l3nz

l3nz

A pmap implementation: any feedback?

Hello,
I have been playing with a pmap (parallel map) implementation, for a task that seems easy: to run a series of functions in parallel and gather their results. The “snag” is that I want to handle errors and timeouts out of a given deadline, all of them returning a results and not aborting the general computation or - worse - killing the parent.

I ended up with something like:

def parallel_map_ordered(enum, myFn, errorFn, timeout) do
    myWrappedFn = fn v ->
      try do
        myFn.(v)
      rescue
        err ->
          with Logger.error(
                 "Crashed pmap: #{Tools.ii(err)} on #{Tools.ii(__STACKTRACE__)} for input #{Tools.ii(v)}"
               ) do
            errorFn.(v, err)
          end
      end
    end

    tasks = Enum.map(enum, &Task.async(fn -> myWrappedFn.(&1) end))

    Task.yield_many(tasks, timeout)
    |> Enum.zip(enum)
    |> Enum.map(fn {{%Task{}, res}, orgval} ->
      case res do
        {:ok, v} ->
          v

        nil ->
          with Logger.error("Timed out pmap: for input #{Tools.ii(orgval)}") do
            errorFn.(orgval, :timeout)
          end
      end
    end)
  end

That you use by passing two functions:

  • one is the main function; it gets called for any element of the Enum and is supposed to return an element
  • one is the error handler, that gets called with the value that was supposed to be processed and the error or the kw :timeout, and returns a value to be inserted into the resulting collection

There is a maximum timeout, after which all computations are aborted.

You use it like this:

 test "simplex" do
      fnOk = fn v ->
        Process.sleep(v * 100)
        v * 100
      end

      fnErr = fn v, e ->
        {:error, v, e}
      end

      assert [100, 200, 300, 100, 200, 100, 200, 500, 100] =
               PsTools.parallel_map_ordered(
                 [1, 2, 3, 1, 2, 1, 2, 5, 1],
                 fnOk,
                 fnErr,
                 5000
               )
    end

This will work; if you set one of the values to (say) 500, the function will time out and terminate in 5 seconds, but all intermediate values will be preserved.

The implementaion is naive, meaning that there is no batching and everything will be run in parallel, so if you have a file reader and run one million in parallel, you will exhaust file descriptors. If you batch, every batch will be limited by the slowest operation, so again it is kind of meh. But it’s a start.

Now for my questions:

  • is there something that I overlooked in the handling? should i catch :exit in the rescue clause?
  • was there a simple way to do that using the Tasks module?

TIA

Marked As Solved

dimitarvp

dimitarvp

Basically you can get away with just this:

defmodule Parallel do
  defp map_executor(func, item) when is_function(func, 1) do
    try do
      # No need to put `:ok` in values that ran successfully
      # because `Task.async_stream` already does it.
      func.(item)
    rescue
      err ->
        # TODO: Also do logging or APM reporting here.
        # Wrap the exception in an `:error` typle.
        {:error, err}
    end
  end

  defp ensure_ok_or_error_tuple({:exit, :timeout}), do: {:error, :timeout}
  defp ensure_ok_or_error_tuple({:ok, _} = ok), do: ok
  defp ensure_ok_or_error_tuple({:error, _} = err), do: err

  def map(list, func, timeout)
    when is_list(list) and is_function(func, 1)
    and is_integer(timeout) and timeout > 0 do
    Task.async_stream(
      list,
      fn(item) -> map_executor(func, item) end,
      timeout: timeout, ordered: true, on_timeout: :kill_task
    )
    |> Enum.map(&ensure_ok_or_error_tuple/1)
  end
end

So more or less: have an executor function where you can neatly take care of exceptions or any special cases before handing off to ensure_ok_or_error_tuple/1 (NOTE: I have not added a catch-all clause there and this is intentional; IMO you’d want the code to crash if you receive an unexpected result so you can add an extra clause to take care of it and not silently ignore it or reshape it into an error tuple that’s not informative; basically: don’t mask a potential bug).

Also depending on your scenario you might want these tasks supervised but the docs cover these cases well so I will not repeat them.

Also Liked

rhcarvalho

rhcarvalho

I wrote a small helper today which I eventually named async_map, a constrained version of Task.Supervisor.async_stream, if you will.

  # Helper function to run supervised tasks concurrently mapping results to the
  # desired output.
  #
  # - `async_fun` runs in a task process, takes a single enumerable item as
  # argument and must return {:ok, result} or {:error, reason}.
  # - `post_fun` runs in the calling process, takes a tuple of the result of
  # async_fun and the corresponding enumerable item as argument, and returns the
  # final result.
  defp async_map(enumerable, async_fun, post_fun, opts) do
    Task.Supervisor.async_stream(
      MyApp.TaskSupervisor,
      enumerable,
      async_fun,
      Keyword.put(opts, :on_timeout, :kill_task)
    )
    |> Enum.map(fn
      {:ok, result} -> result
      {:exit, :timeout} -> {:error, :timeout}
    end)
    |> Enum.zip(enumerable)
    |> Enum.map(post_fun)
  end

It converts timeouts into errors so that I have a single place to handle errors in post_func. It also gives post_func the original input, be it for logging or generating a fallback value.

I know about the :zip_input_on_exit option to Task.Supervisor.async_stream, but the advantage of doing it like this is that I have access to the input in every case of sucess/error/timeout.

Example usage:

1..10
|> async_map(
  fn n ->
    cond do
      n == 5 -> {:error, :bad_input}
      n == 7 -> {:ok, Process.sleep(10_000)}
      true -> {:ok, n*n}
    end
  end,
  fn
    {{:ok, sq}, n} -> "#{n}*#{n} = #{sq}"
    {{:error, reason}, n} -> "Error on #{n}: #{inspect(reason)}"
  end,
  max_concurrency: 5,
  timeout: 500
)

# returns:
# ["1*1 = 1", "2*2 = 4", "3*3 = 9", "4*4 = 16", "Error on 5: :bad_input",
#  "6*6 = 36", "Error on 7: :timeout", "8*8 = 64", "9*9 = 81", "10*10 = 100"]
dimitarvp

dimitarvp

Happens to the best of us, sometimes you just need a second pair of eyes for the final touches. And you seem to have done almost all of this already, so good job!

Where Next?

Popular in Questions Top

JDanielMartinez
Hi! May someone helps me, please! I have two apps into an umbrella project: the first one is Database, which manages queries, and the se...
New
joaquinalcerro
Hi there, I am working with Ecto-Postgresql and I need to call all of the records from a specific table but the table has 40,000 record...
New
aalberti333
As the title describes, I’m trying to run Enum.map() over a list of key/value pairs, where the value is a map. My data looks like this: ...
New
lastday4you
I wanted to check elixir version in phoenix because i found that my elixir is 1.5 but when i use Enum.chunk_by it said the function is un...
New
stefanchrobot
What’s the safe way to decode a JSON string into a struct? I want to avoid calling String.to_atom. Jason.decode can give me a map with st...
New
freewebwithme
Using vs code and installed ElixirLS: support and debugger. And I got an error popped up on start up says Failed to run ‘elixir’ comma...
New
shahryarjb
Hello, I have map which I want to convert it to string like this: the map: %{last_name: "tavakkoli", name: "shahryar"} the string I ne...
New
johnnyicon
Hi all, I've just started learning Elixir and Phoenix Framework, so please pardon my n00bness at this stage. I'm trying to use Postg...
New
kostonstyle
Hi all I want to have a unix time, from the current time plus 1 hour. DateTime.now + 1 hour How to get it in elixir? Thanks
New
vrod
I am using the Starship cross-shell prompt – it seems pretty nice, but I get some errors: [WARN] - (starship::utils): Executing command ...
New

Other popular topics Top

srinivasu
How to handle excepions in elixir? Suppose i have A, B, C ,D, E modules. and each module has get() function. A.get() method will call th...
New
dotdotdotPaul
Okay, I'm having a heck of a time trying to figure out how to best handle the validation of belongs_to associations in Ecto. I'm sure I'...
New
bsollish-terakeet
Credo is smart enough to check for (something like) this: assert length(the_list) == 0 with this response: Checking if an enum is empt...
New
yawaramin
In the Dialyzer docs ( http://erlang.org/doc/man/dialyzer.html#requesting-or-suppressing-warnings-in-source-files ), there is a way to tu...
New
aalberti333
As the title describes, I’m trying to run Enum.map() over a list of key/value pairs, where the value is a map. My data looks like this: ...
New
vac
Hi, I'm quite new in Elixir and I'm trying to format a string to a PEM format. I have the certificate value like MIIDBTCCAe2...... and ...
New
minhajuddin
I have seen a lot of code which picks the first element from a list using Enum.at(0) instead of List.first. Is there a reason why people ...
New
ashish173
I am using Ecto timestamps with postgres, I can see the timestamps() use the :naive_dateime but for my use case I wanted to store the ti...
New
electic
Hi, I am new to Elixir. I am trying to use the DateTime component to insert a date into MySQL however the there seems to be no way to fo...
New
Fl4m3Ph03n1x
About me? ( if you have nothing better to do than reading about some random guy in the internet :stuck_out_tongue: ) Hello all, this is ...
New

We're in Beta

About us Mission Statement