Repository navigation
Stream emitting same error twice under certain circomstances #36053
Description
- Version: 12.18.1 and newer
- Platform: Darwin watson 19.6.0 Darwin Kernel Version 19.6.0: Mon Aug 31 22:12:52 PDT 2020; root:xnu-6153.141.2~1/RELEASE_X86_64 x86_64
- Subsystem: stream
What steps will reproduce the bug?
Run this code on Node.js 12.18.1 or newer and notice how the error is emitted on b twice. On any older Node.js version it's only emitted once.
const { Readable, PassThrough, Writable, pipeline } = require('stream')
const a = new Readable({
read () {
this.destroy(new Error('foo'));
}
})
const b = new PassThrough()
const c = new Writable()
a.once('error', (err) => {
console.log('emitting error on b')
b.emit('error', err) // Note: you're not supposed to emit error manually!
})
b.on('error', (err) => {
console.log('b got error:', err.message)
})
pipeline(a.pipe(b), c, (err) => {
console.log('end of pipeline, err:', err.message)
})If I change b.emit('error', err) to b.destroy(err) the error goes away. Using destroy is best practise and the user isn't supposed to emit error events manually on streams like in the above code. However, this was how we told users to do it previously, so a lot of code most likely exists in the wild expecting this to be possible and emitting the same error twice can have bad consequences.
Thoughts:
- It would be cool if we could warn users if they manually emitted an
errorevent on a stream. - It would be nice if we could make this odd behaviour go away even if the user emits
errormanually.
This regression was most likely introduced in #30869.
Activity
mcollina commentedon Nov 9, 2020
This is not a regression from my point of view, it actually fixed something that was broken before. The problem is that
.emit('error', err)should not really be used as it bypass all the state management logic of streams..destroy(err)has been the recommendation since Node v8.x, which is long in the past - It's time to update.I would also note that this behave exactly as planned:
emit('error', err)causes the error to be emitted on the stream..pipeline()also does that via.destroy(err). I now agree that the change should likely have been semver-major, but I think that ship has long sailed as 12.x is heading into maintenance. Changing this also fixed some key bugs somewhere else, which is important for our users as well.I do not think there is anything to do here :/.
mafintosh commentedon Nov 9, 2020
@mcollina Ya agreed.
Watson and I discussed this issue offline and came to the same conclusion.
However I think there is something to be said here about the landscape being a bit confusing. I bet lots of people still think emit(error) is the way to go for streams. Any simple path where we could warn on that?
mcollina commentedon Nov 9, 2020
Unfortunately not. The problem is that
.emit('error', err)is also used by.destroy()and the rest of the internal machinery. We should really document this better.mafintosh commentedon Nov 9, 2020
@mcollina isn't it only emitted by destroy? Ie we can't just check if .destroyed = true in an overwrite of emit and throw a warning if not and the event is error?
mcollina commentedon Nov 9, 2020
Yes
We could. The problem is that adding an override for
emit()is likely going to slowdown things significantly as we emit a lot.mafintosh commentedon Nov 9, 2020
@mcollina We could just have an internal _emit that is just an alias to normal emit to avoid any slowdown
mcollina commentedon Nov 9, 2020
Unfortunately that's not possible as some descendants override
emit:/.We should probably benchmark this and ship it anyway. I'm happy to review a PR.
Never manually emit errors on streams
mmomtchev commentedon Nov 9, 2020
No statistically significant performance drop for
node benchmark/run.js streams(ie variations across runs are inline with variations with or without the modification)Everything seems to work, but still an avalanche of warnings during
make -j4 test- do as i say not as i do 😄Do you really think it makes sense to add it?
github-actions commentedon Jun 27, 2026
This issue has been marked as stale due to 210 days of inactivity.
It will be automatically closed in 30 days if no further activity occurs. If this is still relevant, please leave a comment or update it to keep it open.
github-actions commentedon Jul 28, 2026
This issue has been automatically closed after 30 days of inactivity following its stale status (no activity for a total of 120 days).
If this is still relevant, feel free to reopen it or leave a comment with additional details so we can continue the discussion.