java binding: marker to detect threads used by the Operator/AsyncOperator

[BaurzhanSakhariev]

I use blocking Operator via Operator.of(NAME, config) which makes blocking operator out of AsyncOperator with default global executor.

Since creation of threads is not controlled by users (I found only API to provide int number of cores), I wonder if there is a way to somehow mark threads created by OpenDAL, for example assign names with pre-defined prefix.

Context: We use RandomizedRunner to run tests and it tries to detect leaked threads.
https://github.com/randomizedtesting/randomizedtesting/blob/master/randomized-runner/src/main/java/com/carrotsearch/randomizedtesting/ThreadLeakControl.java#L629-L633

I always got report about leaked daemon threads after test passes.
Thread dump indicates I have some RUNNABLE non-Java threads and their id-s coincide with reported thread ids.

I think it's a false positive alert and would like to filter OpenDAL threads like https://github.com/crate/crate/blob/master/server/src/testFixtures/java/io/crate/lucene/CrateLuceneTestCase.java#L173
but need some reliable marker.

Thread names look like Thread-{some_int} but ideally would like to rely on more reliable marker name or any other way to identify those threads.

Observation: thread leak report always shows 12 threads and 12 is the number of processors on my machine,
output of sysctl -n hw.ncpu is 12.

Even though I use Operator.of(NAME, config) == AsyncOperator.of(schema, map, null).blocking() it behaves like I used

AsyncOperator
                .of(NAME, config, AsyncExecutor.createTokioExecutor(Runtime.getRuntime().availableProcessors()))
                .blocking()

[Xuanwo]

Hi, I believe all thread created by tokio should default to tokio-runtime-worker. Would you like to give it a try?

[BaurzhanSakhariev]

The default name is “tokio-runtime-worker”.

I checked thread dumps, haven't seen a thread with such name neither with default executor nor with executor created with specified number of cores.

[tisonkun]

Since creation of threads is not controlled by users (I found only API to provide int number of cores), I wonder if there is a way to somehow mark threads created by OpenDAL, for example assign names with pre-defined prefix.

Customized thread prefix is possible. Let me take a look. Maybe you can open an issue for it? Looks like a new feature proposal.

[Xuanwo]

I personally feel that exposing such an API is not advisable since tokio is merely our implementation detail, and we might still introduce other runtimes.

[tisonkun]

The configured thread name may not be respected by the Java runtime also.

[tisonkun]

Perhaps we should figure out how to pass through the name via JNI - attach_current_thread_as_daemon.

[tisonkun]

jni-rs/jni-rs#543

[BaurzhanSakhariev]

Thanks for opening an issue in jni-rs! I will probably use Thread-xxx pattern for now.

Maybe you can open an issue for it? Looks like a new feature proposal.

Opened #5088

[tisonkun]

Work out a way:

Will send a PR soon.