Rongsong Shen
d26ff8870d
The buglog was one 27K-line JSON file whose fields covered only symptom,
root cause and fix, as free text: 2% of entries named the code version a
bug was found on, 12% the build profile, 36% a way to reproduce, and 58% of
the file was OpenWolf hook auto-detections ("Missing guard clause",
"Function not marked async" on .md files).
- .claude/skills/buglog/bug-record.md: the record standard. Eight
sections (Summary as the title, Description, Environment, Reproduction;
Root cause, Fix, Verification, References once fixed), each with the
content it must contain -- e.g. Environment needs the OS, the libc the
failing binary loads (from its ELF interpreter: glibc 2.42 in the nix
shell vs 2.40 on the host), toolchain, build profile and commit.
- .wolf/bugs/<id>.md: 500 records migrated from buglog.json; every field
verified to round-trip. They carry legacy: true and an explicit "not
recorded" where the old log had no such field; component derived from
file paths, severity from keywords (else unknown), fix_commits from
hashes in status_note/fix that git resolves (91 of 463 fixed).
- .wolf/buglog-archive.json: the 710 hook auto-detections, verbatim.
- .wolf/buglog.json is now an index generated from the records, keeping
the fields OpenWolf's hooks and `openwolf bug search` read.
- scripts/buglog.py rewritten: new (template + Environment collected from
the machine and --binary), check, status (refuses a status the record
doesn't support; fixed needs Root cause, Fix, Verification and
--commit hashes git resolves), note, recur, set, show, list, triage,
env, index. 29 unittest cases in scripts/tests/test_buglog.py.
- scripts/git-hooks/pre-commit runs `check --staged`: incomplete staged
records, a legacy record hand-changed to fixed without the fix-time
sections, a new record claiming legacy: true, or a stale index refuse
the commit.
- scripts/gitea-sync.py reads the records: issue body = the record's
sections plus linked fix commits (origin remote, GITEA_COMMIT_REPO to
override), and a body hash so an issue's text is refreshed when the
record changes.
- bug-1866 rewritten to the full standard as the worked example.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
38 lines
3.5 KiB
Markdown
38 lines
3.5 KiB
Markdown
---
|
|
id: "bug-1575"
|
|
status: "fixed"
|
|
status_updated: "2026-09-24"
|
|
status_note: ""
|
|
component: "unknown"
|
|
severity: "unknown"
|
|
timestamp: "2026-08-18T03:15:00Z"
|
|
last_seen: "2026-09-19T09:00:00Z"
|
|
occurrences: 1
|
|
files: [".claude/skills/xce-test-resources/run-resource-verify.sh"]
|
|
tags: ["trap", "cleanup", "pkill", "SIGTERM", "signal-handler", "race-condition", "run-resource-verify.sh", "resource-value-verification", "bash"]
|
|
related: ["bug-1436"]
|
|
legacy: true
|
|
---
|
|
|
|
# bug-1575: run-resource-verify.sh's trap-based cleanup() (EXIT/INT/TERM) non-deterministically left the…
|
|
|
|
## Description
|
|
run-resource-verify.sh's trap-based cleanup() (EXIT/INT/TERM) non-deterministically left the scheduler and/or apiserver process running after the script was interrupted mid-deployment via an external SIGTERM -- confirmed processes still alive (e.g. `xce_scheduler --peers ... --logfile .../scheduler.log`, `/tmp/apiserver -d .../apiserver`) seconds after the orchestrator's own `wait` returned with exit code 143. The SAME cleanup() code, reached via a normal fail()/exit-1 path (no external signal involved) inside the same single foreground invocation, worked perfectly every time (verified zero leftover processes).
|
|
|
|
## Environment
|
|
_Not recorded (migrated from buglog.json)._
|
|
|
|
## Reproduction
|
|
_Not recorded (migrated from buglog.json)._
|
|
|
|
## Root cause
|
|
cleanup() issued four SEPARATE sequential `pkill -9 -f -- "<pattern>"` invocations (one per component: controller/worker/scheduler/apiserver), each forking+execing its own `pkill` subprocess. When cleanup() ran as a normal function call/return within the script's own top-to-bottom control flow (e.g. from fail()), all four completed reliably. When cleanup() instead ran from on_exit as the EXIT trap fired by an asynchronously-delivered SIGTERM (via a separate `trap 'exit 143' TERM` translating the signal into an explicit exit), the sequence of pkill forks was intermittently and non-deterministically cut short partway through (confirmed via added debug instrumentation: one run stopped after the 1st pkill, another after the 3rd, another after the 2nd -- no fixed failure point, consistent with a timing race rather than a deterministic pattern-matching bug). Root cause not fully isolated at the syscall level (ruled out: pattern-matching correctness -- each individual pkill pattern was independently verified correct via direct pgrep; ruled out: cgroup pids/resource exhaustion -- process count stayed low, ~20-25 total throughout). Most likely a bash/signal-trap interaction where starting new child processes (fork/exec for each pkill) from within a trap handler invoked asynchronously (as opposed to synchronous top-level exit) has some window of unreliability in this environment.
|
|
|
|
## Fix
|
|
Rewrote cleanup() to (1) combine all four per-component patterns into ONE `pkill -9 -f` call using extended-regex alternation (`"pat1|pat2|pat3|pat4"`) instead of four separate invocations -- fewer fork/exec round trips through the trap-handler code path -- and (2) wrap it in a retry-until-verified loop (`pgrep -f` check, up to 10 attempts at 0.2s intervals) so a slow/racy single kill attempt can no longer silently leave a process behind. Verified reliable across 5+ repeated live runs deliberately interrupted via external SIGTERM at different points mid-deployment (right after controller-only confirmation, and after controller+scheduler confirmation) -- zero leftover processes every time after the fix, versus non-deterministic leftovers before it.
|
|
|
|
## Verification
|
|
_Not recorded (migrated from buglog.json)._
|
|
|
|
## References
|
|
- Related: bug-1436
|