Elixir Wrocław № 3

Elixir Wrocław № 3

lucjan@surferseo.com

...or how I co-founded Elixir-first startup

💖

Profiling Elixir

Benchmarking vs Profiling

250ms

20%

40%

20%

8%

BENCHMARK

PROFILE

Benchmarking vs Profiling

250ms

20%

40%

20%

8%

BENCHMARK

PROFILE

How long does it take?

What is the slowest part?

Benchmarking vs Profiling

Code optimization loop

IDENTIFY

SLOW CODE

Code optimization loop

IDENTIFY

SLOW CODE

BENCHMARK

Code optimization loop

IDENTIFY

SLOW CODE

BENCHMARK

PROFILE

Code optimization loop

IDENTIFY

SLOW CODE

BENCHMARK

IMPROVE

PROFILE

Code optimization loop

IDENTIFY

SLOW CODE

BENCHMARK

PROFILE

IMPROVE

Code optimization loop

["Elixir", "is", "faster", "than", "Ruby"]

[["Elixir", "is"],

["is", "faster"],

["faster", "than"],

["than", "Ruby"]]

def bigrams(list) do
  list 
    |> Enum.with_index()
    |> Enum.reduce([], 
    fn {item, index}, acc ->
      [Enum.slice(list, index, 2) | 
      acc]
    end) |> Enum.filter(fn grams -> 
      Enum.count(grams) == 2 
    end)
end
def bigrams(list)
  list
    .each_with_index
    .reduce([]) 
    {|acc, (item, index)|
      [list.slice(index, 2)] + 
      acc
    }.select {|grams| 
      grams.length == 2
    }
end

?? ms

?? ms

list = 1..100000 |> Enum.into([])
before_time = :os.system_time(:milli_seconds)
Bigrams.bigrams(list)
IO.inspect(
  :os.system_time(:milli_seconds) - before_time
)
list = (1..10000).to_a
now = (Time.now.to_f * 1000).to_i
bigrams(list)
puts(
  Time.now.to_f * 1000).to_i - now
)

130 ms

620 ms

* for list of 10 000 integers

lower is better

130 ms

620 ms

Elixir supposed to be faster than Ruby!!!1one

Let's find out what the hell is going on

Benchee.run(
  %{
    "initial" => fn -> Bigrams.get(list) end
  }
)

https://github.com/bencheeorg/benchee


Name              ips        average  deviation         median         99th %
initial        2.82 K      354.35 μs    ±16.26%      339.00 μs      583.00 μs

Honorable mention: benchfella

Benchee

https://github.com/alco/benchfella

http://erlang.org/doc/efficiency_guide/profiling.html

Erlang profiling trio

cprof

eprof

fprof

http://erlang.org/doc/efficiency_guide/profiling.html

Erlang profiling trio

speed

details

cprof

eprof

fprof

http://erlang.org/doc/efficiency_guide/profiling.html

Erlang profiling trio

speed

details

speed

details

cprof

eprof

fprof

http://erlang.org/doc/efficiency_guide/profiling.html

Erlang profiling trio

cprof

eprof

fprof

speed

details

speed

details

speed

details

https://hexdocs.pm/mix/Mix.Tasks.Profile.Eprof.html

Elixir profiling trio

mix profile.cprof

mix profile.eprof

mix profile.fprof

Showtime

Running profiler directly from the code

defmodule FProf do
  defmacro profile(do: block) do
    content = quote do
      Mix.Tasks.Profile.Fprof.profile(fn -> unquote(block) end,
        warmup: false, callers: true)
    end
    Code.compile_quoted(content)
  end
end

profile do
  very_slow_function()
end

https://selfamusementpark.com/profiling-a-slow-elixir-test

Visualisations

https://github.com/isacssouza/erlgrind

https://github.com/jrfonseca/gprof2dot

direct erlang fprof

erlgrind


:fprof.apply(Bigrams, :get_array, [list])
:fprof.profile()
:fprof.analyse({:dest, 'out.fprof'})

$ ./erlgrind out.fprof out.cgrind

Visualisations

https://github.com/isacssouza/erlgrind

https://github.com/jrfonseca/gprof2dot

callgrind

KCacheGrind

gprof2dot

webgraphviz.com

Visualisations

eflame

{:eflame, "~> 1.0"}

mix run -e ":eflame.apply(Bigrams, :get, [])" && \
./deps/eflame/stack_to_flame.sh < stacks.out > flame.svg

Shameless plug

WE ARE HIRING!

Thank you!

Made with Slides.com