Repository navigation
fix(tests): main's 3.14 and 3.15 legs time out on the ethernet/header-only-unknown-type/cut7 round trip #1736
Description
Describe the bug
main's Unit Tests fail one subtest on two consecutive runs. In tests/protocols/link/test_link_edge_roundtrip_unit.py::LinkEdgeRoundTripTests::test_cut_is_rejected_or_byte_exact, case ethernet/header-only-unknown-type/cut7 exceeds the test's 30 s per-case timeout:
| sha | leg | result | job |
|---|---|---|---|
fb7027b25 (#1725) |
Python 3.14 | TIMEOUT: did not finish within 30s |
115957996493 |
59dd2613d (#1728) |
Python 3.15 | REBUILD: TimeoutError: did not finish within 30s |
115968208378 |
Every other test in both runs passes (1 failed, 4341 passed / 1 failed, 4345 passed). The previous run, on 51cd24c70, was green on both legs. #1725 touches only CI files, so the cause is either a slow path in the parse or rebuild of that cut, sitting close to the 30 s cap and pushed over it by runner load, or a hang. That has not been measured yet.
Expected behavior
The case finishes well inside its cap, so main stays green.
Additional context
A worker is root-causing it. The fix needs a test that fails without it.
Activity
JarryShaw commentedon Oct 14, 2026
Fixed on
mainine530f39ad, a test-only push. The cause was in the test harness, not pcapkit._edge_roundtrip.pystarted each case's 30 s deadline before importing the case's class.setUppurges pcapkit, so the first case in each class, cut7 here, carried the whole package import (about 480 ms) inside a budget meant for a case that takes about 4 ms.mainby pinning the run to about 1% of a CPU. The fix passed under the same load.test_edge_roundtrip_deadline_1736_unit.pychecks that the module is loaded before the deadline starts. It fails 2 of 2 on the old harness and passes on the fix. Parse and rebuild outputs for all 1015 edge cases are byte-identical before and after.