Project: yarv-prof - The Ruby Toolbox

Prerequisite

DTrace must be installed on your system
You need to edit MRI source code as follows(since pre-compiled MRI doesn't provide insn probe, you need to enable this manually, as of Ruby2.4.0)

$ git clone https://github.com/ruby/ruby.git
$ cd ruby && autoconf
$ ./configure
$ vi vm_opts.h # set the VM_COLLECT_USAGE_DETAILS flag to 1 manually
$ make && make install
$ ./ruby -v

Getting Started

Since we use custom built MRI binary, the profiling command will look like following(unless you have installed custom built MRI and rubygems onto your system)

$ gem install yarv-prof
$ sudo ./ruby -I `gem env gemdir`/gems/yarv-prof*/lib/ -r yarv-prof -e "YarvProf.start; p :hello; YarvProf.end"
$ yarv-prof --load /tmp/yarv-prof/20161128_214131.dump
total number of instruction calls: 26
insn                                   count      total_walltime      mean        variance    stdev
---------------------------------------------------------------------------------------------------
opt_send_without_block                4(15%)         298567(33%)     74642      2362231992    48603
trace                                 4(15%)         139888(15%)     34972       566184557    23795
getinlinecache                         2(7%)           76616(8%)     38308      1093622912    33070
getconstant                            2(7%)           65426(7%)     32713       674546450    25972
setinlinecache                         2(7%)           51301(5%)     25650       218718612    14789
leave                                  2(7%)           46497(5%)     23248        54444612     7379
putobject                              2(7%)           44293(4%)     22146       182576940    13512
getinstancevariable                    2(7%)           42118(4%)     21059        94146642     9703
pop                                    2(7%)           41412(4%)     20706         2056392     1434
opt_not                                1(3%)           30059(3%)     30059             NaN      NaN
jump                                   1(3%)           27644(3%)     27644             NaN      NaN
putself                                1(3%)           23844(2%)     23844             NaN      NaN
branchunless                           1(3%)           15042(1%)     15042             NaN      NaN

Usage

Step1: Recording

This is the sample usage of YarvProf in your code.

YarvProf.start(clock: :cpu, out:'~/log/')
p :hello
YarvProf.end

YarvProf#start can take following 3 optional keyword args.

clock is for the flag to switch measurement mode(:cpu or :wall)
out is to specify the target directory path which the dump file will be stored in.
opt is to give arbitrary CLI option when yarv-prof trigger dtrace command(e.g. opt:'-x bufsize=20m').

Step2: Reporting

Here is the sample usage of yarv-prof CLI command, which is specifically designed to parse and view dumped data.

$ yarvprof --load ./result.dump --insn getlocal_OP__WC__1

yarv-prof command can take following options.

$ yarvprof -h
Usage: yarv-prof [options]
    -v, --version                    Print version
        --load=FILENAME              Load .dump file
        --csv                        Report with csv format
        --raw                        Show raw log data with time series alignment
        --insn=VALUE                 Show a specific instruction only

TBD in future

More detailed information regarding YARV instructions
Improve performance
System Tap support for linux environment(only if needed)
Sampling profiler mode(if needed)
Docker support etc

Other resources

You may want to refer my presentation slide at RubyConf Taiwan 2016

License

MIT

yarv-prof

Runtime