_ios", model.Gauge, crt.WriteIOs - last.WriteIOs}, {"write_merges", model.Gauge, crt.WriteMerges - last.WriteMerges}, {"write_sectors", model.Gauge, crt.WriteSectors - last.WriteSectors}, {"write_ticks", model.Gauge, crt.WriteTicks - last.WriteTicks}, {"in_flight", model.Gauge, crt.InFlight - last.InFlight}, {"io_ticks", model.Gauge, crt.IOTicks - last.IOTicks}, {"time_in_queue", model.Gauge, crt.TimeInQueue - last.TimeInQueue}, }, Tags: map[string]string{"name": crt.Name}, Timestamp: utils.Timestamp(), }) return metrics } (2) // ReadDiskStats reads and parses the file. // // Note: // * Assumes a well formed file and will panic if it isn‘t. func ReadDiskStats(path string) ([]DiskStat, error) { data, err := ioutil.ReadFile(path) if err != nil { return nil, err } devices := strings.Split(string(data), " ") results := make([]DiskStat, len(devices)-1) for i := range results { fields := strings.Fields(devices[i]) Major, _ := strconv.ParseInt(fields[0], 10, strconv.IntSize) results[i].Major = int(Major) Minor, _ := strconv.ParseInt(fields[1], 10, strconv.IntSize) results[i].Minor = int(Minor) results[i].Name = fields[2] results[i].ReadIOs, _ = strconv.ParseUint(fields[3], 10, 64) results[i].ReadMerges, _ = strconv.ParseUint(fields[4], 10, 64) results[i].ReadSectors, _ = strconv.ParseUint(fields[5], 10, 64) results[i].ReadTicks, _ = strconv.ParseUint(fields[6], 10, 64) results[i].WriteIOs, _ = strconv.ParseUint(fields[7], 10, 64) results[i].WriteMerges, _ = strconv.ParseUint(fields[8], 10, 64) results[i].WriteSectors, _ = strconv.ParseUint(fields[9], 10, 64) results[i].WriteTicks, _ = strconv.ParseUint(fields[10], 10, 64) results[i].InFlight, _ = strconv.ParseUint(fields[11], 10, 64) results[i].IOTicks, _ = strconv.ParseUint(fields[12], 10, 64) results[i].TimeInQueue, _ = strconv.ParseUint(fields[13], 10, 64) } return results, nil }
以上是关于008_falcon磁盘io计算方法的主要内容,如果未能解决你的问题,请参考以下文章
Java千百问_02基本使用(008)_eclipse如何关闭代码验证
Java千百问_05面向对象(008)_java中覆盖是什么
Java千百问_05面向对象(008)_java中覆盖是什么