[BUG] Perf framework hangs while executing async scenario with insufficient heap.

[kasobol-msft]

Try the following scenario:

java -Xms100m -Xmx100m -jar azure-storage-perf-1.0.0-beta.1-jar-with-dependencies.jar uploadblob --warmup 0 --duration 1 --size 1048576000

The program keeps reporting zero progress. It is expected to crash with OOM in that case.

It repros easily on local machime.
Test run where this happened: https://dev.azure.com/azure-sdk/internal/_build/results?buildId=899330&view=logs&j=1ebff73e-a26c-5a81-5f48-b56b69f60e6e&t=3f94ee88-7ff1-5b4e-3fcb-67bf1ac93b37

In order to triage whether it's SDK or perf framework I executed similar scenario in simple console app. That one failed immediately with OOM. That suggest problem is in perf framework.

[mikeharder]

The root cause is exceptions like OutOfMemoryError are handled differently by the default Reactor schedulers. Instead of terminating the Flux, the Flux will hang and the exception is only sent to the thread's uncaughtExceptionHandler and the Reactor Schedulers.onHandleError.

reactor/reactor-core#2521

https://github.com/reactor/reactor-core/blob/84e921fbfd84754d519b5e3c78262495dcf48b87/reactor-core/src/main/java/reactor/core/Exceptions.java#L443

https://github.com/reactor/reactor-core/blob/84e921fbfd84754d519b5e3c78262495dcf48b87/reactor-core/src/main/java/reactor/core/Exceptions.java#L443

The perf framework will be fixed by this PR: #21690

This issue could also impact any SDK code which uses the default Reactor schedulers.