myxce/.wolf/bugs/bug-1866.md
Rongsong Shen 25d8a461fa
chore: gitea-sync refreshes 23 recent bug issues to the record format
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-09-28 03:44:30 +00:00

2.9 KiB

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-1866 fixed 2026-09-27 worker leak 2026-09-25T15:42:02Z 2026-09-25T15:42:02Z 1
src/worker/src/xcecd_worker_utils.c
worker
leak
asan
e84a8d43
285 fixed fa33c9cef96e

bug-1866: xcecd_virtual leaks up to five strings per finished job (LeakSanitizer at exit)

Description

Observed:

Direct leak of 12 byte(s) in 12 object(s) allocated from:
    strndup  add_worker (xcecd_worker_utils.c:140)
    <- xcecd_process_command <- handle_message_from_queue

Expected: no leaks reported when xcecd_virtual exits. Impact: one leaked string per job that carries a launch type, plus one per queue/job pre- or post-exec command; grows with the number of finished jobs for the life of the daemon. Found by: LeakSanitizer at xcecd_virtual exit after a job burst run, 2026-09-25.

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), leak detection on at process exit
  • Code: scheduler-framework @ d87b78dc
  • Deployment: xcecd_virtual worker; the burst run's cluster topology was not recorded

Reproduction

  1. Check out d87b78dc and add test_remove_worker_frees_every_startinfo_copy from e84a8d43 to src/worker/testsuite/test_finished_job_retention.c.
  2. /xce-build worker
  3. ASAN_OPTIONS=detect_leaks=0 src/xce.build/worker/testsuite/finished_job_retention_test
  • Frequency: deterministic (58 bytes leaked per add/remove round trip)
  • Evidence: the unit test; the original burst-run log was not kept.

Root cause

add_worker() strndup()s launch_type, q_pre_exec, q_post_exec, job_pre_exec and job_post_exec out of the job's start info, but remove_worker() only freed exechome, execcwd, execuser, job_cmd, pids and job_exdata. LeakSanitizer reported only the launch_type copy because the burst's jobs had no pre/post-exec commands. Pre-existing, origin unknown. remove_worker() is the only place a worker entry is freed, so no other path is affected.

Fix

remove_worker() also frees launch_type, q_pre_exec, q_post_exec, job_pre_exec and job_post_exec (src/worker/src/xcecd_worker_utils.c).

Verification

  • Test: test_remove_worker_frees_every_startinfo_copy in src/worker/testsuite/test_finished_job_retention.c compares ASAN's live-heap bytes across an add/remove round trip (the suite runs with detect_leaks=0): RED before the fix leaked 58 B, the five strings; GREEN after, 0 B
  • Suites: worker 12/12, asan
  • Live: n/a; the leak is in a pure unlink/free path the unit test covers completely

References

  • Gitea: #285