src: avoid draining platform tasks at FreeEnvironment by legendecas · Pull Request #51290 · nodejs/node

legendecas · 2023-12-26T17:48:28Z

At the point of FreeEnvironment and onwards, no JavaScript execution
associated with the Environment should be triggered.

Avoid draining platform tasks that can trigger JavaScript execution in
FreeEnvironment. The holder of node::Environment should immediately
call node::MultiIsolatePlatform::UnregisterIsolate and
v8::Isolate::Dispose to cancel pending foreground tasks and join
concurrent tasks after the environment was freed.

NodePlatform can properly handle the case in RunForegroundTask when
an Isolate out-lives its associated node::Environment.

Fixes: #47748
Fixes: #49344

bcoe · 2023-12-27T16:16:48Z

Why is Isolate::DisallowJavascriptExecutionScope only called when DEBUG is set?

Disallowing JavaScript execution in this function is primarily a rule to avoid footguns, i.e. Environment has been freed and most Node.js APIs would not work. However, it is still valid to evaluate JavaScript as the Isolate has not been freed.

So, to enforce the rule, the Isolate::DisallowJavascriptExecutionScope is opened and will strictly crash the process if JavaScript is invoked in this scope. This would allow us to identify possible problems in debug build. While in the release build, it would be better to be more lenient and let the JavaScript run at the best effort.

I've updated the patch to include a comment about this.

nodejs-github-bot · 2023-12-27T18:27:58Z

CI: https://ci.nodejs.org/job/node-test-pull-request/56561/

At the point of `FreeEnvironment` and onwards, no JavaScript execution associated with the Environment should be triggered. Avoid draining platform tasks that can trigger JavaScript execution in `FreeEnvironment`. The holder of `node::Environment` should immediately call `node::MultiIsolatePlatform::UnregisterIsolate` and `v8::Isolate::Dispose` to cancel pending foreground tasks and join concurrent tasks after the environment was freed. `NodePlatform` can properly handle the case in `RunForegroundTask` when an Isolate out-lives its associated `node::Environment`.

bcoe

LGTM, I'm excited to see if this addresses the deadlock with coverage.

nodejs-github-bot · 2023-12-28T16:00:29Z

CI: https://ci.nodejs.org/job/node-test-pull-request/56569/

mcollina

lgtm

nodejs-github-bot · 2024-01-08T14:39:32Z

Landed in 5db35b4

hhugo · 2024-01-08T15:13:48Z

Thanks. Can we know which version version of node will first include this fix ?

paulrutter · 2024-01-08T21:30:13Z

Will this be ported to the node 20.x branch? Thanks!

mcollina · 2024-01-09T06:07:53Z

If there are no unplanned side effects, yes!

At the point of `FreeEnvironment` and onwards, no JavaScript execution associated with the Environment should be triggered. Avoid draining platform tasks that can trigger JavaScript execution in `FreeEnvironment`. The holder of `node::Environment` should immediately call `node::MultiIsolatePlatform::UnregisterIsolate` and `v8::Isolate::Dispose` to cancel pending foreground tasks and join concurrent tasks after the environment was freed. `NodePlatform` can properly handle the case in `RunForegroundTask` when an Isolate out-lives its associated `node::Environment`. PR-URL: #51290 Fixes: #47748 Fixes: #49344 Reviewed-By: Vinícius Lourenço Claro Cardoso <contact@viniciusl.com.br> Reviewed-By: Matteo Collina <matteo.collina@gmail.com>

At the point of `FreeEnvironment` and onwards, no JavaScript execution associated with the Environment should be triggered. Avoid draining platform tasks that can trigger JavaScript execution in `FreeEnvironment`. The holder of `node::Environment` should immediately call `node::MultiIsolatePlatform::UnregisterIsolate` and `v8::Isolate::Dispose` to cancel pending foreground tasks and join concurrent tasks after the environment was freed. `NodePlatform` can properly handle the case in `RunForegroundTask` when an Isolate out-lives its associated `node::Environment`. PR-URL: nodejs#51290 Fixes: nodejs#47748 Fixes: nodejs#49344 Reviewed-By: Vinícius Lourenço Claro Cardoso <contact@viniciusl.com.br> Reviewed-By: Matteo Collina <matteo.collina@gmail.com>

At the point of `FreeEnvironment` and onwards, no JavaScript execution associated with the Environment should be triggered. Avoid draining platform tasks that can trigger JavaScript execution in `FreeEnvironment`. The holder of `node::Environment` should immediately call `node::MultiIsolatePlatform::UnregisterIsolate` and `v8::Isolate::Dispose` to cancel pending foreground tasks and join concurrent tasks after the environment was freed. `NodePlatform` can properly handle the case in `RunForegroundTask` when an Isolate out-lives its associated `node::Environment`. PR-URL: #51290 Fixes: #47748 Fixes: #49344 Reviewed-By: Vinícius Lourenço Claro Cardoso <contact@viniciusl.com.br> Reviewed-By: Matteo Collina <matteo.collina@gmail.com>

nodejs/node#51290

* chore: bump node in DEPS to v20.12.0 * chore: update build_add_gn_build_files.patch * chore: update patches * chore: bump node in DEPS to v20.12.1 * chore: update patches * build: encode non-ASCII Latin1 characters as one byte in JS2C nodejs/node#51605 * crypto: use EVP_MD_fetch and cache EVP_MD for hashes nodejs/node#51034 * chore: update filenames.json * chore: bump node in DEPS to v20.12.2 * chore: update patches * src: support configurable snapshot nodejs/node#50453 * test: remove test-domain-error-types flaky designation nodejs/node#51717 * src: avoid draining platform tasks at FreeEnvironment nodejs/node#51290 * chore: fix accidentally deleted v8 dep * lib: define FormData and fetch etc. in the built-in snapshot nodejs/node#51598 * chore: rebase on main * chore: remove stray log --------- Co-authored-by: electron-roller[bot] <84116207+electron-roller[bot]@users.noreply.github.com> Co-authored-by: Cheng <zcbenz@gmail.com> Co-authored-by: Shelley Vohr <shelley.vohr@gmail.com> Co-authored-by: PatchUp <73610968+patchup[bot]@users.noreply.github.com>

At the point of `FreeEnvironment` and onwards, no JavaScript execution associated with the Environment should be triggered. Avoid draining platform tasks that can trigger JavaScript execution in `FreeEnvironment`. The holder of `node::Environment` should immediately call `node::MultiIsolatePlatform::UnregisterIsolate` and `v8::Isolate::Dispose` to cancel pending foreground tasks and join concurrent tasks after the environment was freed. `NodePlatform` can properly handle the case in `RunForegroundTask` when an Isolate out-lives its associated `node::Environment`. PR-URL: #51290 Fixes: #47748 Fixes: #49344 Reviewed-By: Vinícius Lourenço Claro Cardoso <contact@viniciusl.com.br> Reviewed-By: Matteo Collina <matteo.collina@gmail.com>

nodejs/node#51290

* chore: bump node in DEPS to v20.13.1 * chore: bump node in DEPS to v20.14.0 * chore: update build_add_gn_build_files.patch * chore: update patches * chore: update patches * build: encode non-ASCII Latin1 characters as one byte in JS2C nodejs/node#51605 * crypto: use EVP_MD_fetch and cache EVP_MD for hashes nodejs/node#51034 * chore: update filenames.json * chore: update patches * src: support configurable snapshot nodejs/node#50453 * test: remove test-domain-error-types flaky designation nodejs/node#51717 * src: avoid draining platform tasks at FreeEnvironment nodejs/node#51290 * chore: fix accidentally deleted v8 dep * lib: define FormData and fetch etc. in the built-in snapshot nodejs/node#51598 * chore: remove stray log * crypto: enable NODE_EXTRA_CA_CERTS with BoringSSL nodejs/node#52217 * test: skip test for dynamically linked OpenSSL nodejs/node#52542 * lib, url: add a `windows` option to path parsing nodejs/node#52509 * src: use dedicated routine to compile function for builtin CJS loader nodejs/node#52016 * test: mark test as flaky nodejs/node#52671 * build,tools: add test-ubsan ci nodejs/node#46297 * src: preload function for Environment nodejs/node#51539 * deps: update c-ares to 1.28.1 nodejs/node#52285 * chore: fixup * events: extract addAbortListener for safe internal use nodejs/node#52081 * module: print location of unsettled top-level await in entry points nodejs/node#51999 * fs: add stacktrace to fs/promises nodejs/node#49849 * chore: fixup indices --------- Co-authored-by: electron-roller[bot] <84116207+electron-roller[bot]@users.noreply.github.com> Co-authored-by: Cheng <zcbenz@gmail.com> Co-authored-by: Shelley Vohr <shelley.vohr@gmail.com> Co-authored-by: PatchUp <73610968+patchup[bot]@users.noreply.github.com>

…velock (#218) ## Problem The recurring `frontend_unit_tests` TIMEOUT flake (`close timed out after 120000ms` → `Failed to terminate worker while running ...`) is caused by a bug in Node 18.19.1 itself, not by test code. gdb backtrace of a live hung worker shows the livelock: on the worker's **normal exit path**, `FreeEnvironment() → NodePlatform::DrainTasks()` tries to drain the foreground task queue, but `FinalizationRegistryCleanupTask`s keep re-spawning faster than they drain — a native infinite loop at 100% CPU that `worker.terminate()` cannot interrupt. The FinalizationRegistry instances come from jsdom's `iterable-weak-set` (one per Document), which is why every test file is equally exposed and the blamed filename in the error is arbitrary. Upstream fix: nodejs/node#51290 (fixes nodejs/node#47748 "Infinite loop at shutdown"), backported to Node 18.20.3+. We are pinned one patch line short at 18.19.1. ## Fix Bump the bazel Node toolchain to 18.20.4 — the newest 18.x bundled with rules_nodejs 6.3.0, so no rules bump is needed. ## Verification A/B/C stress runs of shard 10/10 under identical 9-shard concurrent load (mimicking bazel's parallel shard execution): | Node | Hangs | |---|---| | 18.19.1 (current) | 12 / 31 iterations (~39%) | | 18.20.4 (this PR) | 0 / 60 | | 18.20.8 | 0 / 60 | Full bazel frontend suite on the 18.20.4 toolchain: 10/10 shards pass. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

…velock (#218) ## Problem The recurring `frontend_unit_tests` TIMEOUT flake (`close timed out after 120000ms` → `Failed to terminate worker while running ...`) is caused by a bug in Node 18.19.1 itself, not by test code. gdb backtrace of a live hung worker shows the livelock: on the worker's **normal exit path**, `FreeEnvironment() → NodePlatform::DrainTasks()` tries to drain the foreground task queue, but `FinalizationRegistryCleanupTask`s keep re-spawning faster than they drain — a native infinite loop at 100% CPU that `worker.terminate()` cannot interrupt. The FinalizationRegistry instances come from jsdom's `iterable-weak-set` (one per Document), which is why every test file is equally exposed and the blamed filename in the error is arbitrary. Upstream fix: nodejs/node#51290 (fixes nodejs/node#47748 "Infinite loop at shutdown"), backported to Node 18.20.3+. We are pinned one patch line short at 18.19.1. ## Fix Bump the bazel Node toolchain to 18.20.4 — the newest 18.x bundled with rules_nodejs 6.3.0, so no rules bump is needed. ## Verification A/B/C stress runs of shard 10/10 under identical 9-shard concurrent load (mimicking bazel's parallel shard execution): | Node | Hangs | |---|---| | 18.19.1 (current) | 12 / 31 iterations (~39%) | | 18.20.4 (this PR) | 0 / 60 | | 18.20.8 | 0 / 60 | Full bazel frontend suite on the 18.20.4 toolchain: 10/10 shards pass. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-authored-by: Claude Fable 5 <noreply@anthropic.com> (cherry picked from commit d93131d)

An async WebAssembly.compile() job still running on the platform worker threads when a program was torn down could crash the process in uv_async_send() (SIGSEGV at 0x78, a NULL uv_async_t): the node::CommonEnvironmentSetup destructor unregisters the isolate from the platform before disposing of it, and a compilation step completing in between posts a foreground task to the per-isolate task runner whose async handle is concurrently being nulled and closed. Node's FreeEnvironment() no longer drains the platform (nodejs/node#51290), so deleteIntern() now drains all platform tasks after node::Stop(), while the isolate is still registered, with JavaScript execution disallowed. Also fixes bugs found while testing: - JavaScriptPromise: the promise callback info held a V8 persistent handle without a weak program reference, so releasing a promise after its program was destroyed reset the handle in a disposed isolate (use-after-free) - node's per-process initialization marked the process' stdio file descriptors close-on-exec, so all child processes started after the module was loaded (system(), backquote(), ...) ran without stdin, stdout, and stderr; stdio inheritance is now enabled - JavaScriptObject::toData(): member functions of objects were bound to the function itself instead of the containing object as 'this'

nodejs-github-bot added c++ Issues and PRs that require attention from people who are familiar with C++. needs-ci PRs that need a full CI run. process Issues and PRs related to the process subsystem. labels Dec 26, 2023

legendecas force-pushed the platform-tasks branch from f9c4eb9 to 0cd1a72 Compare December 26, 2023 17:51

legendecas changed the title ~~src: avoid draining tasks at FreeEnvironment~~ src: avoid draining platform tasks at FreeEnvironment Dec 26, 2023

legendecas force-pushed the platform-tasks branch from 0cd1a72 to 8324ab0 Compare December 26, 2023 17:53

bcoe reviewed Dec 27, 2023

View reviewed changes

lpinca added the request-ci Add this label to start a Jenkins CI on a PR. Only starts once the PR has an approving review. label Dec 27, 2023

github-actions Bot removed the request-ci Add this label to start a Jenkins CI on a PR. Only starts once the PR has an approving review. label Dec 27, 2023

legendecas force-pushed the platform-tasks branch from 8324ab0 to beab806 Compare December 28, 2023 14:59

bcoe reviewed Dec 28, 2023

View reviewed changes

targos added the request-ci Add this label to start a Jenkins CI on a PR. Only starts once the PR has an approving review. label Dec 28, 2023

github-actions Bot removed the request-ci Add this label to start a Jenkins CI on a PR. Only starts once the PR has an approving review. label Dec 28, 2023

H4ad approved these changes Dec 28, 2023

View reviewed changes

H4ad added author ready PRs with CI started, the required approvals, and no outstanding review comments. review wanted PRs that need review. labels Dec 28, 2023

mcollina approved these changes Jan 8, 2024

View reviewed changes

mcollina added the commit-queue PRs queued for automated landing through the Commit Queue. label Jan 8, 2024

mcollina mentioned this pull request Jan 8, 2024

v21.6.0 proposal #51342

Merged

nodejs-github-bot removed the commit-queue PRs queued for automated landing through the Commit Queue. label Jan 8, 2024

nodejs-github-bot merged commit 5db35b4 into nodejs:main Jan 8, 2024

legendecas deleted the platform-tasks branch January 8, 2024 14:41

richardlau mentioned this pull request Mar 25, 2024

v20.12.0 proposal #52212

Merged

codebytere added a commit to electron/electron that referenced this pull request Apr 14, 2024

src: avoid draining platform tasks at FreeEnvironment

e0661d0

nodejs/node#51290

codebytere added a commit to electron/electron that referenced this pull request Apr 16, 2024

src: avoid draining platform tasks at FreeEnvironment

76cc0ea

nodejs/node#51290

mcollina mentioned this pull request May 16, 2024

v18.20.3 proposal #53028

Merged

richardlau added the backported-to-v18.x label May 16, 2024

codebytere added a commit to electron/electron that referenced this pull request May 31, 2024

src: avoid draining platform tasks at FreeEnvironment

a1ea056

nodejs/node#51290

YOU54F mentioned this pull request Jun 10, 2024

node sometimes hangs when NODE_V8_COVERAGE is set pact-foundation/pact-js#1074

Open

5 tasks

Yanpei-Wang mentioned this pull request Jul 15, 2026

[PUR-505] Bump Node toolchain to 18.20.4 to fix vitest worker exit livelock Circle-Cat/purrf#218

Merged

Navigation Menu

Uh oh!

src: avoid draining platform tasks at FreeEnvironment - #51290

src: avoid draining platform tasks at FreeEnvironment#51290
nodejs-github-bot merged 1 commit into
nodejs:mainfrom
legendecas:platform-tasks

legendecas commented Dec 26, 2023

Uh oh!

bcoe Dec 27, 2023

Uh oh!

legendecas Dec 28, 2023 •

edited

Loading

Uh oh!

nodejs-github-bot commented Dec 27, 2023

Uh oh!

bcoe left a comment

Uh oh!

nodejs-github-bot commented Dec 28, 2023

Uh oh!

mcollina left a comment

Uh oh!

nodejs-github-bot commented Jan 8, 2024

Uh oh!

hhugo commented Jan 8, 2024

Uh oh!

paulrutter commented Jan 8, 2024

Uh oh!

mcollina commented Jan 9, 2024

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

10 participants

Uh oh!

Conversation

legendecas commented Dec 26, 2023

Uh oh!

bcoe Dec 27, 2023

Choose a reason for hiding this comment

Uh oh!

legendecas Dec 28, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

nodejs-github-bot commented Dec 27, 2023

Uh oh!

bcoe left a comment

Choose a reason for hiding this comment

Uh oh!

nodejs-github-bot commented Dec 28, 2023

Uh oh!

mcollina left a comment

Choose a reason for hiding this comment

Uh oh!

nodejs-github-bot commented Jan 8, 2024

Uh oh!

hhugo commented Jan 8, 2024

Uh oh!

paulrutter commented Jan 8, 2024

Uh oh!

mcollina commented Jan 9, 2024

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

10 participants

legendecas Dec 28, 2023 •

edited

Loading