linux flamegraph火焰图使用

perf可以忽略语言或者一些开发框架，从os的角度看到进程的cpu时间都耗费在哪些调用栈上。
perf与火焰图搭配使用，提供一个比较友好的交互方式，来观察调用栈的耗时
验证环境是ubuntu环境

安装

安装perf

$ sudo apt install linux-tools-common
# 装完使用perf的时候，提示还需要安装其它包
$ sudo apt install linux-tools-4.15.0-91-generic linux-cloud-tools-4.15.0-91-generic

装完之后就可以使用perf命令了。
常用的命令：是 perf record命令，help一下，可以发现

$ perf record -h

 Usage: perf record [<options>] [<command>]
    or: perf record [<options>] -- <command> [<options>]

    -a, --all-cpus        system-wide collection from all CPUs
    -b, --branch-any      sample any taken branches
    -B, --no-buildid      do not collect buildids in perf.data
    -c, --count <n>       event period to sample
    -C, --cpu <cpu>       list of cpus to monitor
    -d, --data            Record the sample addresses
    -D, --delay <n>       ms to wait before starting measurement after program s
    -e, --event <event>   event selector. use 'perf list' to list available even
    -F, --freq <n>        profile at this frequency
    -g                    enables call-graph recording
    -G, --cgroup <name>   monitor event in cgroup name only
    -I, --intr-regs[=<any register>]
                          sample selected machine registers on interrupt, use -I
    -i, --no-inherit      child tasks do not inherit counters
    -j, --branch-filter <branch filter mask>
                          branch stack filter modes
    -k, --clockid <clockid>
                          clockid to use for events, see clock_gettime()
    -m, --mmap-pages <pages[,pages]>
                          number of mmap data pages and AUX area tracing mmap pa
    -N, --no-buildid-cache
                          do not update the buildid cache
    -n, --no-samples      don't sample
    -o, --output <file>   output file name
    -P, --period          Record the sample period
    -p, --pid <pid>       record events on existing process id
    -q, --quiet           don't print any message
    -R, --raw-samples     collect raw sample records from all opened counters
    -r, --realtime <n>    collect data with this RT SCHED_FIFO priority
    -S, --snapshot[=<opts>]
                          AUX area tracing Snapshot Mode
    -s, --stat            per thread counts
    -t, --tid <tid>       record events on existing thread id
    -T, --timestamp       Record the sample timestamps
    -u, --uid <user>      user to profile
    -v, --verbose         be more verbose (show counter open errors, etc)
    -W, --weight          sample by weight (on special events only)
        --all-kernel      Configure all used events to run in kernel space.
        --all-user        Configure all used events to run in user space.
        --buildid-all     Record build-id of all DSOs regardless of hits
        --call-graph <record_mode[,record_size]>
                          setup and enables call-graph (stack chain/backtrace):

                                record_mode:    call graph recording mode (fp|dw
                                record_size:    if record_mode is 'dwarf', max s
                                                default: 8192 (bytes)

                                Default: fp
        --clang-opt <clang options>
                          options passed to clang when compiling BPF scriptlets
        --clang-path <clang path>
                          clang binary to use for compiling BPF scriptlets
        --dry-run         Parse options then exit
        --exclude-perf    don't record events from perf itself
        --filter <filter>
                          event filter
        --group           put the counters into a counter group
        --namespaces      Record namespaces events
        --no-buffering    collect data without buffering
        --overwrite       use overwrite mode
        --per-thread      use per-thread mmaps
        --phys-data       Record the sample physical addresses
        --proc-map-timeout <n>
                          per thread proc mmap processing timeout in ms
        --running-time    Record running/enabled time of read (:S) events
        --sample-cpu      Record the sample cpu
        --switch-events   Record context switch events
        --switch-output[=<signal,size,time>]
                          Switch output when receive SIGUSR2 or cross size,time 
        --tail-synthesize
                          synthesize non-sample events at the end of output
        --timestamp-filename
                          append timestamp to output filename
        --transaction     sample transaction flags (special events only)
        --user-regs[=<any register>]
                          sample selected machine registers on interrupt, use -I
        --vmlinux <file>  vmlinux pathname

安装flamegraph

火焰图是脚本，只需要下载

$ git clone https://github.com/brendangregg/FlameGraph.git

使用

perf的教程主要参考 http://www.brendangregg.com/perf.html

火焰图的教程主要参考 http://www.brendangregg.com/flamegraphs.html

# -F是每秒的采样次数 -p 是进程id， -a 是采集所有cpu， -g是记录调用依赖关系，sleep 
$ sudo perf record -F 99 -p 85356  -a -g -- sleep 10
# 上一步会在当前目录生成 perf.data文件，需要转化为profile未见
$  sudo perf script > out.perf
# 生成一个中间文件 out.folded
$ ./FlameGraph/stackcollapse-perf.pl out.perf > out.folded
# out.svg就是我们要的火焰图文件，可以用浏览器打开
$ ./FlameGraph/flamegraph.pl out.folded > out.svg

用浏览器打开

火焰图怎么看？
教程是这样说的

The x-axis shows the stack profile population, sorted alphabetically (it is not the passage of time), and the y-axis shows stack depth, counting from zero at the bottom. Each rectangle represents a stack frame. The wider a frame is is, the more often it was present in the stacks. The top edge shows what is on-CPU, and beneath it is its ancestry. The colors are usually not significant, picked randomly to differentiate frames.

• x轴是按照字母排列的profile调用栈信息
• y轴代表栈深度。每个矩形代表一个栈帧，矩形越宽，说明调用越频繁。
  上下矩形具有祖先关系，下面矩形是上面矩形的祖先，调用关系的表现在这。
• 颜色是随机的，没有意义。

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。