test_runner: cleanup test timeout abort listener #48915
Conversation
|
Review requested:
|
nodejs-github-bot
added
needs-ci
lib/internal/test_runner/test.js
Outdated
| deferred.promise = PromisePrototypeThen(deferred.promise, () => { | ||
| throw new ERR_TEST_FAILURE( | ||
| `test timed out after ${timeout}ms`, | ||
| kTestTimeoutFailure, | ||
| ); | ||
| }); |
There was a problem hiding this comment.
I wanted to put this in the setTimeout but for some reason, the test-runner-output.mjs test failed for the describe_it file with some missing stack trace
lib/internal/test_runner/test.js
Outdated
| function throwOnAbort() { | ||
| deferred.reject(new AbortError(undefined, { cause: signal.reason })); | ||
| clearTimeout(timer); | ||
| } |
There was a problem hiding this comment.
Done this to match the current behavior:
Line 48 in 78f6952
but we should align it with just using an abort signal with no timeout - no error being thrown or throw an abort error
lib/internal/test_runner/test.js
Outdated
| const cleanAbort = addAbortListener(signal, throwOnAbort)[SymbolDispose]; | ||
| const timer = setTimeout(() => { | ||
| deferred.resolve(); | ||
| cleanAbort(); | ||
| }, timeout); | ||
|
|
||
| timer.unref(); |
There was a problem hiding this comment.
Could/Should this be an AbortSignal.any([signal, AbortSignal.timeout(timeout)]) etc?
There was a problem hiding this comment.
I don't think it would change anything as the timer won't get cleaned...
There was a problem hiding this comment.
What do you mean about the timer not getting clean?
The AbortSignal.any uses weak refs, so once this is GCed it would clear the timeout...
I think this suggestion could potentially reduce the complexity here, and also make it easier to land on v18, which doesn't have dispose & addAbortListener at the moment
WDYT?
There was a problem hiding this comment.
thank you! this is a much cleaner solution! the only problem here is that we create 2 new AbortSignal (AbortSignal.any and AbortSignal.timeout) for every describe/test/hook run, which can be bad for performance...
There was a problem hiding this comment.
@MoLow what do you think? should we remove the addAbortListener and dispose in order to support v18 and v16?
There was a problem hiding this comment.
I don't think this actually simplifies 😬
There was a problem hiding this comment.
@rluvaton I had something like this in mind:
function stopTest(timeout, signal) {
const timeoutSignal = timeout !== kDefaultTimeout && AbortSignal.timeout(timeout);
const composedSignal = AbortSignal.any([signal, timeoutSignal].filter(Boolean));
return PromisePrototypeThen(once(composedSignal, 'abort'), (e) => {
if (e.target.reason !== timeoutSignal.reason) {
return;
}
throw new ERR_TEST_FAILURE(
`test timed out after ${timeout}ms`,
kTestTimeoutFailure,
);
})
};But current solution works for me 🙂
There was a problem hiding this comment.
If we accept the extra creation of an abort signal I highly prefer that
There was a problem hiding this comment.
I assume we would want to profile both and decide based on that?
@MoLow @benjamingr what is your opinion here?
There was a problem hiding this comment.
I've created a benchmark PR and after it is merged we can check different solutions:
lib/internal/test_runner/test.js
Outdated
| const cleanAbort = addAbortListener(signal, throwOnAbort)[SymbolDispose]; | ||
| const timer = setTimeout(() => { | ||
| deferred.resolve(); | ||
| cleanAbort(); | ||
| }, timeout); | ||
|
|
||
| timer.unref(); |
There was a problem hiding this comment.
What do you mean about the timer not getting clean?
The AbortSignal.any uses weak refs, so once this is GCed it would clear the timeout...
I think this suggestion could potentially reduce the complexity here, and also make it easier to land on v18, which doesn't have dispose & addAbortListener at the moment
WDYT?
|
@MoLow updated with your changes and modified to use |
to which listener? the abort listener? because using the |
|
The issue is the listener won't be removed right? kWeakListener means "remove the listener when that resource is gcd" |
|
(I removed the timeout support so I try to make it work for the simple case) I understand, but when I tried changing the const { aborted } = require('internal/abort_controller');
function stopTest(timeout, signal) {
return aborted(signal, {});
}The test I wrote: const { beforeEach, afterEach, test} = require("node:test");
beforeEach(() => {
global.gc();
});
afterEach(() => {
global.gc();
});
for (let i = 1; i <= 11; ++i) {
test(`${i}`, () => {
global.gc();
});
}it still log that: also when doing this it had the same effect: function stopTest(timeout, signal) {
return new Promise((resolve) => {
signal.addEventListener('abort', resolve, { [kWeakHandler]: {} });
setImmediate(() => {
global.gc()
}).unref()
});
}Maybe I'm missing something? |
@rluvaton if we want to keep using the original signal we could keep updating |
|
The problem is that we are not cleaning up the abort listener (see that it still happening even when we GC manually) |
|
The reason why the warning still printed was because of #48951 |
no longer needed after nodejs#48915 fix PR-URL: nodejs#48989 Reviewed-By: Moshe Atlow <moshe@atlow.co.il> Reviewed-By: Chemi Atlow <chemi@atlow.co.il>
|
Due to fact, #48877 didn't land cleanly on v20.x-staging. This PR somehow, depends on it. So we'll need a manual backport. |
RafaelGSS
added
the
backport-requested-v20.x
fix nodejs#48475 PR-URL: nodejs#48915 Reviewed-By: Moshe Atlow <moshe@atlow.co.il> Reviewed-By: Benjamin Gruenbaum <benjamingr@gmail.com> Reviewed-By: Chemi Atlow <chemi@atlow.co.il>
no longer needed after nodejs#48915 fix PR-URL: nodejs#48989 Reviewed-By: Moshe Atlow <moshe@atlow.co.il> Reviewed-By: Chemi Atlow <chemi@atlow.co.il>
rluvaton
added
backport-open-v20.x
fix nodejs#48475 PR-URL: nodejs#48915 Reviewed-By: Moshe Atlow <moshe@atlow.co.il> Reviewed-By: Benjamin Gruenbaum <benjamingr@gmail.com> Reviewed-By: Chemi Atlow <chemi@atlow.co.il>
no longer needed after nodejs#48915 fix PR-URL: nodejs#48989 Reviewed-By: Moshe Atlow <moshe@atlow.co.il> Reviewed-By: Chemi Atlow <chemi@atlow.co.il>
There is an `if` statement that likely should have been an `else` in the original PR. Refs: nodejs#48915
There is an `if` statement that likely should have been an `else` in the original PR. Refs: #48915 PR-URL: #49943 Reviewed-By: Chemi Atlow <chemi@atlow.co.il> Reviewed-By: Moshe Atlow <moshe@atlow.co.il> Reviewed-By: Raz Luvaton <rluvaton@gmail.com> Reviewed-By: Benjamin Gruenbaum <benjamingr@gmail.com> Reviewed-By: LiviaMedeiros <livia@cirno.name>
There is an `if` statement that likely should have been an `else` in the original PR. Refs: nodejs#48915 PR-URL: nodejs#49943 Reviewed-By: Chemi Atlow <chemi@atlow.co.il> Reviewed-By: Moshe Atlow <moshe@atlow.co.il> Reviewed-By: Raz Luvaton <rluvaton@gmail.com> Reviewed-By: Benjamin Gruenbaum <benjamingr@gmail.com> Reviewed-By: LiviaMedeiros <livia@cirno.name>
There is an `if` statement that likely should have been an `else` in the original PR. Refs: #48915 PR-URL: #49943 Reviewed-By: Chemi Atlow <chemi@atlow.co.il> Reviewed-By: Moshe Atlow <moshe@atlow.co.il> Reviewed-By: Raz Luvaton <rluvaton@gmail.com> Reviewed-By: Benjamin Gruenbaum <benjamingr@gmail.com> Reviewed-By: LiviaMedeiros <livia@cirno.name>
targos
added
backported-to-v20.x
There is an `if` statement that likely should have been an `else` in the original PR. Refs: #48915 PR-URL: #49943 Reviewed-By: Chemi Atlow <chemi@atlow.co.il> Reviewed-By: Moshe Atlow <moshe@atlow.co.il> Reviewed-By: Raz Luvaton <rluvaton@gmail.com> Reviewed-By: Benjamin Gruenbaum <benjamingr@gmail.com> Reviewed-By: LiviaMedeiros <livia@cirno.name>
There is an `if` statement that likely should have been an `else` in the original PR. Refs: nodejs#48915 PR-URL: nodejs#49943 Reviewed-By: Chemi Atlow <chemi@atlow.co.il> Reviewed-By: Moshe Atlow <moshe@atlow.co.il> Reviewed-By: Raz Luvaton <rluvaton@gmail.com> Reviewed-By: Benjamin Gruenbaum <benjamingr@gmail.com> Reviewed-By: LiviaMedeiros <livia@cirno.name>
fix #48475 PR-URL: nodejs/node#48915 Reviewed-By: Moshe Atlow <moshe@atlow.co.il> Reviewed-By: Benjamin Gruenbaum <benjamingr@gmail.com> Reviewed-By: Chemi Atlow <chemi@atlow.co.il>
There is an `if` statement that likely should have been an `else` in the original PR. Refs: nodejs/node#48915 PR-URL: nodejs/node#49943 Reviewed-By: Chemi Atlow <chemi@atlow.co.il> Reviewed-By: Moshe Atlow <moshe@atlow.co.il> Reviewed-By: Raz Luvaton <rluvaton@gmail.com> Reviewed-By: Benjamin Gruenbaum <benjamingr@gmail.com> Reviewed-By: LiviaMedeiros <livia@cirno.name>
fix #48475 PR-URL: nodejs/node#48915 Reviewed-By: Moshe Atlow <moshe@atlow.co.il> Reviewed-By: Benjamin Gruenbaum <benjamingr@gmail.com> Reviewed-By: Chemi Atlow <chemi@atlow.co.il>
There is an `if` statement that likely should have been an `else` in the original PR. Refs: nodejs/node#48915 PR-URL: nodejs/node#49943 Reviewed-By: Chemi Atlow <chemi@atlow.co.il> Reviewed-By: Moshe Atlow <moshe@atlow.co.il> Reviewed-By: Raz Luvaton <rluvaton@gmail.com> Reviewed-By: Benjamin Gruenbaum <benjamingr@gmail.com> Reviewed-By: LiviaMedeiros <livia@cirno.name>
fix #48475