mirror of
https://github.com/kuhyx/testsAndMisc.git
synced 2026-07-04 15:23:03 +02:00
Krzysztof kuhy Rudnicki
20d5d1f89b
atop's `-P PRC` output inserts the clock-tick rate (HZ=100) between the
`state` and `utime` columns. Both the Python parser and the native C
aggregator read that constant as utime for every record, charging a flat
1 CPU-second per record — so cpu_seconds collapsed to pid_count and
short-lived fork-storm commands (xset, dd, chronyc) topped the CPU table
(xset showed 67h). The old test fixtures lacked the HZ field, so code and
tests agreed on the bug.
- _parse_prc / atop_agg.c: read utime/stime past the HZ field (after+2/+3,
tokens[10]/[11]); bump the length guards accordingly
- restore C/atop_agg (deleted in 89b4f59) under linux_configuration/C/,
where the build path resolves; corrected test fixtures to include HZ
- _atop_agg_binary: fall back to the Python parser when the C source tree
is gone instead of trusting an orphaned cached binary
- add regression tests proving HZ is not summed as CPU
- bundle the in-progress since-last-report multi-day aggregation (segments,
-b/-e bounding, persisted state, window merging) and its tests/conftest
- meta: gate linux_configuration/tests in pytest_changed_packages.py
Verified by running usage_report.py --date 20260604: Top CPU now led by
SkyrimSE; xset/dd/chronyc fall to ~0. C unit tests + full pytest suite green.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
41 lines
2.7 KiB
JSON
41 lines
2.7 KiB
JSON
{
|
|
"intent": "Stop the usage report from charging atop's per-record HZ field as CPU time, which made short-lived processes (xset, dd, chronyc, sleep) appear as the top CPU consumers (xset reported 67h of CPU in a 5h40m window). After the fix the CPU table reflects real consumers (SkyrimSE, zstd, the video-capture pipeline) and the fork storm shows only in the accurate PID-count column.",
|
|
"scope": [
|
|
"linux_configuration/scripts/periodic_background/system-maintenance/bin/_usage_report_parsing.py",
|
|
"linux_configuration/C/atop_agg/ (restored native helper with the same fix)",
|
|
"linux_configuration/tests/test_usage_report_since.py (regression tests)",
|
|
"Non-goal: rewriting the digital_wellbeing daemons that cause the fork storm"
|
|
],
|
|
"changes": [
|
|
"_parse_prc now reads utime/stime at after+2/after+3, skipping atop's HZ field that sits between state and utime; bumped _PRC_MIN_LEN 11 to 12.",
|
|
"_atop_agg_binary returns None (Python fallback) when the C source tree is absent, instead of trusting an orphaned cached binary; removed the stale ~/.cache/usage_report/atop_agg.",
|
|
"Restored C/atop_agg from git history into linux_configuration/C/atop_agg with the identical HZ-skip fix (tokens[10]/[11]), guard bumped to n<12, redundant PRM length check removed, and test fixtures corrected to include the HZ field.",
|
|
"Added Python regression tests asserting HZ is not summed as CPU and that a missing C source falls back to Python."
|
|
],
|
|
"verification": [
|
|
{
|
|
"command": "python3 usage_report.py --date 20260604 --no-clipboard --quiet",
|
|
"result": "pass",
|
|
"evidence": "Top CPU now led by SkyrimSE.exe 933s; xset/dd/chronyc dropped out entirely (real CPU ~0). Cross-checked against atop directly with corrected field indices."
|
|
},
|
|
{
|
|
"command": "make test (linux_configuration/C/atop_agg)",
|
|
"result": "pass",
|
|
"evidence": "atop_agg tests: OK. Rebuilt binary emits xset cpu_ticks=0 vs 24427000 before."
|
|
},
|
|
{
|
|
"command": "python3 -m pytest test_usage_report_since.py -k 'parse_prc or atop_agg_binary'",
|
|
"result": "pass",
|
|
"evidence": "4 passed. Buggy indices would yield 107 ticks vs the asserted 10, so the regression test fails against the old code."
|
|
}
|
|
],
|
|
"risks": [
|
|
"Native fast path needs a C compiler; without cc the report now falls back to the (slower) Python parser rather than a stale binary.",
|
|
"C helper coverage remains below 100% on defensive OOM/hash-full paths (pre-existing; the suite is not coverage-gated for linux_configuration)."
|
|
],
|
|
"rollback": [
|
|
"git checkout the parsing module and remove linux_configuration/C/atop_agg to revert.",
|
|
"Re-run usage_report.py --date 20260604 and confirm whether xset reappears with inflated CPU."
|
|
]
|
|
}
|