Details
-
Bug
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Incomplete
-
10.2.22, 10.3.12
Description
When variable of thread_handling is assigned as pool-of-threads, there are at most thread_pool_size * thread_pool_oversubscribe working threads to deal with clients request. So when the concurrence of clients is larger than thread_pool_size * thread_pool_oversubscribe and some threads wait for binlog flush or redo log flush, working threads of thread group will decrease and as a result , the system throughput will decrease too.
To improve the throughput, we can let the thread pool worker thread continue to deal with client requests and have some other thread to deal with the waiting user thread.need