Skip to contents

Deterministic execution timeout policies

execution_timeout.Rd

Create explicit elapsed-time budgets and execute analysis plans or registries with timeout handling, dependency propagation, retry integration, and auditable timeout metadata.

Usage

new_execution_timeout_policy(
  default_seconds = Inf,
  module_seconds = numeric(),
  label = "no-timeout"
)

# S3 method for class 'PopgenVCFExecutionTimeoutPolicy'
print(x, ...)

execute_analysis_plan_with_timeouts(
  analysis,
  context,
  registry,
  plan,
  engine = new_execution_engine(fail_fast = FALSE),
  timeout_policy = new_execution_timeout_policy(),
  retry_policy = new_execution_retry_policy()
)

execute_analysis_registry_with_timeouts(
  analysis,
  context,
  registry,
  selected = NULL,
  engine = new_execution_engine(fail_fast = FALSE),
  timeout_policy = new_execution_timeout_policy(),
  retry_policy = new_execution_retry_policy()
)

Arguments

default_seconds

Default elapsed-time budget in seconds. Use Inf for no timeout.

module_seconds

Optional uniquely named numeric vector of positive per-module budgets.

label

Stable policy identifier recorded in execution metadata.

x

A PopgenVCFExecutionTimeoutPolicy object.

...

Unused.

analysis

A PopgenVCFAnalysis object.

context

Runtime context shared by module runners.

registry

A PopgenVCFRegistry object.

plan

A PopgenVCFExecutionPlan object.

engine

Execution engine.

timeout_policy

A PopgenVCFExecutionTimeoutPolicy object.

retry_policy

A PopgenVCFExecutionRetryPolicy object.

selected

Optional module names.

Value

new_execution_timeout_policy() returns a PopgenVCFExecutionTimeoutPolicy. The execution functions return the normal execution result with timeout-aware final and attempt ledgers and timeout policy metadata. The print method returns x invisibly.

Details

Timeouts are disabled by default. A timed-out module contributes no accepted result or artifact, is recorded as timed_out, and blocks dependent modules unless a bounded retry later completes successfully and passes the normal validation contracts.

The budget is enforced with setTimeLimit, which R checks only when control returns to the interpreter. It stops a module that is running R-level code, but it cannot interrupt a single blocking call: a long-running compiled routine (most of SNPRelate, adegenet and phangorn) or a wait on an external process runs to completion however small the budget, and is reported as timed out only afterwards, if at all (measured directly: one compiled call ran 422 seconds under a 1-second budget). A budget is a guard against runaway R-level work, not a hard wall-clock guarantee. The external ancestry backends have their own, genuinely enforced limit (timeout_seconds under analyses.admixture and analyses.faststructure), which kills the child process; use a scheduler or container limit where a hard bound on the whole run is required.