PrincetonUniversity / slurmise / 33889246435

91%

push

github

feat: record known usage without querying sacct (#107)

* feat: accept known usage on raw-record instead of querying sacct

raw-record always asked sacct for max RSS and elapsed time, so a record could
only be written on a machine with a working scheduler. Add --memory and
--runtime to supply both directly; when both are given the record is treated as
already processed and sacct is not consulted.

This lets a database be populated where no scheduler is present, which is what
makes an offline walkthrough of slurmise possible. Supplying only one of the
two still defers to sacct for both, as before.

Assisted-by: Claude Code:claude-opus-5

* fix: reject half-specified usage on raw-record

Passing only --memory or only --runtime fell through to sacct, which then
overwrote the value with its own. Refuse the pair outright rather than
silently discarding what the caller measured.

Decide this at the CLI rather than in raw_record, so the API no longer
overrides a caller that explicitly passed processed_data=False.

Add a no_sacct fixture for tests whose contract is that the scheduler is
never consulted. Leaving sacct unmocked is not equivalent: on a cluster it
answers with live accounting data rather than failing.

Assisted-by: Claude Code:claude-opus-5

* refactor: rename flags, validate in API not CLI

Rename --memory/--runtime to --used-mbs/--used-seconds which is
arguably a bad idea. Mixture of minutes/seconds in slurmise currently.

Infer whether sacct is needed from the job object rather than using
`processed_data` flag. Raise when only time or mem are supplied to API.

Assisted-by: Claude Code:claude-opus-5

* remove processed_data

* fix: Cleanup main merge into PR issue

- Also remove remaining processed_data flag

* chore: Use minutes instead of seconds after PR 112

---------

Co-authored-by: Troy Comi <troycomi@gmail.com>

293 of 334 branches covered (87.72%)

Branch coverage included in aggregate %.

34 of 36 new or added lines in 4 files covered. (94.44%)

2191 of 2407 relevant lines covered (91.03%)

0.91 hits per line