Deadlock when using 'System.exit(1)'

[mschorsch]

Describe the bug

If System.exit is used on the quarkus-main-thread the complete application deadlocks. If i execute System.exit on a different thread then the application terminates as expected.

Maybe this is an expected behaviour as stated in Quarkus#blockingExit but it feels odd.

The behavior should at least be explicitly described in the documentation. What do you think?

Expected behavior

No response

Actual behavior

No response

How to Reproduce?

1. deadlock_reproducer.zip
2. ./mvnw quarkus:dev

Output of uname -a or ver

Linux

Output of java -version

Java 17

GraalVM version (if different from Java)

No response

Quarkus version or git rev

Quarkus 2.13.3.Final

Build tool (ie. output of mvnw --version or gradlew --version)

Maven

Additional information

No response

[quarkus-bot]

/cc @Sanne, @aloubyansky, @geoand, @gsmet, @radcortez, @stuartwdouglas

[geoand]

This is expected as per the Javadoc you linked:

Starts the shutdown process, then waits for the application to shut down. Must not be called by the main thread, or a deadlock will result.

[geoand]

The behavior should at least be explicitly described in the documentation. What do you think?

Which documentation do you mean here? I don't see Quarkus.blockingExit being explicitly mentioned in any docs.

[mschorsch]

The behavior should at least be explicitly described in the documentation. What do you think?

Which documentation do you mean here? I don't see Quarkus.blockingExit being explicitly mentioned in any docs.

Thats the problem, the Quarkus documentation does not mention this behaviour.

[geoand]

The Javadoc does however.

I am interested in hearing where you would expect to see this behavior mentioned. Asking because if we just add it to some random place, there is little chance people will actually see it.

[geoand]

Actually, I have another idea... We can detect if this being called on the main thread and print an error

[mschorsch]

Actually, I have another idea... We can detect if this being called on the main thread and print an error

👍 . This is way better than the documentation.

[geoand]

#28939