Elixir library providing some handy parallel processing facilities that supports configuring number of workers and timeout.
This library is inspired by Parex.
API documentation is available at http://hexdocs.pm/paratize
To use Paratize with your projects, edit your mix.exs
file and add it as a dependency:
defp deps do
[
{:paratize, "~> x.x.x"},
]
end
Paratize is designed to run slow tasks in parallel. There are two processor implementatons, first the chunk based implementation Paratize.Chunk
and the second the worker pool based implementation Paratize.Pool
. Both modules have the same API.
parallel_exec(fun_list, task_options)
parallel_map(arg_list, fun, task_options)
parallel_each(arg_list, fun, task_options)
To execute a list of functions in parallel,
import Paratize.Pool
function_list = [
fn -> Math.fib(40) end,
fn -> :timer.sleep(5000) end,
fn -> HTTPotion.get("http://wwww.reddit.com") end
]
parallel_exec(function_list) # => [102334155, :ok, %HTTPotion.Response{body...}]
function_keyword_list = [
fib: fn -> Math.fib(40) end,
hang: fn -> :timer.sleep(5000) end,
web_request: fn -> HTTPotion.get("http://wwww.reddit.com") end
]
parallel_exec(function_keyword_list) # => [fib: 102334155, hang: :ok, web_request: %HTTPotion.Response{body...}]
To execute a map
in parallel,
(useful when results are needed for further processing)
import Paratize.Pool
slow_func = fn arg -> :timer.sleep(1000); arg + 1 end
workload = 1..100
{time, result} = :timer.tc fn -> workload |> parallel_map(slow_func) |> Enum.join(", ") end
time # => 13034452 (8 CPU cores system, running 8 workers)
To execute a each
in parallel,
(useful when resultset is large, and can be processed individually to prevent memory hog)
import Paratize.Pool
lots_of_urls |> parallel_each(fn url ->
HTTPotion.get(url) |> parse_page |> save_meta_data
end)
Each function accepts task options to customize the parallel processing.
- size - the number of parallel workers, defaults to the number of system cores given by
:erlang.system_info(:schedulers)
- timeout - in milliseconds, the minimum time given for a function to complete, defaults to
5000
. If timeout happens, the entire parallel processing crashes withexit(:timeout,...)
. To disable timeout, use:infinity
.
To achieve maximum parallelism, %Paratize.TaskOptions{}
size should be set to size of your workload,
alias Paratize.Pool
slow_func = fn arg -> :timer.sleep(1000); arg + 1 end
workload = 1..100
{time, result} = :timer.tc fn ->
workload |> Pool.parallel_map(slow_func, size: Enum.count(workload)) |> Enum.join(", ")
end
time # => 1004370 (Running 100 workers)
The %Paratize.TaskOptions{}
timeout
should not be relied upon for precise timing out of each workload, because it is not strictly enforced. It is an implementation detail that reasonably crashes the processor if no further work is completed after the timeout period has lapsed.
This software is licensed under MIT LICENSE.