Parallel Execution
FZ runs cases concurrently across the calculators you provide. Each calculator entry is locked to one case at a time, so N calculator entries = N parallel workers.
Turning On Parallelism
# Sequential — one calculator entry
fz.fzr("input.txt", {"t": [100, 200, 300, 400, 500]}, model,
calculators="sh://bash calc.sh")
# Parallel — repeat the entry
fz.fzr("input.txt", {"t": [100, 200, 300, 400, 500]}, model,
calculators=["sh://bash calc.sh"] * 3) # 3 cases at once
The entries do not have to be identical — mix local, SSH, and SLURM freely:
calculators = [
"sh://bash calc.sh",
"sh://bash calc.sh",
"ssh://user@node1/bash /path/calc.sh",
"ssh://user@node2/bash /path/calc.sh",
]
Load Balancing
Cases are handed out round-robin. With 10 cases and 3 calculators:
| Calculator | Cases |
|---|---|
| 0 | 0, 3, 6, 9 |
| 1 | 1, 4, 7 |
| 2 | 2, 5, 8 |
Controlling the Worker Count
| Method | Example |
|---|---|
| Number of calculator entries | ["sh://bash calc.sh"] * 8 |
| Environment variable | export FZ_MAX_WORKERS=8 |
| Config object | from fz import get_config; get_config().max_workers = 8 |
FZ_MAX_WORKERS caps the thread pool regardless of how many calculator entries you pass.
Choosing a Number
- CPU-bound: about
os.cpu_count()workers. - I/O-bound / remote: more than the core count can help.
- Memory-bound:
available_RAM / RAM_per_case.
Progress and Interrupts
A progress bar with ETA is shown on stderr (auto-disabled when stderr is not a terminal, e.g. in CI or when redirected). Press Ctrl+C for a graceful shutdown: running cases finish, no new cases start, partial results are saved. See Interrupt Handling.
Combine With Cache
calculators = [
"cache://previous_run", # instant on a hit
*["sh://bash calc.sh"] * 4, # otherwise 4 parallel workers
]