Running Jobs

Basic Execution

from ngpb4py import NgpbConfig, NgpbRunner

config = NgpbConfig.from_prm("options.prm")
result = NgpbRunner(nproc=8).run(config=config)

Working Directories

workdir is treated as a scratch parent directory, not the final run directory. Each call to run() creates a unique child directory inside it.

When workdir is omitted, ngpb4py now uses /tmp/ngpb4py by default so short runs avoid the overhead of staging files in the current working tree.

result = NgpbRunner().run(config=config, workdir="scratch", keep_files=True)
print(result.scratch_dir)
print(result.workdir)

This makes concurrent runs safer because they do not overwrite each other's files.

Persistent Container Reuse

NgpbRunner now reuses a warm Apptainer instance across multiple run() calls. Use close() or a context manager to release it explicitly:

with NgpbRunner(nproc=8) as runner:
    result1 = runner.run(config=config1)
    result2 = runner.run(config=config2)

Verbosity

Wrapper logging can be controlled globally on the runner or per run:

runner = NgpbRunner(verbosity=1)
result = runner.run(config=config, verbose=3)

Verbosity levels:

  • 0: warnings and errors only
  • 1: high-level progress messages
  • 2: debug logging from the wrapper
  • 3: debug logging plus streamed backend output

The default runner verbosity is 1, which avoids streamed backend output unless you opt in.

Container Options

You can customize the runtime invocation:

runner = NgpbRunner(
    container_image="/data/images/NextGenPB.sif",
    container_exec_args=["--containall"],
    container_extra_args=["--silent"],
)

Use container_exec_args for flags passed to apptainer exec, and container_extra_args for flags inserted immediately after the runtime command.