Repository navigation
Test modules keep cached tax-benefit systems and simulations for the rest of the test run #2357
Description
Five Python test modules cache tax-benefit systems or simulations in module-level functools.lru_caches that are never cleared, so everything they hold stays alive for the rest of the pytest worker's run:
| Module | Cache | Holds |
|---|---|---|
test_two_child_limit_age_exemption_properties.py |
tax_benefit_system(parameter, value), maxsize=None |
one fresh CountryTaxBenefitSystem plus a reform per (parameter, value) drawn: up to 21 |
test_cliff_evaluated_properties.py |
tax_benefit_system(mtr_adults), maxsize=None |
4 reforms, each over its own fresh system |
test_state_pension_age.py |
differential_simulation(), situation_template() |
a ~10,000-person simulation with every year it calculated, and a template simulation |
test_state_pension_credit_qualifying_age.py |
its own copy of differential_simulation() (pytest imports test_state_pension_age a second time under policyengine_uk.tests.), one_person_units_template() |
as above |
test_uc_work_related_groups_properties.py |
qualifying_age_couples() |
a simulation and the system it built |
Each cached system also keeps the last simulation built on it alive: Simulation.__init__ sets tax_benefit_system.simulation = self.
On CI run 38165445604 (branch mem-measure/2331-run2b, per-worker memprobe artifacts), the worker that ran the two-child-limit file went from 1.10 to 1.65 GiB RSS during that file (13 systems built) and stayed there; malloc_trim after the module did not return it, because the cache still referenced the systems.
Goal: release this memory when each module finishes, and hold less while it runs, without changing which tests run or their outcomes.
Activity