Fix fatal error propagation in async operations #4512
Conversation
|
Hi @armanbilge, @durban Would you mind taking a look at this when you have a moment? Thanks! |
| * @param fut | ||
| * The `java.util.concurrent.CompletableFuture` to suspend in `F[_]` | ||
| */ | ||
| def fromCompletableFuture[A](fut: F[CompletableFuture[A]]): F[A] = cont { |
There was a problem hiding this comment.
Are changes to this method necessary? #4505 (comment) suggests that the problem is actually in async and not directly in fromCompletableFuture.
There was a problem hiding this comment.
That's right that async_ is the core issue, but i think both changes are needed:
fromCompletableFuture: Uses CompletableFuture.handle() which catches all exceptions, including fatal ones which without fix, OutOfMemoryError gets wrapped instead of crashing the JVM.
async_: The IO.delay() wrapper was catching fatal errors during callback registration, causing the same problem.
Both mechanisms need the fix to ensure fatal errors consistently crash the JVM.
There was a problem hiding this comment.
I added more insights on how the error re-surface from onError here: #4505 (comment)
There was a problem hiding this comment.
Thanks @tpetillot
I added more insights on how the error re-surface from
onErrorhere: #4505 (comment)
looks like this code (Attached on the comment) is not detecting the original fatal error, but it's actually:
- Detecting errors thrown BY the error handler
(f(error))but NOT detecting the original fatal error that triggeredonError
The fatal error detection happens after the onError machinery, not during it.
I need to be guided if that is right 😊
There was a problem hiding this comment.
Yep, definitely missed that, tried to draw the route of the error outcome for OOM with IO.cont here: #4505 (comment)
najuna-brian
force-pushed
the
fix/fatal-error-propagation
branch
from
4b68700 to
ee0f110
Compare
Thanks for the review and guidance. Just learnt about all these upstream branches now 😊 |
|
Hello @armanbilge |
3 participants
Description
Fixes #4505
Fixed fatal error propagation in async operations so fatal errors crash the JVM instead of being caught and wrapped.
Changes
kernel/jvm/src/main/scala/cats/effect/kernel/AsyncPlatform.scala: Added fatal error detection infromCompletableFutureto re-throw fatal errors instead of wrapping themcore/shared/src/main/scala/cats/effect/IO.scala: RemovedIO.delay()wrapper inIO.async_to allow fatal errors to propagate during callback registrationWhy both changes?
fromCompletableFutureusesCompletableFuture.handle()which catches all exceptions, including fatal onesasync_hadIO.delay()wrapper catching fatal errors during callback registrationTesting
fromCompletableFutureandasync_fatal error scenarios.onError(_ => IO.unit)workaround mentioned in the issue