about summary refs log tree commit diff
path: root/test/python/server_test_case.py
diff options
context:
space:
mode:
authorAlexander_Kabui <alexanderkabua@gmail.com>2026-10-07 05:35:59 +0300
committerAlexander_Kabui <alexanderkabua@gmail.com>2026-10-07 05:35:59 +0300
commit228bedf08631704a2f884d58f9a70d26783fb1a7 (patch)
tree668de5eec5d37faaa5b741567231b0ec4408ee7d /test/python/server_test_case.py
parent1597f51055ac98b308d7c66112a3d95bd83cacc9 (diff)
downloadgenenetwork4-228bedf08631704a2f884d58f9a70d26783fb1a7.tar.gz
tests/search_engine.rs (timeout stub): stop parking a runtime worker in serve_slow.
The 2s stall ran as a bare std::thread::sleep inside the axum handler,
which parks a tokio worker thread. When the reqwest client future's wake
then lands on the parked worker, the GN3_TIMEOUT_SECS=1 cancel is
processed only after the response completes, and the test measures
~2.0s instead of <1.9s -- a scheduler-dependent ~30% flake, measured
3/10 isolated failures both at pristine C1.1 (1597f51) and at the
P0 baseline run (the first full-suite run of this lane hit it under
load; the isolated 10x runs are the discriminating evidence). Move the
sleep onto the blocking pool via spawn_blocking, the same house pattern
every src/routes handler uses for blocking work. Test semantics
unchanged: the response still stalls 2s, the 1s timeout still fires
first.

Evidence: ~/spikes/port-execution/logs/flakefix.log (10/10 isolated
passes after the fix, full suite EXIT=0); pre-fix logs:
baseline-raw.log, baseline-raw.log.2, c1-2.log.
Diffstat (limited to 'test/python/server_test_case.py')
0 files changed, 0 insertions, 0 deletions