fix(ci): 僵尸run复活防护——重负载job自检+dedupe竞态增强

P1修复(2026-08-30 CI巡检实证):
Gitea 1.26.4 调度器bug:被 concurrency cancel 的 run,其 job 会在数小时后
被复活重跑(attempt=2)。实测 42597 号 run 18:46 被取消,21:46 复活,
与正常流水线并行烧了1.5小时 runner(Validate+UnitTests+staging构建全套),
合并高峰双流水线CPU超卖导致单测 14分→30分。

改动:
1. 新增 scripts/ci/ci_run_selfcheck.sh:job 启动第一步实时查本 run 状态,
   若 conclusion 为 cancelled/skipped(被复活的僵尸job)立即退出(exit 78),
   API 故障时不阻断(unknown 照常执行)
2. ci-pipeline.yml 12 个重量级 job(validate×3、unit/integration/frontend×2、
   build-pr/build-staging/staging-e2e/api-tests/build-production)首步插入自检
3. dedupe-check 情形2增强:同一 head_sha 的 push 流水线不只是在跑时去重,
   已 success 完成时 PR 侧测试同样跳过(消除竞态窗口双跑)
This commit is contained in:
xiaoxia
2026-08-30 23:04:40 +08:00
parent 301b591037
commit 3fcb66e40d
2 changed files with 109 additions and 1 deletions
+34
View File
@@ -0,0 +1,34 @@
#!/bin/bash
# CI job 运行前自检 —— 防 Gitea 1.26 调度 bug:
# 被 concurrency cancel 的 run,其已派发的 job 可能在数小时后被调度器复活重跑
# (2026-08-30 实测:18:46 取消的 run,21:46 复活 attempt=2,与正常流水线抢 runner 1.5 小时)。
# job 启动第一步先实时查本 run 状态,若已终结(cancelled/skipped),容器立即退出不烧资源。
#
# 用法:在重量级 job(Validate/Unit Tests/Integration/Frontend/Build/E2E 类)的
# 首个 step 中加入:bash scripts/ci/ci_run_selfcheck.sh
set -u
echo "▶ CI 运行前自检(僵尸 run 复活防护)..."
if [ -z "${GITHUB_API_URL:-}" ] || [ -z "${GITHUB_REPOSITORY:-}" ] || [ -z "${GITHUB_TOKEN:-}" ] || [ -z "${GITHUB_RUN_ID:-}" ]; then
echo "⚠️ 缺少 GITHUB_* 环境变量(GITHUB_RUN_ID/GITHUB_TOKEN),跳过自检(不阻断)"
exit 0
fi
CONCLUSION=$(curl -sf -H "Authorization: token ${GITHUB_TOKEN}" --max-time 10 \
"${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/actions/runs/${GITHUB_RUN_ID}" 2>/dev/null \
| python3 -c "import json,sys; print(json.load(sys.stdin).get('conclusion') or 'running')" 2>/dev/null || echo "unknown")
case "${CONCLUSION}" in
cancelled|skipped)
echo "::error::本 run(${GITHUB_RUN_ID}) 状态已为 ${CONCLUSION},却仍被调度器派发执行——判定为 Gitea 复活的僵尸 job,立即退出,不再占用 runner"
exit 78 # EX_CONFIG:非零退出让调度器知晓该 job 终结(避免被当成功空跑)
;;
running|unknown|"")
# unknown 时不阻断(API 临时故障),正常执行
echo "✅ run ${GITHUB_RUN_ID} 状态正常(${CONCLUSION:-running}),继续执行"
;;
*)
echo "✅ run ${GITHUB_RUN_ID} 结论=${CONCLUSION}(已终结但非取消),继续执行"
;;
esac