Currently, with threadpool on Unix, timer thread that checks and resolves stalls wakes up every threadpool_stall_limit milliseconds. This might be unnecessary on desktops/laptops and consume some CPU or even prevent machines from going to sleep mode, in case there is no client activity. We need to:

• detect the condition "no client activity"
• Put timer thread into the long sleep
• wake timer when the first client s active again.