What was accepted
bin/fm-test-run.sh sets DEFAULT_PER_SCRIPT_TIMEOUT_SECS=480, and that value is settled. This issue records the exposure that value leaves on one lane, so that accepting it is a written decision rather than a comment nobody revisits.
Evidence
.github/workflows/ci.yml runs bin/fm-test-run.sh --family real-herdr-gated, which is standard mode, so the 480s per-script default is armed.- The slowest measured script in that family is the Herdr presentation E2E at 341s, recorded in
bin/fm-test-run.sh.
- Nothing enforces a per-script ceiling on that lane. The only enforced ceiling is the family step's
timeout-minutes: 20 (1200s).
- So the real margin is 480/341 = 1.41x, materially thinner than the 2x+ margin the portable lanes carry.
The bound-rationale comment previously argued a ~420s per-script ceiling as "a hard constraint rather than a statistical one". That premise is not enforced anywhere, and it is being corrected to state the measured margin instead, under fm-test-suite-robustness.
The exposure
A healthy-but-slow Herdr E2E running 1.41x its measured time under CI load is terminated at 480s and reported as exit=124, on a required lane. That reads as a hang when it is a slow pass.
Deliberately NOT done
Widening the bound or adding a family exemption was refused. Widening a bound to buy comfort converts a true intermittent signal into a permanent false pass that nobody revisits, and the 480s value is settled on its own merits.
Reopen trigger
The first exit=124 on real-herdr-gated where the terminated script is later shown to have been healthy. At that point the question is not "widen the bound" but "why did that script get 1.41x slower", and the answer decides which fix is correct.
Refs #178, #220.
What was accepted
bin/fm-test-run.shsetsDEFAULT_PER_SCRIPT_TIMEOUT_SECS=480, and that value is settled. This issue records the exposure that value leaves on one lane, so that accepting it is a written decision rather than a comment nobody revisits.Evidence
.github/workflows/ci.ymlrunsbin/fm-test-run.sh --family real-herdr-gated, which is standard mode, so the 480s per-script default is armed.bin/fm-test-run.sh.timeout-minutes: 20(1200s).The bound-rationale comment previously argued a ~420s per-script ceiling as "a hard constraint rather than a statistical one". That premise is not enforced anywhere, and it is being corrected to state the measured margin instead, under fm-test-suite-robustness.
The exposure
A healthy-but-slow Herdr E2E running 1.41x its measured time under CI load is terminated at 480s and reported as
exit=124, on a required lane. That reads as a hang when it is a slow pass.Deliberately NOT done
Widening the bound or adding a family exemption was refused. Widening a bound to buy comfort converts a true intermittent signal into a permanent false pass that nobody revisits, and the 480s value is settled on its own merits.
Reopen trigger
The first
exit=124onreal-herdr-gatedwhere the terminated script is later shown to have been healthy. At that point the question is not "widen the bound" but "why did that script get 1.41x slower", and the answer decides which fix is correct.Refs #178, #220.