wegamekinglc / calc-flow / 35389729921
93%

DEFAULT BRANCH: main
Ran
Jobs 3
Files 198
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 93.173% (+0.001%) from 93.172%
35389729921

push

github

web-flow 
perf: optimize DataFusion benchmark harness (DAL-276) (#293)

* perf: remove bench sample retention and RSS polling (DAL-276)

The paired SQL/DataFusion benchmark's timed windows carried two
harness-owned costs: a helper thread polling /proc/self/status every
millisecond inside every timed sample, and one retained output batch
per timed sample (never read after collection, correctness uses the
warm-up pair).

Read the kernel-maintained VmHWM high-water mark instead, re-anchored
per sample through clear_refs, so peaks are exact and cost nothing
inside the timed window; retain outputs only for the warm-up
correctness pair and drop them before the sampling loop.

Outputs, phase boundaries, and the evidence schema are unchanged.
On the nightly matched-adaptive profile this removes 9-18% of
measured median latency and 30-69% of peak RSS while tightening
sample CV. The bench self-tests now run through
scripts/run_rust_tests.py so the new RSS-window and retention
contracts bind in CI.

Co-authored-by: multica-agent <github@multica.ai>

* fix: gate rss-window self-test to Linux (#293)

The rss-window self-test asserts Linux-only /proc contracts
(VmRSS/VmHWM and the clear_refs reset). Now that
scripts/run_rust_tests.py drives this bench's self-tests, the
Windows rust-tests job would execute it too and fail
deterministically when both reads fall back to the floor value.
Compile the test and its call site only on Linux; the sampler
itself keeps its cross-platform floor. Also fold the duplicated
kilobyte-to-byte mapping into one status_field_bytes helper.

Co-authored-by: multica-agent <github@multica.ai>

* refactor: split retention self-test below CCN 8 (#293)

Codacy measures cyclomatic complexity with a CCN 8 limit and flagged
run_sample_retention_tests at 15; the repository's own ratchet only
counts lizard-forgive markers, so the finding surfaced in CI. Split
the test into retention_case_args, correctness_is_complete, and
timed_sample_drops_output helpers in... (continued)

103537 of 111123 relevant lines covered (93.17%)

70284.07 hits per line

Coverage Regressions

Lines Coverage ∆ File
1
88.75
-0.09% src/calc_flow_studio/run_manager.py
Jobs
ID Job ID Ran Files Coverage
1 rust - 35389729921.1 147
93.01
GitHub Action Run
2 python - 35389729921.2 41
95.36
GitHub Action Run
3 studio - 35389729921.3 10
92.17
GitHub Action Run
Source Files on build 35389729921