skuld_process

Parallel > | Umbrella →

Multi-process execution for Skuld — true BEAM-process parallelism and process-level shared mutable state. Use when you need real parallelism across CPU cores, or when forked tasks need to share state without copying env.

What's included

• Skuld.Effects.Parallel — fork-join concurrency using Task.Supervisor. Parallel.all runs computations in separate BEAM processes and collects results; Parallel.race returns the first to complete and cancels the rest; Parallel.map applies a function over items concurrently. A with_sequential_handler runs everything synchronously for deterministic testing.
• Skuld.Effects.AtomicState — thread-safe mutable state for concurrent contexts. Unlike regular State (stored in env.state and copied when forking), AtomicState uses an external Agent (production) or env.state (testing via Sync handler). Supports multiple independent tagged states and compare-and-swap operations.

When to use Parallel vs FiberPool

FiberPool (from skuld_concurrency) runs fibers cooperatively within a single BEAM process — ideal for I/O-bound work. Parallel spawns real BEAM processes — use for CPU-bound work or when you need OS-level scheduling. Both compose: FiberPool-managed fibers can fan out with Parallel when they hit CPU-heavy sections.

Installation

def deps do
  [
    {:skuld_process, "~> 0.32"}
  ]
end

Example: parallel fan-out with shared state

use Skuld.Syntax
alias Skuld.Effects.{Parallel, AtomicState, Throw}

comp do
  # Initialize shared counter
  _ <- AtomicState.put(:processed, 0)

  # Process items in parallel
  _ <- Parallel.map(1..10, fn i ->
    comp do
      result <- do_expensive_work(i)
      _ <- AtomicState.modify(:processed, &(&1 + 1))
      result
    end
  end)

  count <- AtomicState.get(:processed)
  count
end
|> AtomicState.Agent.with_handler(0, tag: :processed)
|> Parallel.with_handler()
|> Throw.with_handler()
|> Comp.run!()
#=> 10

Further reading

• Parallel — operations, error handling, test handler
• AtomicState — tagged states, CAS, Agent vs Sync

See the architecture guide for how this fits into the Skuld ecosystem.

Parallel > | Umbrella →