Enforce Job Object cleanup and resource limits in AppContainer runner by npollock1 · Pull Request #317 · microsoft/mxc

npollock1 · 2026-05-14T19:56:10Z

📖 Description

Adds Windows Job Object enforcement to the AppContainer runner so descendant processes are guaranteed to be terminated when wxc-exec finishes (success, timeout, or failure), and introduces an optional process.resourceLimits section for memory, process-count, CPU-rate, and child-process restrictions.

What changed

Schema: New process.resourceLimits object on schemas/dev/mxc-config.schema.0.6.0-dev.json with memoryMb, maxProcesses, cpuRatePercent (0..=100), and allowChildProcesses (default true). All optional; 0 for any cap means no limit.
Rust models / parser: ResourceLimits struct on CodexRequest; cpuRatePercent > 100 is rejected at parse time. Permissive defaults preserve behavior for omitted sections.
AppContainer runner: Always creates a Job Object with JOB_OBJECT_LIMIT_KILL_ON_JOB_CLOSE — even when no caps are configured — so descendants cannot outlive the sandbox. Optional caps applied via JOBOBJECT_EXTENDED_LIMIT_INFORMATION and JOBOBJECT_CPU_RATE_CONTROL_INFORMATION. allowChildProcesses: false adds PROC_THREAD_ATTRIBUTE_CHILD_PROCESS_POLICY. The process is spawned CREATE_SUSPENDED, assigned to the job, then resumed (no race window where children can escape). Timeout cleanup uses TerminateJobObject to kill the entire tree.
SDK: ResourceLimits in SandboxPolicy with TypeScript-side validation, propagated into ContainerConfig.process.resourceLimits.
Docs: docs/schema.md and sdk/README.md updated.

Notable behavior change

GUI launcher-style apps (e.g. modern Windows 11 Notepad) that rely on a launcher → COM activation → separate UI process pattern will have their UI killed when wxc-exec exits, because the activated UI process becomes a member of the job. This is intentional for the one-shot AppContainer model — long-lived/interactive use cases should use the state-aware sandbox lifecycle.

Known gap (out of scope for this PR)

The BaseContainer FlatBuffer (external/windows-sdk/BaseContainerSpecification.fbs) does not currently expose resource-limit fields. The same Job Object approach can be applied externally to the process handle returned by Experimental_CreateProcessInSandbox in a follow-up; nested job objects are supported on Windows 8+. Documented in docs/schema.md.

🔗 References

Closes #316

🔍 Validation

cargo build --workspace --target x86_64-pc-windows-msvc — clean
cargo test --workspace --target x86_64-pc-windows-msvc --lib — 344/344 wxc_common tests pass, including:
- 5 new parser tests for resourceLimits (omitted, full, partial, zero, out-of-range)
- 3 new Job Object tests in appcontainer_runner: job_limit_info_uses_request_resource_limits, job_object_kills_orphans_on_drop, job_object_terminates_tree
cargo fmt --all -- --check — clean
cargo clippy --workspace --all-targets -- -D warnings — clean
npm run build && npm test (sdk) — 137/137 tests pass, including new createConfigFromPolicy > resourceLimits coverage (omitted, propagated, invalid CPU, negative caps, non-integers, allowChildProcesses: false)

✅ Checklist

Signed the Contributor License Agreement
Linked to an issue
Updated documentation (if applicable)
Updated Copilot instructions (if build, architecture, or conventions changed) — N/A, no build/convention changes

📋 Issue Type

Bug fix
Feature
Task

Microsoft Reviewers: Open in CodeFlow

Introduces process.resourceLimits with memoryMb, maxProcesses, cpuRatePercent, and allowChildProcesses fields. Adds ResourceLimits struct to CodexRequest with permissive defaults (no caps, child processes allowed). cpuRatePercent values >100 are rejected at parse time. Includes 5 new unit tests covering omitted, full, partial, zero, and out-of-range inputs. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Create an AppContainer job object for every run so kill-on-job-close cleans up descendant processes. Apply configured memory, process, CPU, and child-process restrictions, assign the suspended process before resuming, and terminate the job on timeout. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

Adds Windows Job Object enforcement to the AppContainer runner so descendant processes are always cleaned up, and introduces an optional process.resourceLimits configuration section (memory, process count, CPU rate, child-process policy) across the schema, Rust parser/models, AppContainer runner, and TypeScript SDK.

Changes:

New ResourceLimits model + parser validation in wxc_common, propagated into CodexRequest.
AppContainer runner always creates a Job Object with KILL_ON_JOB_CLOSE, optionally applies memory/process/CPU caps, uses CREATE_SUSPENDED + assign + resume to avoid escape, and uses TerminateJobObject for timeouts.
SDK exposes ResourceLimits on SandboxPolicy with validation, schema/docs updated.

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
src/wxc_common/src/models.rs	Adds `ResourceLimits` struct and field on `CodexRequest`.
src/wxc_common/src/config_parser.rs	Parses and validates `process.resourceLimits` JSON.
src/wxc_common/src/appcontainer_runner.rs	Creates/assigns Job Object, adds child-process attribute, switches timeout to `TerminateJobObject`, adds tests.
sdk/src/types.ts	Adds `ResourceLimits` interface and `SandboxPolicy.resourceLimits`.
sdk/src/sandbox.ts	Validates and propagates `resourceLimits` into `ContainerConfig`.
sdk/src/index.ts	Re-exports `ResourceLimits`.
sdk/tests/unit/sandbox.test.ts	Unit tests for `resourceLimits` validation/propagation.
schemas/dev/mxc-config.schema.0.6.0-dev.json	Adds `resourceLimits` JSON schema.
docs/schema.md	Documents resource limits and backend support matrix.
sdk/README.md	Documents `resourceLimits` field.

+            spawn_test_process("cmd.exe /c ping -n 30 127.0.0.1", PROCESS_CREATION_FLAGS(0));
+
+        unsafe { AssignProcessToJobObject(job.get(), child.process.get()) }.expect("assign");


+        let child_process_policy = PROCESS_CREATION_CHILD_PROCESS_RESTRICTED;
+        if request.resource_limits.child_processes_blocked() {


        process: {
            commandLine: '',
            timeout: policy.timeoutMs ?? 0,
+            ...(policy.resourceLimits ? { resourceLimits: policy.resourceLimits } : {}),


alexsnitkovskiy · 2026-05-26T19:20:44Z

As long as this is only for NON-Tesera scenarios and as long as it does not interfere with the Tesera's future resourcing policy this would be OK. So it should be used perhaps down-level and so on...similar or integrated with work Gudge is doing.

npollock1 · 2026-05-26T21:11:00Z

I'm not attached to this implementation. I just wanted to see if it would actually address the referenced bug and I figured it couldn't hurt to share.

kanismohammed · 2026-05-30T00:48:27Z

Cached the Pull Request and closing it to clean up the repo for Public release

npollock1 and others added 6 commits May 14, 2026 10:50

Add SDK resource limit policy support

0f565b8

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Merge AppContainer job object enforcement

5dccb77

Merge SDK resource limit support

4c33975

Document process.resourceLimits and Job Object cleanup behavior

120d2fd

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot AI review requested due to automatic review settings May 14, 2026 19:56

Copilot AI reviewed May 14, 2026

View reviewed changes

Copilot started reviewing on behalf of npollock1 May 14, 2026 20:15 View session

kanismohammed requested a review from a team May 26, 2026 19:10

kanismohammed closed this May 30, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Enforce Job Object cleanup and resource limits in AppContainer runner#317

Enforce Job Object cleanup and resource limits in AppContainer runner#317
npollock1 wants to merge 6 commits into
microsoft:mainfrom
npollock1:feat/job-object-limits

npollock1 commented May 14, 2026 •

edited by microsoft-github-policy-service Bot

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

alexsnitkovskiy commented May 26, 2026

Uh oh!

npollock1 commented May 26, 2026

Uh oh!

kanismohammed commented May 30, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

		spawn_test_process("cmd.exe /c ping -n 30 127.0.0.1", PROCESS_CREATION_FLAGS(0));

		unsafe { AssignProcessToJobObject(job.get(), child.process.get()) }.expect("assign");

		let child_process_policy = PROCESS_CREATION_CHILD_PROCESS_RESTRICTED;
		if request.resource_limits.child_processes_blocked() {

Uh oh!

Conversation

npollock1 commented May 14, 2026 • edited by microsoft-github-policy-service Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

📖 Description

What changed

Notable behavior change

Known gap (out of scope for this PR)

🔗 References

🔍 Validation

✅ Checklist

📋 Issue Type

Microsoft Reviewers: Open in CodeFlow

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

alexsnitkovskiy commented May 26, 2026

Uh oh!

npollock1 commented May 26, 2026

Uh oh!

kanismohammed commented May 30, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

npollock1 commented May 14, 2026 •

edited by microsoft-github-policy-service Bot

Loading