Performance Settings (Quick Guide)

mdxplain exposes a small set of stability-focused resource controls through pipeline.config.performance. These settings are applied immediately when you change them.

Key settings

  • use_stability_config (PipelineManager init): - None (default): apply stability settings only when use_memmap=True - True: always apply stability settings - False: keep OS defaults unless you change the config manually

  • pipeline.config.performance.resource_nice: lower CPU priority to keep the system responsive

  • pipeline.config.performance.resource_io_priority: reduce I/O priority to avoid starving the OS

  • pipeline.config.performance.resource_cpu_affinity: keep a few cores free for the system

  • pipeline.config.performance.auto_blas_thread_limit: prevent BLAS/OpenMP oversubscription

Examples

Enable stability settings explicitly:

pipeline = PipelineManager(use_stability_config=True)

Change resource limits at runtime:

pipeline.config.performance.update(
    resource_nice=10,
    resource_io_priority="low",
    reserve_cores=2,
)

You can also update a single field directly:

pipeline.config.performance.reserve_cores = 1

For the full rationale and OS-specific behavior, see What we observed.