test: count a group of zombies as ended, and kill create once it wrote the manifest by ThomasK33 · Pull Request #314 · coder/agent-tty

ThomasK33 · 2026-11-04T00:48:40Z

Why

"destroy ends a child of the program whose main thread exited while another thread runs" needed 2 retries on main and on #311, on this shared host at load averages of 60 to 80. The #311 advisor also found that the early-destroy test still needed the host to start the program before create gave up after about 5 s. Follow-up of #311.

Cause

groupEnds in test/helpers.ts polled kill(-pgid, 0) for 2 s. That call also finds a zombie. A program's background process is an orphan once its parent shell dies, so init reaps it, not the host. On a loaded host, init reaped it after the 2 s deadline. The test then failed although destroy had already ended every process of the group.

I reproduced this with a child subreaper that reaps only every 10 s (slow-reaper.py below). Main fails groupEnds in 2 or 3 of the 5 selected tests per run:

"destroy ends a child of the program whose main thread exited while another thread runs"
"destroy ends a background process that outlives the program"
"destroy ends a background process left by a program that exited"

This branch passes 5 of 5.

Repro wrapper (not committed)

# REAP_DELAY=10 python3 slow-reaper.py npx vitest run --retry=0 test/integration/lifecycle.test.ts -t "whose main thread exited|destroy ends a background process"
import ctypes, os, sys, time
ctypes.CDLL(None).prctl(36, 1, 0, 0, 0)  # PR_SET_CHILD_SUBREAPER
delay = float(os.environ.get('REAP_DELAY', '3'))
child = os.fork()
if child == 0:
    os.execvp(sys.argv[1], sys.argv[1:])
code = None
while True:
    time.sleep(delay)
    try:
        while True:
            pid, status = os.waitpid(-1, os.WNOHANG)
            if pid == 0:
                break
            if pid == child:
                code = os.waitstatus_to_exitcode(status)
    except ChildProcessError:
        sys.exit(code if code is not None else 1)

What changes

Only test helpers and one test change.

On Linux, groupEnds now reads the threads of each group member in /proc. A group whose threads are all zombies counts as ended. A live thread, such as the running thread of zombie-leader.py whose main thread exited, still counts as running. Two scans must agree, because a process can fork and end between the /proc listing and its own read. In some cases /proc cannot show the whole group: the listing fails, a read is refused, /proc has hidepid=2, 4, invisible or ptraceable, or mountinfo is unreadable. Then the kill(-pgid, 0) probe decides alone, as before. The probe also decides alone when a scan finds the group id but shows no member at all, as a /proc of another pid namespace does. Off Linux it keeps kill(-pgid, 0), because launchd reaps orphans at once. The deadline is now 5 s, which costs nothing when the group ends sooner.
The early-destroy test kills create as soon as the manifest exists. The host reads an existing manifest also after its creator died (waitForManifestFromCreate, src/host/hostMain.ts). So the test no longer needs the host to start the program before create gives up. Its envelope wait is now 30 s instead of 15 s.
I checked the advisor's second point with a throwaway test. That test held the host's start-up write and failed on purpose. Afterwards, cleanupHome had killed the held host, because it SIGKILLs the hostPid that the manifest names. No change is needed.

Validation

tsc --noEmit, oxlint and oxfmt --check pass.
With the 10 s reaper, main failed and this branch passed 5 of 5 (results above).
With a delay of 1.5 s and 2.5 s at the start of the host and of each destroy (the test: keep the lifecycle tests stable on a slow runner and cut fixed waits #311 preload), the early-destroy test passes.
A throwaway test checked groupEnds against a live zombie-leader.py. It still reads the group as running after 1.5 s, and it reads the group as ended right after a SIGKILL.

Generated with xum • Model: anthropic:claude-opus-5-5 • Thinking: xhigh

…e the manifest - groupEnds waited for init to reap the program's killed orphans: kill(-pgid, 0) finds a zombie. On a loaded host init reaped them after the 2 s deadline, so "destroy ends a child of the program whose main thread exited" and the tests of a background process failed although destroy had ended every process. On Linux groupEnds now reads the threads in /proc and counts a group whose threads are all zombies as ended. - The early-destroy test killed create once the program ran, so it still needed the host to start the program before create gave up. It now kills create once the manifest exists: the host reads an existing manifest also after its creator died. Its envelope wait is 30 s.

ThomasK33 · 2026-11-04T00:48:52Z

@codex review

ThomasK33 · 2026-11-04T00:48:55Z

@codex security review

chatgpt-codex-connector · 2026-11-04T00:49:03Z

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-11-04T01:33:12.152388Z	`524b2ba`	Manual request
🔒 Security Review	✅ Completed 2026-11-04T01:33:58.038975Z	`524b2ba`	Manual request

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

chatgpt-codex-connector · 2026-11-04T00:51:25Z

🛡️ Codex Security Review

Security review completed. No security issues were found in this pull request.

Reviewed commit: f9f11d1180

View security finding report

Only the user who started this review can view the report in Codex.

ℹ️ About Codex security reviews in GitHub

This is an experimental Codex feature. Security reviews are triggered when:

You comment "@codex security review"
A regular code review gets triggered (for example, "@codex review" or when a PR is opened), and you’re opted in so security review runs alongside code review

Once complete, Codex will leave suggestions, or a comment if no findings are found.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: f9f11d1180

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

…proc hides processes Codex on #314: a process can fork and end between the /proc listing and its own read, and a /proc that is missing, refuses a read or hides other users cannot show the whole group.

ThomasK33 · 2026-11-04T01:06:59Z

@codex review

ThomasK33 · 2026-11-04T01:07:02Z

@codex security review

chatgpt-codex-connector · 2026-11-04T01:09:19Z

🛡️ Codex Security Review

Security review completed. No security issues were found in this pull request.

Reviewed commit: c836c62308

View security finding report

Only the user who started this review can view the report in Codex.

ℹ️ About Codex security reviews in GitHub

This is an experimental Codex feature. Security reviews are triggered when:

You comment "@codex security review"
A regular code review gets triggered (for example, "@codex review" or when a PR is opened), and you’re opted in so security review runs alongside code review

Once complete, Codex will leave suggestions, or a comment if no findings are found.

chatgpt-codex-connector · 2026-11-04T01:09:50Z

Codex Review: Didn't find any major issues. 🚀

Reviewed commit: c836c62308

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

… only ENOENT pass the sessions read #314 advisor: a /proc of another pid namespace shows no member of the group, so the probe decides there. Only a missing sessions folder means that create has not written yet.

ThomasK33 · 2026-11-04T01:30:57Z

@codex review

ThomasK33 · 2026-11-04T01:31:01Z

@codex security review

chatgpt-codex-connector · 2026-11-04T01:33:10Z

Codex Review: Didn't find any major issues. Already looking forward to the next diff.

Reviewed commit: 524b2ba870

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-11-04T01:33:56Z

🛡️ Codex Security Review

Security review completed. No security issues were found in this pull request.

Reviewed commit: 524b2ba870

View security finding report

Only the user who started this review can view the report in Codex.

ℹ️ About Codex security reviews in GitHub

This is an experimental Codex feature. Security reviews are triggered when:

You comment "@codex security review"
A regular code review gets triggered (for example, "@codex review" or when a PR is opened), and you’re opted in so security review runs alongside code review

Once complete, Codex will leave suggestions, or a comment if no findings are found.

chatgpt-codex-connector Bot reviewed Nov 4, 2026

View reviewed changes

Comment thread test/helpers.ts Resolved

Comment thread test/helpers.ts Resolved

test: scan /proc twice in groupEnds, and let the probe decide where /…

c836c62

…proc hides processes Codex on #314: a process can fork and end between the /proc listing and its own read, and a /proc that is missing, refuses a read or hides other users cannot show the whole group.

test: read a scan that shows no group member as inconclusive, and let…

524b2ba

… only ENOENT pass the sessions read #314 advisor: a /proc of another pid namespace shows no member of the group, so the probe decides there. Only a missing sessions folder means that create has not written yet.

ThomasK33 added this pull request to the merge queue Nov 4, 2026

Merged via the queue into main with commit 8668515 Nov 4, 2026
18 checks passed

ThomasK33 deleted the test/zombie-leader-load branch November 4, 2026 02:01

Conversation

ThomasK33 commented Nov 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Why

Cause

What changes

Validation

Uh oh!

ThomasK33 commented Nov 4, 2026

Uh oh!

ThomasK33 commented Nov 4, 2026

Uh oh!

chatgpt-codex-connector Bot commented Nov 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codex Review Summary

Uh oh!

chatgpt-codex-connector Bot commented Nov 4, 2026

🛡️ Codex Security Review

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

Uh oh!

ThomasK33 commented Nov 4, 2026

Uh oh!

ThomasK33 commented Nov 4, 2026

Uh oh!

chatgpt-codex-connector Bot commented Nov 4, 2026

🛡️ Codex Security Review

Uh oh!

chatgpt-codex-connector Bot commented Nov 4, 2026

Uh oh!

ThomasK33 commented Nov 4, 2026

Uh oh!

ThomasK33 commented Nov 4, 2026

Uh oh!

chatgpt-codex-connector Bot commented Nov 4, 2026

Uh oh!

chatgpt-codex-connector Bot commented Nov 4, 2026

🛡️ Codex Security Review

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

ThomasK33 commented Nov 4, 2026 •

edited

Loading

chatgpt-codex-connector Bot commented Nov 4, 2026 •

edited

Loading