Parallel

Factories

Parallel.of(String name, List<Action> children)
Parallel.of(String name, int parallelism, List<Action> children)
Parallel.of(String name, Action... children)
Parallel.of(String name, int parallelism, Action... children)
Parallel.of(String name, ExecutorService executorService, List<Action> children)
Parallel.of(String name, ExecutorService executorService, Action... children)

Semantics

• children are submitted concurrently
• the action waits for all children to finish
• parent status is computed from child results
• when parallelism is provided, a semaphore limits concurrent children inside this node

Parallel uses an ExecutorService.

If no executor is supplied directly to the action, it uses context.getExecutorService().

Custom executor ownership

If you pass an ExecutorService to Parallel.of(...), Paramixel uses it but does not manage its lifecycle for you.

Runner executor ownership

When no ExecutorService is supplied to Runner.builder(), each run() call creates two thread pools:

1. Runner pool — used for top-level Parallel tasks (threads named paramixel-runner-N)
2. Parallel pool — used for nested Parallel tasks using the default executor (threads named paramixel-parallel-N)

Both pools are shut down when run() completes. A Runner can be called multiple times; each call gets fresh pools.

When Runner.builder().executorService(myPool) is used:

• myPool is used for top-level action dispatch
• The parallel pool is still created and managed per run() call
• myPool is never shut down by the runner — you control its lifecycle

Pre-execution validation

DefaultRunner validates the action tree for structural problems before execution begins.

Cycle detection

CycleLoopDetector walks the action graph and rejects parent-child cycles. If a cycle is detected, it throws CycleDetectedException:

CycleDetectedException: Cycle detected in action graph: actionA[id1] -> actionB[id2] -> actionA[id1]

Deadlock detection

DeadlockDetector detects nested Parallel configurations that would cause thread starvation.

// This throws DeadlockDetected when paramixel.parallelism=1:
Action danger = Parallel.of("A",
        Parallel.of("B",
                Parallel.of("C",
                        Direct.of("leaf1", ctx -> {}),
                        Direct.of("leaf2", ctx -> {}))));

Runner.builder()
    .configuration(Map.of("paramixel.parallelism", "1"))
    .build()
    .run(danger); // throws DeadlockDetected

The fix

There are two ways to resolve this:

1. Supply dedicated executors to inner Parallel actions (preferred for fine-grained control):

ExecutorService innerEs = Executors.newFixedThreadPool(4);
Action safe = Parallel.of("A",
        Parallel.of("B", innerEs,
                Parallel.of("C",
                        Direct.of("leaf1", ctx -> {}),
                        Direct.of("leaf2", ctx -> {}))));

2. Increase paramixel.parallelism to match the nesting depth:

Runner.builder()
    .configuration(Map.of("paramixel.parallelism", "2"))
    .build()
    .run(danger); // safe with 2 threads

When detection resets

Detection resets the depth counter at any Parallel node that has a custom ExecutorService, because custom executors provide their own thread pool and do not contend for shared pool threads.

RoutingExecutorService

Internally, DefaultRunner uses a RoutingExecutorService that routes submitted tasks between two delegate executor services based on execution depth:

• Depth 0 (root level) → runner executor service
• Depth > 0 (nested parallel) → parallel executor service

This prevents thread starvation in typical nested Parallel configurations without requiring users to supply custom executors.

Interrupt handling

If the executing thread is interrupted during concurrency semaphore acquisition, the Parallel action transitions to a FAIL result with the InterruptedException as the cause, fires afterAction, and then re-interrupts the thread before re-throwing a RuntimeException wrapping the cause. This ensures the lifecycle contract (result transitions to a terminal state and afterAction is always invoked) is honored even under interrupt conditions.

Examples

Default executor

Action action = Parallel.of(
        "tests",
        Direct.of("a", context -> {}),
        Direct.of("b", context -> {}));

Node-local concurrency limit

Action action = Parallel.of(
        "tests",
        2,
        Direct.of("a", context -> {}),
        Direct.of("b", context -> {}),
        Direct.of("c", context -> {}));

Custom executor

ExecutorService executorService = Executors.newFixedThreadPool(4);
Action action = Parallel.of("tests", executorService, children);