[v12.x] backport BaseObjectPtr, SetImmediate(), Worker+report#32301

addaleax · 2020-03-16T13:14:16Z

Fun group backport of the following PRs because they all depend on each other:

addaleax · 2020-03-17T09:02:30Z

@codebytere Just rebased this, can you take a look?

nodejs-github-bot · 2020-03-17T09:03:14Z

CI: https://ci.nodejs.org/job/node-test-pull-request/29881/

nodejs-github-bot · 2020-03-20T13:18:02Z

CI: https://ci.nodejs.org/job/node-test-pull-request/29948/

addaleax · 2020-03-20T19:02:32Z

This should not be landed without #32393.

nodejs-github-bot · 2020-03-23T19:25:34Z

CI: https://ci.nodejs.org/job/node-test-pull-request/29996/

nodejs-github-bot · 2020-03-27T15:08:03Z

CI: https://ci.nodejs.org/job/node-test-pull-request/30126/

nodejs-github-bot · 2020-03-28T13:34:04Z

CI: https://ci.nodejs.org/job/node-test-pull-request/30152/

Referring to `BaseObject` instances using standard C++ smart pointers can interfere with BaseObject’s own cleanup mechanisms (explicit delete, delete-on-GC and delete-on-cleanup). Introducing custom smart pointers allows referring to `BaseObject`s safely while keeping those mechanisms intact. Refs: nodejs/quic#141 Refs: nodejs/quic#149 Reviewed-By: James M Snell <[email protected]> PR-URL: nodejs#30374 Refs: nodejs/quic#165 Reviewed-By: David Carlier <[email protected]>

Refs: nodejs/quic#141 Reviewed-By: James M Snell <[email protected]> PR-URL: nodejs#30374 Refs: nodejs/quic#149 Refs: nodejs/quic#165 Reviewed-By: David Carlier <[email protected]>

PR-URL: nodejs#30374 Refs: nodejs/quic#141 Refs: nodejs/quic#149 Refs: nodejs/quic#141 Refs: nodejs/quic#165 Reviewed-By: James M Snell <[email protected]> Reviewed-By: David Carlier <[email protected]>

This is no longer necessary now that the copyable `BaseObjectPtr` is available (as opposed to the only-movable `v8::Global`). PR-URL: nodejs#30374 Refs: nodejs/quic#141 Refs: nodejs/quic#149 Refs: nodejs/quic#141 Refs: nodejs/quic#165 Reviewed-By: James M Snell <[email protected]> Reviewed-By: David Carlier <[email protected]>

This allows keeping `BaseObjectPtr`s to `HandleWrap` instances. Previously, the pointer kept the `HandleWrap` object alive, leaving the Environment cleanup code that waits for the handle list to drain in a busy loop, because only the `HandleWrap` destructor removed the item from the list. Refs: nodejs/quic#165 Reviewed-By: James M Snell <[email protected]> Reviewed-By: Daniel Bevenius <[email protected]> PR-URL: nodejs#30374 Refs: nodejs/quic#141 Refs: nodejs/quic#149 Refs: nodejs/quic#141 Reviewed-By: David Carlier <[email protected]>

This was overlooked in a489583. Refs: nodejs#30374 PR-URL: nodejs#30666Fixes: nodejs#30643 Reviewed-By: Rich Trott <[email protected]> Reviewed-By: James M Snell <[email protected]>

This is in preparation for running native `SetImmediate()` callbacks during shutdown. PR-URL: nodejs#30666Fixes: nodejs#30643 Refs: nodejs#30374 Reviewed-By: Rich Trott <[email protected]> Reviewed-By: James M Snell <[email protected]>

Guard against running `SetImmediate()` from the destructor. The object will not be alive or usable in the callback, so it does not make sense to attempt to schedule the `SetImmediate()`. PR-URL: nodejs#30666Fixes: nodejs#30643 Refs: nodejs#30374 Reviewed-By: Rich Trott <[email protected]> Reviewed-By: James M Snell <[email protected]>

This can be necessary, because some parts of the Node.js code base perform cleanup operations in the Immediate callbacks, e.g. HTTP/2. This resolves flakiness in an HTTP/2 test that failed when a `SetImmediate()` callback was not run or destroyed before the `Environment` destructor started, because that callback held a strong reference to the `Http2Session` object and the expectation was that no such objects exist once the `Environment` constructor starts. Another, slightly more direct, alternative would have been to clear the immediate queue rather than to run it. However, this approach seems to make more sense as code generally assumes that the `SetImmediate()` callback will always run; For example, N-API uses an immediate callback to call buffer finalization callbacks. Unref’ed immediates are skipped, as the expectation is generally that they may not run anyway. Fixes: nodejs#30643 PR-URL: nodejs#30666 Refs: nodejs#30374 Reviewed-By: Rich Trott <[email protected]> Reviewed-By: James M Snell <[email protected]>

Refactor for clarity and reusability. Make it more obvious that the list is a FIFO queue. PR-URL: nodejs#31386 Refs: openjs-foundation/summit#240 Reviewed-By: Gireesh Punathil <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Colin Ihrig <[email protected]> Reviewed-By: Rich Trott <[email protected]>

There is no real reason to manage a count manually, given that checking whether there are C++ callbacks is a single pointer comparison. This makes it easier to add other kinds of native C++ callbacks that are managed in a similar way. PR-URL: nodejs#31386 Refs: openjs-foundation/summit#240 Reviewed-By: Gireesh Punathil <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Colin Ihrig <[email protected]> Reviewed-By: Rich Trott <[email protected]>

Prevent mistakes like the one fixed by the previous commit by destroying the callback immediately after it has been called. PR-URL: nodejs#31468 Refs: nodejs#31386 Reviewed-By: Colin Ihrig <[email protected]> Reviewed-By: Gireesh Punathil <[email protected]>

Fix the condition for deleting the underlying data pointed to by a `BaseObjectWeakPtr`, which erroneously skipped that deletion when `ptr->get()` was `nullptr`. This fixes a memory leak reported by some of the tests. Refs: nodejs#30374 (comment) PR-URL: nodejs#32393 Reviewed-By: Matheus Marchini <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Colin Ihrig <[email protected]>

MylesBorins · 2020-04-01T06:34:06Z

Rebased against staging

nodejs-github-bot · 2020-04-01T06:34:26Z

CI: https://ci.nodejs.org/job/node-test-pull-request/30319/

nodejs-github-bot · 2020-04-01T12:49:02Z

CI: https://ci.nodejs.org/job/node-test-pull-request/30324/

nodejs-github-bot · 2020-04-01T16:48:44Z

CI: https://ci.nodejs.org/job/node-test-pull-request/30330/

nodejs-github-bot · 2020-04-01T20:30:03Z

CI: https://ci.nodejs.org/job/node-test-pull-request/30337/

Referring to `BaseObject` instances using standard C++ smart pointers can interfere with BaseObject’s own cleanup mechanisms (explicit delete, delete-on-GC and delete-on-cleanup). Introducing custom smart pointers allows referring to `BaseObject`s safely while keeping those mechanisms intact. Refs: nodejs/quic#141 Refs: nodejs/quic#149 Reviewed-By: James M Snell <[email protected]> Backport-PR-URL: #32301 PR-URL: #30374 Refs: nodejs/quic#165 Reviewed-By: David Carlier <[email protected]>

Refs: nodejs/quic#141 Reviewed-By: James M Snell <[email protected]> Backport-PR-URL: #32301 PR-URL: #30374 Refs: nodejs/quic#149 Refs: nodejs/quic#165 Reviewed-By: David Carlier <[email protected]>

Backport-PR-URL: #32301 PR-URL: #30374 Refs: nodejs/quic#141 Refs: nodejs/quic#149 Refs: nodejs/quic#141 Refs: nodejs/quic#165 Reviewed-By: James M Snell <[email protected]> Reviewed-By: David Carlier <[email protected]>

This is no longer necessary now that the copyable `BaseObjectPtr` is available (as opposed to the only-movable `v8::Global`). Backport-PR-URL: #32301 PR-URL: #30374 Refs: nodejs/quic#141 Refs: nodejs/quic#149 Refs: nodejs/quic#141 Refs: nodejs/quic#165 Reviewed-By: James M Snell <[email protected]> Reviewed-By: David Carlier <[email protected]>

This allows keeping `BaseObjectPtr`s to `HandleWrap` instances. Previously, the pointer kept the `HandleWrap` object alive, leaving the Environment cleanup code that waits for the handle list to drain in a busy loop, because only the `HandleWrap` destructor removed the item from the list. Refs: nodejs/quic#165 Reviewed-By: James M Snell <[email protected]> Reviewed-By: Daniel Bevenius <[email protected]> Backport-PR-URL: #32301 PR-URL: #30374 Refs: nodejs/quic#141 Refs: nodejs/quic#149 Refs: nodejs/quic#141 Reviewed-By: David Carlier <[email protected]>

This was overlooked in a489583. Refs: #30374 Backport-PR-URL: #32301 PR-URL: #30666Fixes: #30643 Reviewed-By: Rich Trott <[email protected]> Reviewed-By: James M Snell <[email protected]>

This is in preparation for running native `SetImmediate()` callbacks during shutdown. Backport-PR-URL: #32301 PR-URL: #30666Fixes: #30643 Refs: #30374 Reviewed-By: Rich Trott <[email protected]> Reviewed-By: James M Snell <[email protected]>

Guard against running `SetImmediate()` from the destructor. The object will not be alive or usable in the callback, so it does not make sense to attempt to schedule the `SetImmediate()`. Backport-PR-URL: #32301 PR-URL: #30666Fixes: #30643 Refs: #30374 Reviewed-By: Rich Trott <[email protected]> Reviewed-By: James M Snell <[email protected]>

This can be necessary, because some parts of the Node.js code base perform cleanup operations in the Immediate callbacks, e.g. HTTP/2. This resolves flakiness in an HTTP/2 test that failed when a `SetImmediate()` callback was not run or destroyed before the `Environment` destructor started, because that callback held a strong reference to the `Http2Session` object and the expectation was that no such objects exist once the `Environment` constructor starts. Another, slightly more direct, alternative would have been to clear the immediate queue rather than to run it. However, this approach seems to make more sense as code generally assumes that the `SetImmediate()` callback will always run; For example, N-API uses an immediate callback to call buffer finalization callbacks. Unref’ed immediates are skipped, as the expectation is generally that they may not run anyway. Fixes: #30643 Backport-PR-URL: #32301 PR-URL: #30666 Refs: #30374 Reviewed-By: Rich Trott <[email protected]> Reviewed-By: James M Snell <[email protected]>

Refactor for clarity and reusability. Make it more obvious that the list is a FIFO queue. Backport-PR-URL: #32301 PR-URL: #31386 Refs: openjs-foundation/summit#240 Reviewed-By: Gireesh Punathil <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Colin Ihrig <[email protected]> Reviewed-By: Rich Trott <[email protected]>

There is no real reason to manage a count manually, given that checking whether there are C++ callbacks is a single pointer comparison. This makes it easier to add other kinds of native C++ callbacks that are managed in a similar way. Backport-PR-URL: #32301 PR-URL: #31386 Refs: openjs-foundation/summit#240 Reviewed-By: Gireesh Punathil <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Colin Ihrig <[email protected]> Reviewed-By: Rich Trott <[email protected]>

Add a variant of `SetImmediate()` that can be called from any thread. This allows removing the `AsyncRequest` abstraction and replaces it with a more generic mechanism. Backport-PR-URL: #32301 PR-URL: #31386 Refs: openjs-foundation/summit#240 Reviewed-By: Gireesh Punathil <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Colin Ihrig <[email protected]> Reviewed-By: Rich Trott <[email protected]>

Remove `AsyncRequest` from the source code, and replace its usage with threadsafe `SetImmediate()` calls. This has the advantage of being able to pass in any function, rather than one that is defined when the `AsyncRequest` is “installed”. This necessitates two changes: - The stopping flag (which was only used in one case and ignored in the other) is now a direct member of the `Environment` class. - Workers no longer have their own libuv handles, requiring manual management of their libuv ref count. As a drive-by fix, the `can_call_into_js` variable was turned into an atomic variable. While there have been no bug reports, the flag is set from `Stop(env)` calls, which are supposed to be possible from any thread. Backport-PR-URL: #32301 PR-URL: #31386 Refs: openjs-foundation/summit#240 Reviewed-By: Gireesh Punathil <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Colin Ihrig <[email protected]> Reviewed-By: Rich Trott <[email protected]>

Allow doing what V8’s `v8::Isolate::RequestInterrupt()` does for V8. This also works when there is no JS code currently executing. Backport-PR-URL: #32301 PR-URL: #31386 Refs: openjs-foundation/summit#240 Reviewed-By: Gireesh Punathil <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Colin Ihrig <[email protected]> Reviewed-By: Rich Trott <[email protected]>

This is a) the right thing to do anyway because these functions can not be inlined by the compiler and b) avoids compilation warnings in the following commit. Backport-PR-URL: #32301 PR-URL: #31386 Refs: openjs-foundation/summit#240 Reviewed-By: Gireesh Punathil <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Colin Ihrig <[email protected]> Reviewed-By: Rich Trott <[email protected]>

Include a report for each sub-Worker of the current Node.js instance. This adds a feature that is necessary for eventually making the report feature stable, as was discussed during the last collaborator summit. Refs: openjs-foundation/summit#240 Backport-PR-URL: #32301 PR-URL: #31386 Reviewed-By: Gireesh Punathil <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Colin Ihrig <[email protected]> Reviewed-By: Rich Trott <[email protected]>

Make `SetImmediate()` behave more like `process.nextTick()` (which matches how we use it) by also running tasks that have been added during previous `SetImmediate()` calls. Backport-PR-URL: #32301 PR-URL: #31502 Reviewed-By: Yongsheng Zhang <[email protected]> Reviewed-By: David Carlier <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Rich Trott <[email protected]> Reviewed-By: Colin Ihrig <[email protected]> Reviewed-By: Minwoo Jung <[email protected]>

de2c68c moved this call to the destructor, under the assumption that that would essentially be equivalent to running it as part of the callback since the worker would be destroyed along with the callback. However, the actual code in `Environment::RunAndClearNativeImmediates()` comes with the subtlety that testing whether a JS exception has been thrown happens between the invocation of the callback and its destruction, leaving a possible exception from `JoinThread()` potentially unhandled (and unintentionally silenced through the `TryCatch`). This affected exceptions thrown from the `'exit'` event of the Worker, and made the `parallel/test-worker-message-type-unknown` test flaky, as the invalid message was sometimes only received during the Worker thread’s exit handler. Fix this by moving the `JoinThread()` call back to where it was before. Refs: #31386 Backport-PR-URL: #32301 PR-URL: #31468 Reviewed-By: Colin Ihrig <[email protected]> Reviewed-By: Gireesh Punathil <[email protected]>

Prevent mistakes like the one fixed by the previous commit by destroying the callback immediately after it has been called. Backport-PR-URL: #32301 PR-URL: #31468 Refs: #31386 Reviewed-By: Colin Ihrig <[email protected]> Reviewed-By: Gireesh Punathil <[email protected]>

Fix the condition for deleting the underlying data pointed to by a `BaseObjectWeakPtr`, which erroneously skipped that deletion when `ptr->get()` was `nullptr`. This fixes a memory leak reported by some of the tests. Refs: #30374 (comment) Backport-PR-URL: #32301 PR-URL: #32393 Reviewed-By: Matheus Marchini <[email protected]> Reviewed-By: James M Snell <[email protected]> Reviewed-By: Colin Ihrig <[email protected]>

MylesBorins · 2020-04-01T21:37:59Z

green ci!

landed in d568efc...fbd5be6

nodejs-github-bot added c++ Issues and PRs that require attention from people who are familiar with C++. lib / src Issues and PRs related to general changes in the lib or src directory. v12.x labels Mar 16, 2020

codebytere force-pushed the v12.x-staging branch from 2047cca to 0606e1dCompare March 17, 2020 02:49

addaleax force-pushed the backport-report-worker-v12.x branch from 633a412 to 30bb555Compare March 17, 2020 09:02

addaleax added the blocked PRs that are blocked by other issues or PRs. label Mar 20, 2020

codebytere force-pushed the v12.x-staging branch from a01fb4a to e5dc2fbCompare March 23, 2020 16:01

addaleax force-pushed the backport-report-worker-v12.x branch from 3207231 to fb98111Compare March 23, 2020 19:24

addaleax added author ready PRs that have at least one approval, no pending requests for changes, and a CI started. and removed blocked PRs that are blocked by other issues or PRs. labels Mar 23, 2020

codebytere force-pushed the v12.x-staging branch 2 times, most recently from 63a03d2 to d577190Compare March 31, 2020 23:57

addaleax added 11 commits April 1, 2020 02:33

http2: use custom BaseObject smart pointers
e1b1267
Refs: nodejs/quic#141 Reviewed-By: James M Snell <[email protected]> PR-URL: nodejs#30374 Refs: nodejs/quic#149 Refs: nodejs/quic#165 Reviewed-By: David Carlier <[email protected]>

src: use BaseObjectPtr for keeping channel alive in dns bindings
95bf451
PR-URL: nodejs#30374 Refs: nodejs/quic#141 Refs: nodejs/quic#149 Refs: nodejs/quic#141 Refs: nodejs/quic#165 Reviewed-By: James M Snell <[email protected]> Reviewed-By: David Carlier <[email protected]>

src: keep object alive in stream_pipe code
b18531d
This was overlooked in a489583. Refs: nodejs#30374 PR-URL: nodejs#30666Fixes: nodejs#30643 Reviewed-By: Rich Trott <[email protected]> Reviewed-By: James M Snell <[email protected]>

src: add more can_call_into_js() guards
ece9faf
This is in preparation for running native `SetImmediate()` callbacks during shutdown. PR-URL: nodejs#30666Fixes: nodejs#30643 Refs: nodejs#30374 Reviewed-By: Rich Trott <[email protected]> Reviewed-By: James M Snell <[email protected]>

addaleax added 2 commits April 1, 2020 02:33

MylesBorins force-pushed the backport-report-worker-v12.x branch from fb98111 to d85cb07Compare April 1, 2020 06:33

MylesBorins closed this Apr 1, 2020

Uh oh!

[v12.x] backport BaseObjectPtr, SetImmediate(), Worker+report#32301

[v12.x] backport BaseObjectPtr, SetImmediate(), Worker+report #32301

Uh oh!

Conversation

addaleax commented Mar 16, 2020• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

addaleax commented Mar 17, 2020

Uh oh!

nodejs-github-bot commented Mar 17, 2020

Uh oh!

nodejs-github-bot commented Mar 20, 2020

Uh oh!

addaleax commented Mar 20, 2020

Uh oh!

nodejs-github-bot commented Mar 23, 2020

Uh oh!

nodejs-github-bot commented Mar 27, 2020

Uh oh!

nodejs-github-bot commented Mar 28, 2020

Uh oh!

MylesBorins commented Apr 1, 2020

Uh oh!

nodejs-github-bot commented Apr 1, 2020

Uh oh!

nodejs-github-bot commented Apr 1, 2020

Uh oh!

nodejs-github-bot commented Apr 1, 2020

Uh oh!

nodejs-github-bot commented Apr 1, 2020

Uh oh!

MylesBorins commented Apr 1, 2020

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

addaleax commented Mar 16, 2020•
edited
Loading