lei.chen/myxce
1
0
Fork 1
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

feat(controller-peers-env-fallback): fall back to XCE_CONTROLLER_PEERS in the worker (task 3.2)

Browse source 
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
This commit is contained in:
Rongsong Shen 2026-10-03 07:07:19 +00:00
parent 55193fcb4b
commit 05a7e6dd0c
Signed by untrusted user: rshen
GPG key ID: 80533536B0EB651C
3 changed files with 203 additions and 17 deletions
Download patch file Download diff file Expand all files Collapse all files

4
.kiro/specs/controller-peers-env-fallback/tasks.md
View file

@ -57,7 +57,7 @@
- Done when, after rebuilding the scheduler, the scheduler suite passes and every scheduler case in the regression script passes
- _Requirements: 1.1, 1.2, 1.3, 1.4, 1.5, 1.7, 1.8_
- [ ] 3.2 Fall back to the environment variable in the worker
- [x] 3.2 Fall back to the environment variable in the worker
- Add the worker cases to the regression script first and see them fail, using the virtual worker: env list used ("HA cluster config: N candidate(s)" on stdout, since these lines are logged before the worker's log file exists); `--peers` winning; neither set (extended no-peers line); parse error and port-0 validation error from each source naming that source on stderr; `-h` naming the variable, the opt-out, and that `sudo` drops it unless preserved
- Use the variable only when no `--peers` is given; name the source in both error messages; extend the no-peers line; update the usage text
- Done when, after rebuilding the worker, the worker suite passes and every worker case in the regression script passes
@ -137,3 +137,5 @@
- (2.3) The 1.8 wording cannot contain the old "No --peers flag; running ..." substring, so the 2.2 flush case now matches the full new line. Env-sourced "HA mode" lines end " (from XCE_CONTROLLER_PEERS)"; explicit-`--peers` lines and error texts are unchanged apart from naming the source. Reword the script header's "while the case list is empty" comment in 3.x when the shellcheck disable is dropped.
- (3.1 → 3.2) The scheduler's validation message now quotes the value, names the source and hints "(empty host or port 0?)" (the old "missing identity?" was wrong: identity is always set there, and `xce_parse_peers` caps the candidate count itself). Use the same shape for the worker's `xce_cluster_config_validate()` message.
- (3.1 → 5.2) On this machine `test-version-startup-consistency.sh` and `test-dirty-fingerprint-startup-log-unchanged.sh` exit 1 for an environmental reason: their apiserver `go build` fails with "module lookup disabled by GOPROXY=off" (no Go module cache). Their scheduler checks pass. Task 5.2's "both scripts pass" needs a machine with the Go module cache, or the result must be reported as environment-limited with the scheduler checks shown passing.
- (3.2) The worker's early stdout lines are not buffered (visible within 0.3s while it runs). `xcecd_virtual` ignores SIGTERM (pre-existing), so `start_worker` SIGKILLs it after the asserted line; the virtual worker leaves only its scratch log. Its peers parse/validation early returns now free `logdir` (they leaked under ASAN). Out of scope, pre-existing: `-l <dir>` combined with `-h`/`--version` still leaks `logdir`.
- (3.2 → 3.3) The file-wide `shellcheck disable=SC2329` must stay (cases and helpers run via the `"$fn"` registry and `trap`); the SC2034 disable now covers only the `HARNESS_BIN` line — remove it once the harness cases use it.