Performance Config

GitHub Link to Code.

Performance configuration for PipelineManager.

This module defines a small configuration container that triggers a callback whenever its values change. PipelineManager uses it to re-apply process-level resource limits when the user edits pipeline.config.performance at runtime.

class mdxplain.pipeline.manager.performance_config.PerformanceConfig(defaults: Dict[str, Any], on_change: Callable[[], None] | None = None)

Mutable performance settings that apply process limits on change.

The configuration stores process-level tuning knobs such as CPU priority, I/O priority, CPU affinity, and BLAS/OpenMP thread limits. It does not apply any settings by itself; instead it calls a provided callback whenever one of its fields changes. This keeps the configuration lightweight while allowing PipelineManager to centralize the actual resource management.

Fields

auto_resource_limitsbool: If True, compute a recommended CPU affinity based on reserve_cores and apply process-level limits when configuration changes.
reserve_coresint: Number of CPU cores to keep free when auto_resource_limits is enabled.
resource_niceint or None: POSIX nice value (or Windows priority mapping). None means “do not set”.
resource_io_prioritystr or None: I/O priority hint (“idle”, “low”, “normal”, “high”). None means “do not set”.
resource_cpu_affinitysequence of int or None: Explicit CPU affinity list. When set, it overrides auto selection.
auto_blas_thread_limitbool: If True, cap BLAS/OpenMP threads to the active CPU set size.

__init__(defaults: Dict[str, Any], on_change: Callable[[], None] | None = None) → None

Create a performance configuration with explicit defaults.

Parameters

defaultsdict: Mapping from field name to initial value. All fields must be provided so that pipeline.config.performance has a complete set of tunables visible to the user.
on_changecallable, optional: Callback invoked after a value changes. PipelineManager passes a function that re-applies resource limits based on this config.

Returns

None: Creates a configuration object with defaults applied.

update(**kwargs: Any) → None

Update multiple fields and apply once.

This helper avoids repeated re-application when several settings are changed together. It validates field names and triggers the callback only once after all updates are applied.

Parameters

kwargs: Field names and values to update.

Returns

None: Applies all updates and triggers a single on_change callback.