[ThreadPool Resource Leak]There are more and more scheduler threads in our production environment

[caiconghui]

[caiconghui]

I get the thread snapshot of fe process, and find there are many schduler thread.
grep "ScheduledThreadPoolExecutor.java:899" temp | wc -l
3826

[caiconghui]

Here, we start the thread in MasterTaskExecutor

https://github.com/apache/incubator-doris/blob/1fc82cd6e48ef5fed4d29b7e5cb4df63c3130027/fe/src/main/java/org/apache/doris/task/MasterTaskExecutor.java#L39-L43

in catalog ,we init loadTaskScheduler, we cannot gc loadTaskScheduler when checkpoint catalog is set to be null
https://github.com/apache/incubator-doris/blob/c9a7c373a709a9d8c195eaf4245fb06c1cc667fe/fe/src/main/java/org/apache/doris/catalog/Catalog.java#L527
It would warm up Config.async_load_task_pool_size threads when catalog is created