Skip to content

fix(sandbox): OOM-bias session hosts that no cgroup scope bounds - #22683

Closed
unstablebrainiac wants to merge 1 commit into
kirodotdev:mainfrom
unstablebrainiac:fix/oom-bias-session-host-without-cgroup
Closed

unstablebrainiac wants to merge 1 commit into
kirodotdev:mainfrom
unstablebrainiac:fix/oom-bias-session-host-without-cgroup

Conversation

@unstablebrainiac

@unstablebrainiac unstablebrainiac commented Oct 21, 2026 •

Copy link
Copy Markdown
Contributor

Problem / Motivation

Goal: When a Linux host without cgroup v2 runs out of memory, the kernel kills a process in one agent session instead of the gateway.

On such a host nothing bounds an agent's memory, so the kernel OOM killer is the only ceiling left. Every session host runs at oom_score_adj 0, the same as the gateway, and so does everything it runs. The kernel then ranks by size, and the gateway is usually the largest process (#22681).

Why it matters

Losing the gateway ends every chat and subagent on the host at once, and the restarted gateway starts into the same memory shortage. Losing one agent process ends at most the sessions on that agent's runtime.

Not a goal

  • Bounding agent memory on hosts without cgroup v2.
  • Finding or limiting whichever process exhausted the memory.
  • Any change on a host where the cgroup scope wraps the session host, or off Linux.
  • The pod boot probe's 20-second kiro-cli check, which also uses session_host but runs no agent work.

What changed (motivation → approach → change)

When no cgroup scope bounds a session host, it now starts at oom_score_adj 1000, and so does everything it runs. Out of memory, the kernel then kills a process in one agent's tree instead of the gateway.

Session host spawn cgroup scope Profile
ACP launch tail acp/launch.py, for acp/client.py and acp/runtime.py wraps it session_host: NOFILE headroom, no OOM bias, as today
the same launch tail could not wrap it new unbounded_session_host: the same NOFILE headroom plus --oom-bias
pod boot probe agent_sdk/pod_child_probe.py either session_host, unchanged

The launch tail compares the argv before and after cgroup_scope_argv. An unchanged argv means no scope, so no memory.max bounds the tree, and the launch picks the biased profile. The spawn shim then writes oom_score_adj=1000 on the session host, and every process in its tree inherits it. That includes the shell commands the harness runs itself, which the tool profile never reaches. A scoped session host stays unbiased, because its scope already bounds it.

sandbox_rlimit.py defines the profile and maps it to the bias, sandbox.py forwards the name, and docs/architecture/resource-protection.md lists it. A shim-less install keeps the session host's descriptor headroom and drops only the bias.

Backwards compatibility

Compatible: a session host the cgroup scope wraps spawns exactly as before. One it cannot wrap gets today's limits plus the OOM bias, which only changes which process the kernel kills once memory is gone.

Removes nothing: the change adds one profile and picks it in one new case. No path, key, record or public function is taken away.

Tests

  • test/test_acp_launch_goldens.py: the launch picks session_host when the scope wraps the spawn and unbounded_session_host when it does not.
  • test/test_spawn_exec_shim.py:
    • the new profile's shim argv keeps RLIMIT_NOFILE:hard and adds --oom-bias;
    • a shim-less install falls back to the session host's preexec, not the tool's 1024 cap;
    • a real child spawned with the new profile, and a command that child runs, both read oom_score_adj 1000, while session_host passes the parent's own value through. The test skips when the runner is itself at 1000, because both profiles would then read alike. Its waits are bounded, and a probe still running at the bound is killed.
  • test/test_sandbox_refactor_facade.py: sandbox forwards the new name.

Disabling each piece in turn (the bias-map entry, the shim's NOFILE rule for the new profile, the shim-less fallback, and the launch's choice inverted, always biased, or ignored) makes at least one of these tests fail. Each new or changed test passed 20 runs at -n0, 10 runs at -n 4 --dist loadgroup beside its sibling files, and one shuffled run (seed 15928), all from a shell at oom_score_adj 0.

Manual verification

On Linux 5.10 (Amazon Linux 2, cgroup v1, systemd 219), from a shell at oom_score_adj 0:

  • Every live session host's /proc/<pid>/cgroup reads memory:/system.slice/kirocrew.service, the gateway's own cgroup, so cgroup_scope_argv did not wrap them. _compute_cgroup_scope_probe() gives the reason: (False, 'no XDG_RUNTIME_DIR (no systemd user session)'). On this host the launch therefore picks unbounded_session_host every time.
  • Through the real spawn shim, a child and a command it runs read 1000 under unbounded_session_host and 0 under session_host.
  • With --timeout=20, making the probe's grandchild hang fails the real-child test in 7 s with subprocess.TimeoutExpired, and making the probe itself hang fails it in 12 s naming the probe. Neither leaves a process behind.

A local patch that biases every session_host spawn has run on this host since 2026-10-19. No spawn here is scoped, so for agent sessions it does what this PR does. /proc/<pid>/oom_score reads 1336 to 1340 for the largest agent processes and 693 for the gateway, so the kernel's first choice is now an agent process.

Related Issues

Closes #22681

Related: #20489 (the loop-stall restart that ended the incident in #22681 before the kernel chose a victim)

Pattern harvest

Not generalizable: the session host is the only long-lived trusted spawn left unbiased on purpose, and acp/launch.py is its only launch site, so there is no wider class for a rule to catch.

Checklist

  • At most two commits (one is the norm), with a Conventional Commits title (feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)
  • Existing tests pass and new tests added for new functionality
  • Self-review completed; code follows project style guidelines
  • Documentation updated (if applicable)
  • No secrets, credentials, or internal references in the diff

Sensitive change review

  • Head commit: 483d4b6
  • What rule changed: when cgroup_scope_argv cannot wrap a session host, the ACP launch tail spawns it with the new unbounded_session_host profile instead of session_host. That profile is session_host's RLIMIT_NOFILE:hard plus the spawn shim's --oom-bias. A scoped session host keeps session_host.
  • Before -> after (what was allowed/blocked, now): before, an unscoped session host and everything it ran had the gateway's oom_score_adj, 0 on the reporting host. After, they run at 1000. No mount, path mask, seccomp rule, environment scrub or descriptor limit changes, and nothing is newly allowed or blocked.
  • Worst case if wrong (breaks commands / leaks secret / locks user out): the kernel kills an agent process, at worst a session host and the sessions on its runtime, before it would have killed the gateway. No command breaks, no secret moves and no user is locked out, because the bias only raises oom_score_adj, which any unprivileged process may do.
  • Evidence checked (which sandbox result proves it): on Linux 5.10 with cgroup v1, spawned through the real shim, a child and a command it runs read 1000 under unbounded_session_host and 0 under session_host. Every live session host there reads memory:/system.slice/kirocrew.service, so none is scoped, and _compute_cgroup_scope_probe() returns (False, 'no XDG_RUNTIME_DIR (no systemd user session)'). Not run: the OS sandbox backend, because the shell these checks ran in refuses unshare(CLONE_NEWUSER) with EPERM.
  • No regression:
    • Tools still work (which tools/commands were run, result): this PR was built in an agent session on that host whose whole tree runs at 1000 through an equivalent local patch, and shell commands, git, gh, pytest and MCP tools all worked there.
    • Backward compatible (old config / old data / old callers still work, how checked): a scoped session host spawns with session_host exactly as before, pinned by test_only_a_session_host_without_a_cgroup_scope_is_an_oom_target. No config key, path or public name is removed, and a shim-less install keeps session_host_preexec, pinned by test_unbounded_session_host_fallback_keeps_the_session_host_preexec.
    • Existing tests (which suites passed on this head): test/test_spawn_exec_shim.py, test/test_acp_launch_goldens.py and test/test_sandbox_refactor_facade.py, 660 tests, passed on this head. 5,262 related acp_*, spawn_exec* and sandbox* tests passed on the same production change.
  • How to undo: revert the commit. The launch tail then passes session_host for every session host again, and no config or data needs cleaning up. Session hosts already running keep 1000 until they restart.

@github-actions github-actions Bot added fork Pull request from a fork (external contributor) readiness: checking Automated validation is still running labels Oct 21, 2026
@github-actions

github-actions Bot commented Oct 21, 2026 •

Copy link
Copy Markdown
Contributor

UX Review (Opus 5.5, fork) — ⏭️ skipped

Revision 483d4b683f46ac891218f91fc9247a09bb67076e touches no user-facing surface (no changes under website/ or committed screenshots), so the UX review was skipped. Does not block merge.

@github-actions

github-actions Bot commented Oct 21, 2026 •

Copy link
Copy Markdown
Contributor

Design Review (Opus 5.5, fork) — ✅ PASS

Design-level review of 483d4b683f46ac891218f91fc9247a09bb67076e via the fork AI-review pipeline — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS does not.

Design-Verdict: PASS

Nothing blocks: the change only raises oom_score_adj, it is limited to the one launch site where no memory.max bounds the tree, and the shim's bias already does nothing off Linux.

Notes

  • The check that argv is unchanged before and after cgroup_scope_argv is the same test _name_scope (acp/runtime_spawn_launch.py:754) already uses. That makes two places read "was the spawn scoped?" off argv identity. If they ever disagree, the unit name could be reported for a session host that was biased. A small shared helper that answers "was this argv scoped?" would keep both readers on one answer.

[DESIGN-REVIEWED] 483d4b6

Description read

[DESCRIPTION-READ] 19c4fed54cd425eba4226d98462635f5200b2f10553ba21c0fe6c1eaa0b1ccfe

That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with .github/scripts/pr-description-capture.sh over the same inputs; if it differs, the description or that evidence changed after this verdict was formed and any finding drawn from either is unproven.

@github-actions

github-actions Bot commented Oct 21, 2026 •

Copy link
Copy Markdown
Contributor

First Principles Review (Opus 5.5, fork) — ✅ PASS

Premise-level review of 483d4b683f46ac891218f91fc9247a09bb67076e via the fork AI-review pipeline — why this exists and whether the shipped surface is the smallest honest version. Updated in place on each push. A BLOCK verdict blocks PR readiness; PASS does not.

First-Principles-Verdict: PASS

Nothing blocks: this is a point fix for #22681 that adds one profile, picked at the only long-lived session-host launch site.

Notes

What this change ships

Inventory (4 items) — 3 justified

Intent: when memory runs out on a Linux host with no cgroup scope, the kernel should kill an agent process, not the gateway. This is a FIX.

  1. Session hosts with no cgroup scope, and everything they run, now start at oom_score_adj 1000 — inherited (narrows a prior pin)
  2. Without the spawn shim, the new profile keeps the session host's descriptor headroom and drops only the bias — justified
  3. A new profile name, unbounded_session_host, is exported through sandbox and has 1 consumer (acp/launch.py) — justified
  4. The resource-protection doc lists six profiles — justified

[FIRST-PRINCIPLES-REVIEWED] 483d4b6

Description read

[DESCRIPTION-READ] e47377801ef6127f0b8692b763bf37b90bb991cf5138422ce4332fed244bc6bc

That sha256 names the PR description this verdict read. Recompute it with .github/scripts/pr-description-capture.sh; if it differs, the description changed after this verdict was formed and any finding drawn from the description is unproven.

@github-actions

github-actions Bot commented Oct 21, 2026 •

Copy link
Copy Markdown
Contributor

Opus 5.5 Review (fork) — ✅ no blocking findings

Reviewed 783b647222a2e73d54736c12381a7bd3b5885b4d via the fork AI-review pipeline; updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] 783b647

@github-actions

github-actions Bot commented Oct 21, 2026 •

Copy link
Copy Markdown
Contributor

GPT 6.1 Review (fork) — ✅ no blocking findings

Reviewed 483d4b683f46ac891218f91fc9247a09bb67076e via the fork AI-review pipeline; updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] 483d4b6

@bolichen97 bolichen97 left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks good to ship on the code. Two maintainer checks remain before approval.

Head reviewed: 51e7c36. Bots read: Design PASS, First Principles PASS, UX skipped, Opus no blocking, GPT no blocking, readiness checking.

Findings: 0 must-fix, 0 should-fix.

What I verified on this head:

  • cgroup_scope_argv returns the same argv only on its two degrade paths (no cgroup v2 delegation, or no trusted systemd-run). Every wrapped path prepends systemd-run. So argv == unscoped_argv is a correct "no scope" test.
  • request.scope_argv runs after the choice and only adds --unit to an argv that is already wrapped. It cannot turn an unscoped spawn into a scoped one, so the profile choice cannot be stale.
  • The new profile reuses the session host NOFILE spec and preexec, and only adds the bias entry in _PROFILE_OOM_BIAS. Off Linux the shim skips the bias, and on Windows the shim prefix is empty.
  • The bias only raises oom_score_adj. It widens no fence and lowers no limit.
  • The golden test goes red if the launch choice is inverted or ignored. The real-child test checks the grandchild, which is the path the harness shell commands take, and its waits are bounded.

Remaining maintainer checks:

  1. PR Readiness is still pending on this head. Backend test shards are still running.
  2. The diff changes sandbox.py and sandbox_rlimit.py. A change under the sandbox modules needs a maintainer review, and the Sensitive Change Review check is red on this head. Please fill the sensitive change template in the description so that check can pass.

Not repeated: the Design Review note that a killed session host can take every session on that runtime, not one session. The doc row and the sandbox_rlimit.py comment say "one session lost". Consider "one agent runtime lost" in a later push, but this does not block.

@unstablebrainiac

Copy link
Copy Markdown
Contributor Author

Sensitive Change Review cannot pass on this PR, with or without the template. On main, evaluate() in .github/scripts/sensitive_change_review.py returns its fork refusal before it reads any template, and the records it reads are the author's PR comments and approving reviews, not the description. So it stays red on every head of a fork PR that touches sandbox*.py, as it does on #22636 and #20822. The check's own message gives the path: a maintainer takes the change over on a branch of this repository with Co-authored-by: and Supersedes #22683, the way #21764 did for #19442.

The filled template is in the description now, at the end, and passes the checker's own template_problems(). Whoever takes this over can paste it into their author comment with their own head SHA. Its Evidence line says what was not run: the OS sandbox backend, because the shell these checks ran in refuses unshare(CLONE_NEWUSER).

"One session lost": agreed. A session host can serve several sessions, a parent and the subagents created on its runtime, so killing it ends all of them. The sandbox_rlimit.py comment now says "at worst that runtime and the sessions it serves". The doc row never claimed one session, it only says the kernel takes the largest process in some agent tree, so it stays. The same claim in the description and in #22681 is corrected too. The comment fix goes out with a rebase once the backend shards on this head finish.

@unstablebrainiac
unstablebrainiac force-pushed the fix/oom-bias-session-host-without-cgroup branch from 51e7c36 to c83441a Compare October 21, 2026 08:03
@github-actions github-actions Bot added readiness: action required A blocking check or review needs attention and removed readiness: checking Automated validation is still running labels Oct 21, 2026
@unstablebrainiac
unstablebrainiac force-pushed the fix/oom-bias-session-host-without-cgroup branch from c83441a to 865c993 Compare October 21, 2026 11:00
@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: passed Eligible automated validation passed for the current revision and removed readiness: action required A blocking check or review needs attention readiness: checking Automated validation is still running labels Oct 21, 2026
@iamwhatever

Copy link
Copy Markdown
Contributor

Hold for maintainer review at 865c993.

This changes sandbox code and which process the kernel kills under memory pressure. An unscoped session host and everything it runs now start at oom_score_adj 1000 (launch.py#L1051, sandbox_rlimit.py#L326). Sandbox and process-killing changes are reserved for a human reviewer. I found no defect: a scoped host and the pod probe keep session_host, and the shim-less fallback keeps the NOFILE headroom. The Sensitive Change Review check is red on this fork head. It clears when a maintainer takes the change over on a same-repo branch and posts the template there, then gives the human review.

— crew-pr-reviewer (r117)

@iamwhatever
iamwhatever enabled auto-merge (squash) October 21, 2026 21:14
@github-actions github-actions Bot added the merge conflict Branch has merge conflicts with its base — author must resolve before merge label Oct 21, 2026
@unstablebrainiac
unstablebrainiac force-pushed the fix/oom-bias-session-host-without-cgroup branch from 865c993 to f453df0 Compare October 21, 2026 21:44
@github-actions github-actions Bot added readiness: checking Automated validation is still running merge conflict Branch has merge conflicts with its base — author must resolve before merge and removed readiness: passed Eligible automated validation passed for the current revision merge conflict Branch has merge conflicts with its base — author must resolve before merge labels Oct 21, 2026
@unstablebrainiac
unstablebrainiac force-pushed the fix/oom-bias-session-host-without-cgroup branch from f453df0 to 1fe83c4 Compare October 21, 2026 23:15
@github-actions github-actions Bot removed the merge conflict Branch has merge conflicts with its base — author must resolve before merge label Oct 21, 2026
@github-actions github-actions Bot added the readiness: action required A blocking check or review needs attention label Nov 6, 2026
@unstablebrainiac
unstablebrainiac force-pushed the fix/oom-bias-session-host-without-cgroup branch from 1212174 to 783b647 Compare November 6, 2026 01:47
@unstablebrainiac

Copy link
Copy Markdown
Contributor Author

@bolichen97 Dropped the runtime skip in test_unbounded_session_host_biases_what_its_harness_runs in 783b647222a2; the comment now says a runner already at 1000 still checks both expected scores but cannot tell the two profiles apart. Still one commit, nothing else changed.

The test passes from a shell at oom_score_adj 0 and from one at 1000 (no skip), the three changed test files pass (725), and test_flake_pattern_ratchet.py passes.

@github-actions github-actions Bot added readiness: checking Automated validation is still running and removed readiness: action required A blocking check or review needs attention labels Nov 6, 2026
@bolichen97

Copy link
Copy Markdown
Contributor

Thanks, 783b647 fixes my point: the runtime pytest.skip is gone (only the Linux-only skipif remains), and GPT 6.1 now reports no blocking findings.

Could you update this branch onto main? Backend Tests (3.12, 6) fails only on test_dashboard_server_composition_contract.py::test_the_lazy_imports_stay_inside_the_functions_that_need_them (131 vs 130). That pin was off on main itself and was fixed by #30984 (5dcda3b), which merged after this branch's base (8c2e758). Coverage Gate follows from that shard. Once it is green I will approve.

@github-actions github-actions Bot added readiness: action required A blocking check or review needs attention and removed readiness: checking Automated validation is still running labels Nov 6, 2026
@unstablebrainiac
unstablebrainiac force-pushed the fix/oom-bias-session-host-without-cgroup branch from 783b647 to ea352fe Compare November 6, 2026 07:48
@unstablebrainiac

Copy link
Copy Markdown
Contributor Author

@bolichen97 Rebased onto main f5340cc65851, which includes #30984, as ea352fe12e68. The patch is identical (git range-diff reports =). Locally the three changed test files and test_dashboard_server_composition_contract.py pass (774), as does test_flake_pattern_ratchet.py.

@github-actions github-actions Bot added readiness: checking Automated validation is still running and removed readiness: action required A blocking check or review needs attention labels Nov 6, 2026
@bolichen97

Copy link
Copy Markdown
Contributor

@unstablebrainiac thanks for the update onto main. The patch is unchanged, and my earlier point stays fixed.

One thing blocks on ea352fe. Sensitive Change Review reports "Sensitive path changed without a reasoning template for this head." The rebase gave the PR a new head sha, so the lane needs the ## Sensitive change review form again, with Head commit: ea352fe12e6814e5fd26fe7f73941be611ba9b22. Could you post it? The content can match your last form, since the patch did not change.

Opus 5.5, Internal Content Scan and Security Scope Review are also red. They failed when the CI app hit its API rate limit at 07:56 UTC, not on your change. I'll approve once the lanes are green.

@unstablebrainiac

Copy link
Copy Markdown
Contributor Author

Sensitive change review

  • Head commit: ea352fe
  • What rule changed: when cgroup_scope_argv cannot wrap a session host, the ACP launch tail spawns it with the new unbounded_session_host profile instead of session_host. That profile is session_host's RLIMIT_NOFILE:hard plus the spawn shim's --oom-bias. A scoped session host keeps session_host.
  • Before -> after (what was allowed/blocked, now): before, an unscoped session host and everything it ran had the gateway's oom_score_adj, 0 on the reporting host. After, they run at 1000. No mount, path mask, seccomp rule, environment scrub or descriptor limit changes, and nothing is newly allowed or blocked.
  • Worst case if wrong (breaks commands / leaks secret / locks user out): the kernel kills an agent process, at worst a session host and the sessions on its runtime, before it would have killed the gateway. No command breaks, no secret moves and no user is locked out, because the bias only raises oom_score_adj, which any unprivileged process may do.
  • Evidence checked (which sandbox result proves it): on Linux 5.10 with cgroup v1, spawned through the real shim, a child and a command it runs read 1000 under unbounded_session_host and 0 under session_host. Every live session host there reads memory:/system.slice/kirocrew.service, so none is scoped, and _compute_cgroup_scope_probe() returns (False, 'no XDG_RUNTIME_DIR (no systemd user session)'). Not run: the OS sandbox backend, because the shell these checks ran in refuses unshare(CLONE_NEWUSER) with EPERM.
  • No regression:
    • Tools still work (which tools/commands were run, result): this PR was built in an agent session on that host whose whole tree runs at 1000 through an equivalent local patch, and shell commands, git, gh, pytest and MCP tools all worked there.
    • Backward compatible (old config / old data / old callers still work, how checked): a scoped session host spawns with session_host exactly as before, pinned by test_only_a_session_host_without_a_cgroup_scope_is_an_oom_target. No config key, path or public name is removed, and a shim-less install keeps session_host_preexec, pinned by test_unbounded_session_host_fallback_keeps_the_session_host_preexec.
    • Existing tests (which suites passed on this head): test/test_spawn_exec_shim.py, test/test_acp_launch_goldens.py and test/test_sandbox_refactor_facade.py, with test/test_dashboard_server_composition_contract.py, 774 tests, passed on this head, and test/test_flake_pattern_ratchet.py passed. 5,262 related acp_*, spawn_exec* and sandbox* tests passed on the same production change.
  • How to undo: revert the commit. The launch tail then passes session_host for every session host again, and no config or data needs cleaning up. Session hosts already running keep 1000 until they restart.

@github-actions github-actions Bot added the merge conflict Branch has merge conflicts with its base — author must resolve before merge label Nov 11, 2026
…ateway

On a host without cgroup v2 delegation the agent tree has no memory.max,
so the kernel's host-wide OOM killer is the only memory ceiling. Session
hosts were spawned with oom_score_adj 0, tying the gateway, so the kernel
picked by size and killed the gateway, ending every session.

The ACP launch tail now picks unbounded_session_host when
cgroup_scope_argv did not wrap the spawn. It keeps session_host's NOFILE
headroom and adds the OOM bias, which every process in the agent tree
inherits, including shell commands the harness runs itself.
@unstablebrainiac
unstablebrainiac force-pushed the fix/oom-bias-session-host-without-cgroup branch from ea352fe to 483d4b6 Compare November 12, 2026 21:43
@KiroCrewCaptain

Copy link
Copy Markdown
Collaborator

Code change approver recommendation: Approve

Why: the decider estimates a 10% chance that this change introduces a defect, within the 12% limit for approving on its own, and every check passes.

Adds an unbounded_session_host profile: the same NOFILE headroom as session_host plus --oom-bias. The ACP launch tail picks it when cgroup_scope_argv leaves the argv unchanged, meaning no scope was added. The detection is sound: cgroup_scope_argv returns the argv unchanged whenever it can't scope, and the later request.scope_argv only names a scope that already exists. Every sibling branch (_rlimit_spec, _preexec_for_profile, _PROFILE_OOM_BIAS) handles the new profile. Other session_host callers (the pod probe and the Coder CLI) are unchanged, and the bias does nothing off Linux. Shared daemons such as gatewayd are started by the gateway, not from inside agent trees, so they don't inherit the bias. Tests cover the launch choice, the shim argv, the shim-less fallback, and a real child and grandchild reading oom_score_adj. Scoped hosts behave exactly as before.

An automated recommendation, not a review: a maintainer still decides. It approves on its own only when the estimated chance of a defect is at most 12% and the change has at most 400 changed lines. Decider v3c (217fe607) · commit 483d4b683f · 130 changed lines. If this recommendation is wrong, reply here: replies are how the approver is evaluated.

@github-actions github-actions Bot added readiness: action required A blocking check or review needs attention and removed merge conflict Branch has merge conflicts with its base — author must resolve before merge readiness: checking Automated validation is still running labels Nov 12, 2026
@iamwhatever

Copy link
Copy Markdown
Contributor

I'm picking this up to address the requested changes and the red CI. Per the maintainer review, a sandbox change from a fork cannot pass Sensitive Change Review, so I'll carry @unstablebrainiac's commit on a branch of this repository with authorship kept, rebase it onto current main, run the real Linux sandbox check on an unscoped host, and link the new PR here. Thanks @unstablebrainiac for the fix and the careful iterations.

@iamwhatever

Copy link
Copy Markdown
Contributor

Carried on a repository branch as #39189: the same commit, authored by @unstablebrainiac, rebased onto current main. That PR also has a real-host check on Amazon Linux 2 (cgroup v1, no user bus): a live session's host, harness command, grandchild, MCP servers and a subagent all read oom_score_adj 1000 while the gateway stays at 0. Review and CI continue there.

@iamwhatever

Copy link
Copy Markdown
Contributor

Landed as #39189 (merge commit f47c07b), carrying @unstablebrainiac's commit with authorship kept; #22681 is closed. Closing this one as superseded. Thanks for the fix.

auto-merge was automatically disabled November 18, 2026 20:58

Pull request was closed

@github-actions github-actions Bot removed the readiness: action required A blocking check or review needs attention label Nov 18, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

fork Pull request from a fork (external contributor)

Projects

None yet

Development

Successfully merging this pull request may close these issues.

On hosts without cgroup v2, the kernel OOM killer picks the gateway over agent session trees

4 participants