sage.utils.servers

Filename : servers.py Description : Single source of truth for per-server settings.

This module centralises everything that changes when Sage is run on a different machine: where data is saved, where the conda env lives, and how batch jobs are launched. A run is wired to a server with one switch – the SAGE_SERVER environment variable (or hostname auto-detection) – and every path, both in Python (config.py / dataset.py) and in the shell run scripts (sage/utils/run_base.sh), is derived from the same entry here.

Add a new machine by adding one Server to SERVERS below.

Shell side

The shell helpers read these same values via:

eval "$(python -m sage.utils.servers env)"

so there is no duplicated path list to keep in sync.

CLI

python -m sage.utils.servers name # resolved server name python -m sage.utils.servers show # human-readable dump python -m sage.utils.servers env # export KEY=VAL lines for the shell

Attributes

`REPO_ROOT`
`SERVERS`

Classes

Server

Everything that is specific to one machine.

Functions

get_server([name])

Resolve the active server.

Module Contents

REPO_ROOT = ''[source]

class Server[source]

Everything that is specific to one machine.

Paths

home : the user’s home directory on that machine. data_root : parent dir under which data_release/ is created. This is

the big-storage mount where downloaded noise lands and from which training reads. Mirrors save_parent_dir of sage.data.primer.DataReleaseDownloader.

work_root : scratch / cache root (model compile caches, torch hub, etc.). repo_root : location of the sage checkout. Defaults to this checkout.

Launch

pythonabsolute path to the interpreter (conda env) for jobs. If: None the shell falls back to whatever python is on PATH.

scheduler : “slurm” or “local”. partition / qos / gres / time / cpus / mem / account / mail :

SLURM sbatch parameters (ignored when scheduler=”local”).

name: str[source]

hostmatch: tuple[str, Ellipsis][source]

home: str[source]

data_root: str[source]

work_root: str[source]

repo_root: str = ''[source]

python: str | None = None[source]

scheduler: str = 'local'[source]

partition: str = ''[source]

qos: str = ''[source]

gres: str = ''[source]

time: str = ''[source]

cpus: int = 4[source]

mem: str = ''[source]

account: str = ''[source]

mail: str = ''[source]

property data_release_root: str[source]

Return type:: str

dataset_dir(run)[source]

e.g. <data_root>/data_release/o3b_dataset

Parameters:: run (str)
Return type:: str

data_dir(run)[source]

The per-run data_dir (recolour PSDs etc.).

Parameters:: run (str)
Return type:: str

noise_bin(detector, run)[source]

Monolithic noise .bin for one detector / observing run.

Parameters:

detector (str)
run (str)

Return type:

str

SERVERS: dict[str, Server][source]

get_server(name=None)[source]

Resolve the active server.

Order: explicit name arg -> $SAGE_SERVER -> hostname auto-detect. Raises a clear error (listing valid names and the current hostname) if it cannot be determined, so a run never silently writes to the wrong place.

Parameters:: name (str | None)
Return type:: Server