Skip to content

test: Deflake memory estimation test and restore its RSS regression coverage #2202

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
vdusek merged 2 commits into from
Sep 1, 2026
Merged

test: Deflake memory estimation test and restore its RSS regression coverage #2202

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
ba024b0
test: Deflake memory estimation test and restore its RSS regression c…
vdusek Aug 31, 2026
f907bd0
test: Tighten comments in the memory estimation test
vdusek Aug 31, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
52 changes: 33 additions & 19 deletions tests/unit/_utils/test_system.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -235,6 +235,7 @@ def parent_process() -> None:
children_count = 4
# Memory calculation is not exact, so allow for some tolerance.
test_tolerance = 0.3
measurement_rounds = 3

def no_extra_memory_child(ready: synchronize.Barrier, measured: synchronize.Barrier) -> None:
ready.wait()
Expand All @@ -254,7 +255,10 @@ def shared_extra_memory_child(
ready: synchronize.Barrier, measured: synchronize.Barrier, memory: SharedMemory
) -> None:
assert memory.buf is not None
print(f'Using the memory... {memory.buf[-1]}')
# Fault every page in: untouched pages never enter the RSS (hiding the overcount this test guards
# against) and are reclaimed first under memory pressure, which drops them from every mapper's PSS.
page_sum = sum(memory.buf[::4096])
print(f'Using the memory... {page_sum}')
ready.wait()
measured.wait()

Expand Down Expand Up @@ -293,28 +297,38 @@ def get_additional_memory_estimation_while_running_processes(

return (memory_during - memory_before).to_mb() / count

additional_memory_simple_child = get_additional_memory_estimation_while_running_processes(
target=no_extra_memory_child, count=children_count
)
additional_memory_extra_memory_child = (
get_additional_memory_estimation_while_running_processes(target=extra_memory_child, count=children_count)
- additional_memory_simple_child
)
additional_memory_shared_extra_memory_child = (
get_additional_memory_estimation_while_running_processes(
target=shared_extra_memory_child, count=children_count, use_shared_memory=True
# Under memory pressure the kernel reclaims cold pages, which silently leave the PSS readings and skew a
# round's deltas, so a distorted round is re-measured. A genuine overcount of shared memory misses the
# expectation several times over in every round, so the retries cannot mask it.
for _ in range(measurement_rounds):
additional_memory_simple_child = get_additional_memory_estimation_while_running_processes(
target=no_extra_memory_child, count=children_count
)
additional_memory_extra_memory_child = (
get_additional_memory_estimation_while_running_processes(
target=extra_memory_child, count=children_count
)
- additional_memory_simple_child
)
additional_memory_shared_extra_memory_child = (
get_additional_memory_estimation_while_running_processes(
target=shared_extra_memory_child, count=children_count, use_shared_memory=True
)
- additional_memory_simple_child
)
- additional_memory_simple_child
)

memory_estimation_difference_ratio = (
abs((additional_memory_shared_extra_memory_child * children_count) - additional_memory_extra_memory_child)
/ additional_memory_extra_memory_child
)
memory_estimation_difference_ratio = (
abs(
(additional_memory_shared_extra_memory_child * children_count)
- additional_memory_extra_memory_child
)
/ additional_memory_extra_memory_child
)

estimated_memory_expectation.value = memory_estimation_difference_ratio < test_tolerance
if memory_estimation_difference_ratio < test_tolerance:
estimated_memory_expectation.value = True
break

if not estimated_memory_expectation.value:
print(
f'{additional_memory_shared_extra_memory_child=}\n'
f'{children_count=}\n'
Expand Down
Loading