myxce/.wolf/bugs/bug-1924.md

id	status	status_updated	status_note	component	severity	timestamp	last_seen	occurrences	files	tags	related	fix_commits	gitea_issue	gitea_synced_status	gitea_synced_hash
bug-1924	fixed	2026-10-08		scheduler	noise	2026-10-08T09:30:37Z	2026-10-08T09:30:37Z	1	src/scheduler/testsuite/test_clock.c	test flaky clock tlog		e56f27f9	349	fixed	03309e9ff280

bug-1924: scheduler core clock test fails when the test PID contains 1999 (tlog line includes getpid())

Description

Observed: "scheduler core clock test" failed once in a full scheduler gate run (task 2.2 of scheduler-sim-decision-trace, 2026-10-08), then passed 5/5 on rerun:

../scheduler/testsuite/test_clock.c:351:test_tlog_format_shows_frozen_virtual_time_not_infotime:FAIL: formatted log line must NOT show info->time's wall-clock year while the virtual clock is frozen
15 Tests 1 Failures 0 Ignored

Expected: the test passes on every run. The formatter output is correct; only the assertion is too broad. Impact: noise only. A random, rare failure of the scheduler suite (and of /xce-build scheduler) that has nothing to do with the change under test, costing a rerun and an investigation. No production effect. Found by: spec task scheduler-sim-decision-trace 2.2 (full-suite gate run).

The test sets info->time.year = 1999 and asserts strstr(buff, "1999") == NULL over the WHOLE formatted line. xce_core_tlog_format() (src/scheduler/src/core_clock.c:238-275) prints [%5d] with getpid() in that line, and the frozen virtual time (2033-05-18 03:33:20,000) is constant. So the assertion fails whenever the test process's PID contains the digits "1999" (e.g. 2619991). Confirmed by forcing such a PID (see Reproduction).

Environment

• OS: no /etc/os-release; Docker container; NixOS kernel; kernel Linux 6.18.33 x86_64
• libc: glibc 2.42-47 (interpreter /nix/store/wb6rhpznjfczwlwx23zmdrrw74bayxw4-glibc-2.42-47/lib/ld-linux-x86-64.so.2)
• Toolchain: gcc (GCC) 15.2.0, meson 1.9.1
• Build: asan (xce.build)
• Code: scheduler-framework @ 54f9139e (first seen on the tree with task 2.2 uncommitted; test_clock.c and core_clock.c unchanged by it)
• Deployment: none (unit test binary sched_core_clock_test, meson suite scheduler, test name "scheduler core clock test")

Reproduction

1. Build: bash src/tools/xce-build.sh scheduler (ASAN, src/xce.build).
2. sh /home/claude/tmp/bug-1924/repro.sh src/xce.build/scheduler/testsuite burns PIDs until the test binary runs under a PID containing "1999", then prints the result.

• Observed 2026-10-08: pid=2612000 PASS; pid=2619991 test_tlog_format_shows_frozen_virtual_time_not_infotime: FAIL (15 Tests 1 Failures).
• Frequency: deterministic for a PID containing "1999", otherwise never. Of the PIDs 1..4194304 (this host's pid_max), 2240 contain "1999" (0.053%); in the 2.0M-2.7M range the host was using, 0.034%. The original failure was one such unlucky run.
• Evidence: /home/claude/tmp/bug-1924/ (clock-pid1999.out, the forced-PID run; gate22-excerpt.log, the original gate failure; repro.sh).

Root cause

test_tlog_format_shows_frozen_virtual_time_not_infotime (src/scheduler/testsuite/test_clock.c) set info->time.year = 1999 and asserted strstr(buff, "1999") == NULL over the whole formatted line. xce_core_tlog_format() (src/scheduler/src/core_clock.c:238-275) also writes getpid() into that line ([%5d]). With the virtual clock frozen, the timestamp is constant (2033-05-18 03:33:20,000 UTC, via gmtime_r in ns_to_tlog_time), so the only variable text is the PID. Any PID containing "1999" made the assertion fail although the formatter was correct. The defect was in the test, not the formatter; pre-existing since the test was added (scheduler-core task 4.5). Scope: the other tlog-format tests in test_clock.c assert specific substrings and do not search the whole line for a value the PID can contain.

Fix

test_clock.c: the test now asserts the exact timestamp field, TEST_ASSERT_EQUAL_STRING_LEN_MESSAGE("[2033-05-18 03:33:20,000] ", buff, 26, ...), instead of searching the whole line for "1999". That is both stricter (it pins the frozen virtual time) and independent of the PID. It still fails if the formatter prints info->time (1999-01-01), so it still catches the defect it was written for.

Verification

• Test: scheduler core clock test, test_tlog_format_shows_frozen_virtual_time_not_infotime in src/scheduler/testsuite/test_clock.c. RED before the fix: forced with /home/claude/tmp/bug-1924/repro.sh, the old test FAILED under pid=2661999 (15 Tests 1 Failures). GREEN after: the fixed test PASSED under pid=2671999 (15 Tests 0 Failures), and passes in a normal run.
• Suites: scheduler core clock test OK on the branch build (asan, src/xce.build) after the cherry-pick; standalone run 15 Tests 0 Failures.
• Live: n/a, a test-only change with no production code touched.

References

• Related: none
• Found by: .kiro/specs/scheduler-sim-decision-trace task 2.2 gate run