You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
ci(timing): the project ordering leg will hit its 20-minute step cap within about a week #2254
Problem. The weekly CI timing report (run 40288101249, on main91147a182) fails its two hard checks for Plain unittest ordering (project), step "Run tests/project and the root-level modules under plain unittest":
H2: the step's p90 is 11.8 min, at or above 50% of its 20-minute cap.
H4: on a linear fit it reaches the 20-minute cap in 7.3 days.
tests/project keeps growing: this leg, run locally, went from 2290 tests to 2388 over rounds 5 and 6. The report also names tests.test_tier_guard as the slowest module (over 30 s summed on 3.13), with several WorkflowAgreementTests cases over 10 s each.
Options
A. Raise the step cap. This buys time, but the trend continues.
B. Shard the step into two project ordering legs by module, as the other ordering legs are split. Each half stays well under the cap.
C. Make the slow tests cheaper. WorkflowAgreementTests looks like it re-reads or re-expands the workflows for each case; caching that parse per class would cut most of the module's time.
Recommendation: B plus C. B removes the trend, and C is cheap and helps every leg that runs the tier guard. A alone only defers the problem.
The ci-timing-report artifact of run 40288101249 is artifact 12636501186; its slowest-module table (Python 3.13, median of 5 push runs, top 8 of the 10 rows) follows.
Activity
JarryShaw commentedon Oct 26, 2026
B plus C
JarryShaw commentedon Oct 26, 2026
and might worth review and consolidate and remove unnecessary duplicated tests.
ci(ordering): shard the project ordering leg and cache the tier guard…
ci(ordering): shard the project ordering leg and cache the tier guard…
ci(ordering): shard the project ordering leg and cache the tier guard…
JarryShaw commentedon Nov 1, 2026
The
ci-timing-reportartifact of run 40288101249 is artifact 12636501186; its slowest-module table (Python 3.13, median of 5 push runs, top 8 of the 10 rows) follows.tests.test_tier_guardtests.protocols.test_frame_layer_limit_1610_unittests.protocols.internet.test_ext_chain_limit_rebuild_1619_unittests.protocols.misc.test_pcapng_unittests.test_support_helperstests.utilities.test_exceptions_excepthooktests.const.test_const_enum_builtin_paritytests.foundation.test_extraction_no_eofA fresh baseline was dispatched as run 41303273918 (#2554), after #2258 and #2244 changed the job shape and selection.