Linux operation and maintenance troubleshooting ideas, this article is enough~-Linux Operation and Maintenance-php.cn

#1. Background

Sometimes you encounter some difficult and complicated diseases, and the monitoring plug-in cannot be used immediately at a glance Discover the root of the problem. At this time, you need to log in to the server to further analyze the root cause of the problem. Then analyzing problems requires a certain amount of technical experience accumulation, and some problems involve very wide areas in order to locate the problem. Therefore, analyzing problems and stepping into pitfalls is a great exercise for one's growth and self-improvement. If we have a good set of analysis tools, it will be twice the result with half the effort, helping everyone quickly locate problems and saving everyone a lot of time to do more in-depth work.

Linux operation and maintenance troubleshooting ideas, this article is enough~

2. Description

This article The article mainly introduces various problem positioning tools and analyzes problems based on cases.

3. Methodology for analyzing problems

Applying the 5W2H method, several questions about performance analysis can be raised

What-What is the phenomenon like
When-When does it happen
Why-Why did it happen
Where-Where did the problem happen
How much-how many resources were consumed
How to do-how to solve the problem

4. cpu

4.1 Description

For applications, we usually focus on the kernel CPU scheduler function and performance.

Thread status analysis mainly analyzes where thread time is used, and the classification of thread status is generally divided into:

on- CPU: During execution, the time during execution is usually divided into user mode time user and system mode time sys.
off-CPU: Waiting for the next round of CPU, or waiting for I/O, lock, page change, etc., its status can be subdivided into executable , anonymous paging, sleep, lock, idle and other states.

If a lot of time is spent on the CPU, profiling the CPU can quickly explain the cause; if the system spends a lot of time in the off-cpu state, locating the problem will take a lot of time. But there are still some concepts that need to be clear:

Processor
Core
Hardware thread
CPU memory cache
Clock frequency
CPI and instructions per cycle IPC
CPU instructions
Usage rate
User time/kernel time
Scheduler
Run Queue
Preemption
Multiple processes
Multiple threads
Word length

##4.2 Analysis tools

Linux operation and maintenance troubleshooting ideas, this article is enough~

Note:

##uptime, vmstat, mpstat, top, pidstat can only query the usage of cpu and load.
perf can follow the time-consuming status of specific functions within the process, and can specify kernel functions for statistics, and specify which ones to hit.

4.3 Usage

//查看系统cpu使用情况top
//查看所有cpu核信息mpstat -P ALL 1
//查看cpu使用情况以及平均负载vmstat 1
//进程cpu的统计信息pidstat -u 1 -p pid
//跟踪进程内部函数级cpu使用情况 perf top -p pid -e cpu-clock

Copy after login

5. Memory

5.1 Note

Memory is designed to improve efficiency. When actually analyzing the problem, memory problems may not only affect performance, but also affect services or cause other problems.同样对于内存有些概念需要清楚：

牛逼啊！接私活必备的 N 个开源项目！赶快收藏

Copy after login

主存
虚拟内存
常驻内存
地址空间
OOM
页缓存
缺页
换页
交换空间
交换
用户分配器libc、glibc、libmalloc和mtmalloc
LINUX内核级SLUB分配器

5.2 分析工具

Linux operation and maintenance troubleshooting ideas, this article is enough~

说明：

free，vmstat，top，pidstat，pmap只能统计内存信息以及进程的内存使用情况。
valgrind 可以分析内存泄漏问题。
dtrace 动态跟踪。需要对内核函数有很深入的了解，通过D语言编写脚本完成跟踪。

5.3 使用方式

//查看系统内存使用情况free -m//虚拟内存统计信息vmstat 1//查看系统内存情况top//1s采集周期，获取内存的统计信息pidstat -p pid -r 1//查看进程的内存映像信息pmap -d pid//检测程序内存问题valgrind --tool=memcheck --leak-check=full --log-file=./log.txt  ./程序名

Copy after login

6. 磁盘IO

6.1 说明

磁盘通常是计算机最慢的子系统，也是最容易出现性能瓶颈的地方，因为磁盘离 CPU 距离最远而且 CPU 访问磁盘要涉及到机械操作，比如转轴、寻轨等。访问硬盘和访问内存之间的速度差别是以数量级来计算的，就像1天和1分钟的差别一样。要监测 IO 性能，有必要了解一下基本原理和 Linux 是如何处理硬盘和内存之间的 IO 的。

在理解磁盘IO之前，同样我们需要理解一些概念，例如：

File System
VFS
# #File system cache
page cachepage cache
buffer cache buffer cache
Directory cache
##inode
inode cache
noop calling strategy

6.2 Analysis tool

6.3 使用方式

//查看系统io信息iotop//统计io详细信息iostat -d -x -k 1 10//查看进程级io的信息pidstat -d 1 -p  pid//查看系统IO的请求，比如可以在发现系统IO异常时，可以使用该命令进行调查，就能指定到底是什么原因导致的IO异常perf record -e block:block_rq_issue -ag^Cperf report

Copy after login

7. 网络

7.1 说明

网络的监测是所有 Linux 子系统里面最复杂的，有太多的因素在里面，比如：延迟、阻塞、冲突、丢包等，更糟的是与 Linux 主机相连的路由器、交换机、无线信号都会影响到整体网络并且很难判断是因为 Linux 网络子系统的问题还是别的设备的问题，增加了监测和判断的复杂度。现在我们使用的所有网卡都称为自适应网卡，意思是说能根据网络上的不同网络设备导致的不同网络速度和工作模式进行自动调整。

7.2 分析工具

Linux operation and maintenance troubleshooting ideas, this article is enough~

7.3 使用方式

//显示网络统计信息netstat -s//显示当前UDP连接状况netstat -nu//显示UDP端口号的使用情况netstat -apu//统计机器中网络连接各个状态个数netstat -a | awk '/^tcp/ {++S[$NF]} END {for(a in S) print a, S[a]}'//显示TCP连接ss -t -a//显示sockets摘要信息ss -s//显示所有udp socketsss -u -a//tcp,etcp状态sar -n TCP,ETCP 1//查看网络IOsar -n DEV 1//抓包以包为单位进行输出tcpdump -i eth1 host 192.168.1.1 and port 80 //抓包以流为单位显示数据内容tcpflow -cp host 192.168.1.1

Copy after login

8. 系统负载

8.1 说明

Load 就是对计算机干活多少的度量（WikiPedia：the system Load is a measure of the amount of work that a compute system is doing）简单的说是进程队列的长度。Load Average 就是一段时间（1分钟、5分钟、15分钟）内平均Load。

8.2 分析工具

Linux operation and maintenance troubleshooting ideas, this article is enough~

8.3 使用方式

//查看负载情况uptimetopvmstat//统计系统调用耗时情况strace -c -p pid//跟踪指定的系统操作例如epoll_waitstrace -T -e epoll_wait -p pid//查看内核日志信息dmesg

Copy after login

9. 火焰图

9.1 说明

火焰图（Flame Graph是 Bredan Gregg 创建的一种性能分析图表，因为它的样子近似 ?而得名。

火焰图主要是用来展示 CPU的调用栈。

y 轴表示调用栈，每一层都是一个函数。调用栈越深，火焰就越高，顶部就是正在执行的函数，下方都是它的父函数。

x 轴表示抽样数，如果一个函数在 x 轴占据的宽度越宽，就表示它被抽到的次数多，即执行的时间长。注意，x 轴不代表时间，而是所有的调用栈合并后，按字母顺序排列的。

火焰图就是看顶层的哪个函数占据的宽度最大。只要有”平顶”（plateaus），就表示该函数可能存在性能问题。颜色没有特殊含义，因为火焰图表示的是 CPU 的繁忙程度，所以一般选择暖色调。

常见的火焰图类型有 On-CPU、Off-CPU、Memory、Hot/Cold、Differential等等。

9.2 安装依赖库

//安装systemtap，默认系统已安装yum install systemtap systemtap-runtime//内核调试库必须跟内核版本对应，例如：uname -r 2.6.18-308.el5kernel-debuginfo-2.6.18-308.el5.x86_64.rpmkernel-devel-2.6.18-308.el5.x86_64.rpmkernel-debuginfo-common-2.6.18-308.el5.x86_64.rpm//安装内核调试库debuginfo-install --enablerepo=debuginfo search kerneldebuginfo-install --enablerepo=debuginfo  search glibc

Copy after login

9.3 安装

git clone https://github.com/lidaohang/quick_location.gitcd quick_location

Copy after login

9.4 CPU级别火焰图

cpu占用过高，或者使用率提不上来，你能快速定位到代码的哪块有问题吗？

一般的做法可能就是通过日志等方式去确定问题。现在我们有了火焰图，能够非常清晰的发现哪个函数占用cpu过高，或者过低导致的问题。另外，搜索公众号Linux就该这样学后台回复“猴子”，获取一份惊喜礼包。

9.4.1 on-CPU

cpu占用过高，执行中的时间通常又分为用户态时间user和系统态时间sys。

使用方式：

//on-CPU usersh ngx_on_cpu_u.sh pid//进入结果目录 cd ngx_on_cpu_u//on-CPU kernelsh ngx_on_cpu_k.sh pid//进入结果目录 cd ngx_on_cpu_k//开一个临时端口 8088 python -m SimpleHTTPServer 8088//打开浏览器输入地址127.0.0.1:8088/pid.svg

Copy after login

DEMO：

Copy after login

#include #include 
void foo3(){  }
void foo2(){    int i;    for(i=0 ; i < 10; i++)           foo3();}
void foo1(){    int i;  for(i = 0; i< 1000; i++)      foo3();}
int main(void){    int i;    for( i =0; i< 1000000000; i++) {          foo1();          foo2();    }}

Copy after login

DEMO火焰图：

Linux operation and maintenance troubleshooting ideas, this article is enough~

9.4.2 off-CPU

cpu过低，利用率不高。等待下一轮CPU，或者等待I/O、锁、换页等等，其状态可以细分为可执行、匿名换页、睡眠、锁、空闲等状态。

使用方式：

// off-CPU usersh ngx_off_cpu_u.sh pid//进入结果目录cd ngx_off_cpu_u//off-CPU kernelsh ngx_off_cpu_k.sh pid//进入结果目录cd ngx_off_cpu_k//开一个临时端口8088python -m SimpleHTTPServer 8088//打开浏览器输入地址127.0.0.1:8088/pid.svg

Copy after login

官网DEMO：

Linux operation and maintenance troubleshooting ideas, this article is enough~

9.5 内存级别火焰图

如果线上程序出现了内存泄漏，并且只在特定的场景才会出现。这个时候我们怎么办呢？有什么好的方式和工具能快速的发现代码的问题呢？同样内存级别火焰图帮你快速分析问题的根源。

使用方式：

sh ngx_on_memory.sh pid//进入结果目录cd ngx_on_memory//开一个临时端口8088python -m SimpleHTTPServer 8088//打开浏览器输入地址127.0.0.1:8088/pid.svg

Copy after login

官网DEMO：

Linux operation and maintenance troubleshooting ideas, this article is enough~

9.6 性能回退-红蓝差分火焰图

你能快速定位CPU性能回退的问题么？如果你的工作环境非常复杂且变化快速，那么使用现有的工具是来定位这类问题是很具有挑战性的。当你花掉数周时间把根因找到时，代码已经又变更了好几轮，新的性能问题又冒了出来。主要可以用到每次构建中，每次上线做对比看，如果损失严重可以立马解决修复。

通过抓取了两张普通的火焰图，然后进行对比，并对差异部分进行标色：红色表示上升，蓝色表示下降。差分火焰图是以当前（“修改后”）的profile文件作为基准，形状和大小都保持不变。因此你通过色彩的差异就能够很直观的找到差异部分，且可以看出为什么会有这样的差异。

使用方式：

cd quick_location//抓取代码修改前的profile 1文件perf record -F 99 -p pid -g -- sleep 30perf script > out.stacks1//抓取代码修改后的profile 2文件perf record -F 99 -p pid -g -- sleep 30perf script > out.stacks2//生成差分火焰图:./FlameGraph/stackcollapse-perf.pl ../out.stacks1 > out.folded1./FlameGraph/stackcollapse-perf.pl ../out.stacks2 > out.folded2./FlameGraph/difffolded.pl out.folded1 out.folded2 | ./FlameGraph/flamegraph.pl > diff2.svg

Copy after login

DEMO：

//test.c#include #include 
void foo3(){  }
void foo2(){    int i;    for(i=0 ; i < 10; i++)      foo3();}
void foo1(){    int i;    for(i = 0; i< 1000; i++)       foo3();}
int main(void){    int i;  for( i =0; i< 1000000000; i++) {      foo1();      foo2();    }}
//test1.c#include #include 
void foo3(){
}
void foo2(){  int i;  for(i=0 ; i < 10; i++)         foo3();}
void foo1(){    int i;    for(i = 0; i< 1000; i++)         foo3();}
void add(){    int i;    for(i = 0; i< 10000; i++)       foo3();}
int main(void){    int i;    for( i =0; i< 1000000000; i++) {    foo1();    foo2();    add();  }}

Copy after login

DEMO红蓝差分火焰图：

Linux operation and maintenance troubleshooting ideas, this article is enough~

10. Case Analysis

10.1 Access layer nginx cluster anomaly

Through the monitoring plug-in, it was found that at 19 o'clock on September 25, 2017, a large number of 499, 5xx appeared in the nginx cluster request traffic status code. And it was found that the machine's CPU usage has increased, and it continues. In addition, search the top algorithm background of the official account and reply "algorithm" to get a surprise gift package.

10.2 Analyze nginx related indicators

a) **Analyze nginx request traffic:

Linux operation and maintenance troubleshooting ideas, this article is enough~

Conclusion:

From the above figure, we find that the traffic did not increase suddenly, but decreased, which has nothing to do with the sudden increase in request traffic.

b) **Analysis of nginx response time
Linux operation and maintenance troubleshooting ideas, this article is enough~

Conclusion:

Through the above figure, we find that the increase in nginx response time may be related to nginx itself or to the backend upstream response time.

c) **Analysis of nginx upstream response time

Linux operation and maintenance troubleshooting ideas, this article is enough~

##Conclusion:

It is found from the above figure that the nginx upstream response time has increased. It is currently speculated that the backend upstream response time may be holding back nginx, causing nginx to experience abnormal request traffic.

10.3 Analyze system CPU situation

a) **Observe system indicators through top

top

Linux operation and maintenance troubleshooting ideas, this article is enough~

Conclusion:

It is found that nginx worker cpu is relatively high

b) **Analyze the internal cpu situation of nginx process

perf top -p pid

结论：

发现主要开销在free,malloc,json解析上面

10.4 火焰图分析cpu
a) **生成用户态cpu火焰图

//on-CPU usersh ngx_on_cpu_u.sh pid//进入结果目录cd ngx_on_cpu_u//开一个临时端口8088python -m SimpleHTTPServer 8088//打开浏览器输入地址127.0.0.1:8088/pid.svg

Copy after login

Linux operation and maintenance troubleshooting ideas, this article is enough~

结论：

发现代码里面有频繁的解析json操作，并且发现这个json库性能不高，占用cpu挺高。

10.5 案例总结

a) 分析请求流量异常，得出nginx upstream后端机器响应时间拉长

b) Analyzing the high cpu of the nginx process, it is concluded that the nginx internal module code has time-consuming json parsing and memory allocation and recycling operations

10.5 .1 In-depth analysis

Based on the conclusions of the above two points of analysis, we will further analyze in depth.

The backend upstream response is lengthened, which may affect nginx's processing capabilities at most. But it is unlikely to affect nginx internal modules taking up too much CPU operations. And the modules that occupied a lot of CPU at that time were only executed when requested. It is unlikely that the upstram backend is holding nginx back, thereby triggering the time-consuming operation of this CPU.

10.5.2 Solution

When encountering this kind of problem, we give priority to solving known and very clear problems. That's the problem with high CPU. The solution is to downgrade and close the module that consumes too much CPU, and then observe. After downgrading and shutting down the module, the CPU dropped, and nginx request traffic became normal. The reason why upstream time is lengthened is because the interface called by the upstream backend service may be a loop and go back to nginx again.

11.参考资料

http://www.brendangregg.com/index.html
http://www.brendangregg.com/FlameGraphs/cpuflamegraphs.html
http://www.brendangregg.com/FlameGraphs/memoryflamegraphs.html
http://www.brendangregg.com/FlameGraphs/offcpuflamegraphs.html
http://www.brendangregg.com/blog/2014-11-09/differential-flame-graphs.html
https://github.com/openresty/openresty-systemtap-toolkit
https://github.com/brendangregg/FlameGraph
https://www.slideshare.net/brendangregg/blazing-performance-with-flame-graphs

The above is the detailed content of Linux operation and maintenance troubleshooting ideas, this article is enough~. For more information, please follow other related articles on the PHP Chinese website!