AWS installer (beta): an interrupt during the deploy wait can occasionally be ignored

未关闭
#1,413 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

还没有人认领这个 Issue。

评估

难度
3/5
预计耗时
1-2 天
新手友好度
74/100
Issue 类型
缺陷
描述清晰度
描述清楚
活跃度
活跃
技术栈
aws, bash
领域
cli

调研方向

Start with deploy/aws-installer/lib/steps.sh, especially cv_deploy_wait, and inspect the heartbeat command substitutions in steps.sh and checks.sh. Run the installer test test_an_interrupt_leaves_a_usable_record_and_no_child[during-the-deploy-to-the-installer-INT] with the wait tick set to 0 to reproduce the race. Done means both direct-process and process-group SIGINT deliveries stop the installer, leave a usable record and no child, and the regression test covers both deliveries.

由索引模型根据 Issue 内容生成。

描述

bug post-launch

What happens

While controlvar-aws install waits for a CloudFormation deploy, an interrupt (SIGINT) can occasionally be ignored,
so the install carries on instead of stopping and writing its record. This applies to an interrupt sent to the
installer's own process (for example kill -INT <pid>) and, more rarely, to Ctrl-C in the terminal (which signals the
whole process group): pressing Ctrl-C again stops it. SIGTERM is acted on.

Why

bash does not replay a trapped SIGINT that arrives while the shell is collecting a command substitution $(...)
whose child has finished (command substitutions are not in bash's jobs table). The deploy-wait loop takes the time with
now=$(date +%s) on every tick (deploy/aws-installer/lib/steps.sh, cv_deploy_wait), and the first heartbeat runs
a few more substitutions (steps.sh, checks.sh).

Measured

The installer test test_an_interrupt_leaves_a_usable_record_and_no_child[during-the-deploy-to-the-installer-INT]
(which sets the wait tick to 0, so the loop spins) ignores the INT in about 1 of 200 runs on bash 5.3 and 3.2; with
the INT sent to the whole group, 3 of 300 runs. When the date child is made to linger after its output, an INT to
the installer alone is ignored 8 of 8 times. With the tick at its default of 5 seconds the window is a tiny fraction of
each tick.

Proposed fix

Take the time from bash's $SECONDS instead of $(date +%s) in the wait loop (start and deadline from the same
clock), keep command substitutions out of the main shell's wait loop (the heartbeat included), and keep the test above
(both deliveries) as the regression test.

主要语言
Python
星标
0
派生
4
平均合并
5 小时 36 分钟
30 天内合并 PR
509

环境准备

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

Controlvar/controlvar 的其他 Issue

查看 Controlvar/controlvar 的全部 Issue

相似的 Issue

更多 Python Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。