Persistent project calculations
Calculations are untracked by default. A persistent project stores run
records for tracking progress and loading results later. This is
independent of the execution backend. Reconnecting to a known running
job is a separate capability: direct execution supports limited output
reattachment without guaranteeing the lost exit code, whereas the
kkrunner supervisor retains completion information. See
First calculation: aluminium SCF and DOS for ordinary calculations and Remote running, MPI and execution profiles
for the differences between persistence, reattachment and recovery.
Create and track a run
Create a project once and pass its ID to the calculation:
from ase.build import bulk
from ase2sprkkr import SPRKKR
from ase2sprkkr.sprkkr.projects import default_project_repository
project = default_project_repository().create("Aluminium")
calculator = SPRKKR(atoms=bulk("Al", "fcc", a=4.05), directory="al-scf")
run = calculator.create_run(project=project.id)
print("Project ID:", project.id)
print("Run ID:", run.id)
run.submit()
result = run.wait()
create_run() prepares the calculation, submit() starts execution,
and wait() returns the SPR-KKR result. The project ID identifies
the project; the run ID identifies the calculation. The output files
must remain available for loading and postprocessing.
For a synchronous calculation with project tracking, use
calculator.calculate(project=project.id) instead of
create_run(), submit() and wait().
On the command line, select an existing project by UUID and optionally an execution profile:
ase2sprkkr run example.pot --task SCF --project-id PROJECT_UUID --profile cluster
The command prints the run ID after submission and normally waits for the result. To submit a project calculation without waiting, use:
ase2sprkkr run example.pot --task SCF --project-id PROJECT_UUID --profile cluster --submit-only
--submit-only requires --project-id and a managed kkrunner or
psij backend. A tracked direct run must remain under CLI observation
until completion. The working directory defaults to the project’s directory
when configured, or to the current directory. Use --directory PATH to
override it. See Command-line reference for all options.
Reconnect or load history
After restarting the client, reopen the project manager to resume observing a known nonterminal run, where backend capabilities permit:
from ase2sprkkr.sprkkr.projects import project_execution_manager
from ase2sprkkr.outputs.project_results import attach_project_run
manager = project_execution_manager(project="PROJECT_UUID")
run = attach_project_run(manager, "RUN_ID")
result = run.wait()
For a completed historical run, load the parsed result rather than reattaching the process:
from ase2sprkkr.outputs.task_result import TaskResult
result = TaskResult.from_run("RUN_ID", manager=manager)
Replace the placeholder IDs with those printed when the project and run
were created. History requires access to the same project run store and
retained output files. Failed or interrupted postprocessing may be
retried without resubmitting SPR-KKR using
retry_project_postprocessing(manager, run_id).
Inspect and follow a run
ase2sprkkr remote status RUN_ID
ase2sprkkr remote follow RUN_ID --info --from-start
ase2sprkkr remote cancel RUN_ID
These commands use the saved project run ID. See Command-line reference for other options, or Remote running, MPI and execution profiles for backend requirements.