Twenty suites still cap a server start at 20 s, and one of them failed a 10-way run #1720
Labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
jaylouisw/packetforge#1720
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
The measurement
PF_TEST_CONCURRENCY=10 node scripts/run-tests.jsonf4e22080, 2026-11-16 20:05 UTC:Not an assertion — a crash, in the restart half of the suite (line 83), after the deadline expired:
The
|is the child's captured output, and it is empty: the second server never printed its bannerin 20 s, so it had not reached
listen.Why 20 s is the wrong number
/api/catalog200So a start is 4.8 s idle-load and this suite spends ~11 s more than that just being slow inside a
10-way run, which leaves the restart's 20 s cap a couple of seconds of headroom. It missed once in
10227 assertions, which is exactly how a red CI leg looks: nothing is broken, the bound is too tight.
#1714 measured this and #1715 set the bound to 60 s for 13 suites. These 20 still say 20 s — the
class was fixed where it had already failed, not where it was about to.
The fix
test/*.test.jsgets one minute, so a slow machine fails for beingbroken and not for being busy.
Date.now() + <literal>under a minuteanywhere in
test/.board-id-is-a-namegains the #1715 assertion — a port nothing is listening on fails at thedeadline it was given, not after a fixed number of tries — because a bound nothing exercises is
how this class came back.
Found in the issues loop; no code under test is at fault.