Worker Leaking Issue

[jiazou-bigdata]

I've found that the number of available workers (waiting workers) in the frontend PDB server will gradually reduce along the execution, until a point when the total number of available workers become 0 and the whole program will freeze.

Increasing the number of workers can fix the problem temporarily.

I suspect the problem is because some of the threads cannot finish because allocation blocks in these threads cannot be freed successfully. These worker threads stuck somewhere and could not return.