FredS

FredS replied to a topic

Threads in PPL thread pool not returning to idle as expected

Using Delphi 10.3.2 I've encountered the following problem (specifically on Windows, not sure about other platforms): After using TParallel or TTask, not all threads in the thread pool return to idle as expected. I've prepared the following simple example: procedure TForm.Button1Click(Sender: TObject); var SW:TStopWatch; procedure Load; begin TParallel.For(0, 99999999, procedure(i: Integer) var T:Single; begin T:=Sin(i/PI); end); end; begin Memo1.Lines.Add('PPL Test ---------------'); Memo1.Lines.Add('Before: '+TThreadPoolStats.Current.Formatted); SW:=TStopWatch.StartNew; Load; Memo1.Lines.Add('Finished in '+SW.Elapsed.ToString); Memo1.Lines.Add('After: '+TThreadPoolStats.Current.Formatted); Memo1.Lines.Add('------------------------'); end; Repeatedly pressing the button results in the following output: PPL Test --------------- Before: 0 Worker Threads (56 .. 1400) / 0 Idle / 0 Retired / 0 Suspended / Avg. CPU = 0 / Curr. CPU = 0 Finished in 00:00:09.8423686 After: 56 Worker Threads (56 .. 1400) / 43 Idle / 0 Retired / 0 Suspended / Avg. CPU = 30 / Curr. CPU = 100 ------------------------ PPL Test --------------- Before: 56 Worker Threads (56 .. 1400) / 43 Idle / 0 Retired / 0 Suspended / Avg. CPU = 36 / Curr. CPU = 0 Finished in 00:00:04.7841342 After: 65 Worker Threads (56 .. 1400) / 50 Idle / 0 Retired / 0 Suspended / Avg. CPU = 43 / Curr. CPU = 50 ------------------------ PPL Test --------------- Before: 65 Worker Threads (56 .. 1400) / 50 Idle / 0 Retired / 0 Suspended / Avg. CPU = 42 / Curr. CPU = 0 Finished in 00:00:04.7090980 After: 74 Worker Threads (56 .. 1400) / 55 Idle / 0 Retired / 0 Suspended / Avg. CPU = 42 / Curr. CPU = 48 ------------------------ PPL Test --------------- Before: 74 Worker Threads (56 .. 1400) / 55 Idle / 0 Retired / 0 Suspended / Avg. CPU = 38 / Curr. CPU = 0 Finished in 00:00:02.3904253 After: 74 Worker Threads (56 .. 1400) / 50 Idle / 0 Retired / 0 Suspended / Avg. CPU = 57 / Curr. CPU = 98 ------------------------ PPL Test --------------- Before: 74 Worker Threads (56 .. 1400) / 50 Idle / 0 Retired / 0 Suspended / Avg. CPU = 51 / Curr. CPU = 0 Finished in 00:00:05.2865123 After: 84 Worker Threads (56 .. 1400) / 55 Idle / 0 Retired / 0 Suspended / Avg. CPU = 44 / Curr. CPU = 41 ------------------------ PPL Test --------------- Before: 84 Worker Threads (56 .. 1400) / 55 Idle / 0 Retired / 0 Suspended / Avg. CPU = 36 / Curr. CPU = 0 Finished in 00:00:06.8032435 After: 96 Worker Threads (56 .. 1400) / 64 Idle / 0 Retired / 0 Suspended / Avg. CPU = 36 / Curr. CPU = 32 ------------------------ PPL Test --------------- Before: 96 Worker Threads (56 .. 1400) / 64 Idle / 0 Retired / 0 Suspended / Avg. CPU = 0 / Curr. CPU = 0 Finished in 00:00:07.7921758 After: 111 Worker Threads (56 .. 1400) / 77 Idle / 0 Retired / 0 Suspended / Avg. CPU = 27 / Curr. CPU = 23 ------------------------ PPL Test --------------- Before: 111 Worker Threads (56 .. 1400) / 77 Idle / 0 Retired / 0 Suspended / Avg. CPU = 20 / Curr. CPU = 0 Finished in 00:00:02.3559000 After: 111 Worker Threads (56 .. 1400) / 77 Idle / 0 Retired / 0 Suspended / Avg. CPU = 48 / Curr. CPU = 100 ------------------------ PPL Test --------------- Before: 111 Worker Threads (56 .. 1400) / 77 Idle / 0 Retired / 0 Suspended / Avg. CPU = 48 / Curr. CPU = 0 Finished in 00:00:06.6491982 After: 122 Worker Threads (56 .. 1400) / 87 Idle / 0 Retired / 0 Suspended / Avg. CPU = 33 / Curr. CPU = 100 ------------------------ Even after the first run, only 43 threads in the pool seem to have returned to idle. Instead, with almost every further run, the total number of worker threads is increased to make up the difference. Despite this, the total CPU usage is going down slowly. If I wait for the inactive threads to be automatically retired and run the test again, I get something like this: PPL Test --------------- Before: 0 Worker Threads (56 .. 1400) / -36 Idle / 0 Retired / 0 Suspended / Avg. CPU = 0 / Curr. CPU = 0 Finished in 00:00:09.5530676 After: 56 Worker Threads (56 .. 1400) / 20 Idle / 0 Retired / 0 Suspended / Avg. CPU = 30 / Curr. CPU = 100 ------------------------ A negative number of idle threads? And after the run, only 20 threads are stated as idle again. Further runs result in the behavior described above. If someone could shed some light on all this for me, I would highly appreciate it! Am I doing something wrong here? What do I need to do to get 100% CPU usage on every run, without the thread pool ballooning? Thanks!
Then there is that undocumented 'Feature' where after you wait for termination your elapsed time goes from ~1s to ~81s: PPL Test --------------- Before: Worker: 0, (Min: 4..Max: 100), Idle: -9, Retired: 0, Suspended: 0, CPU(Avg): 5, CPU: 2 Finished in 00:00:01.6734632 After: Worker: 4, (Min: 4..Max: 100), Idle: -5, Retired: 0, Suspended: 0, CPU(Avg): 20, CPU: 100 ------------------------ After waiting for threads to terminate: PPL Test --------------- Before: Worker: 4, (Min: 4..Max: 100), Idle: -5, Retired: 0, Suspended: 0, CPU(Avg): 1, CPU: 1 Finished in 00:00:30.7872728 After: Worker: 4, (Min: 4..Max: 100), Idle: -3, Retired: 0, Suspended: 0, CPU(Avg): 20, CPU: 50 ------------------------ PPL Test --------------- Before: Worker: 4, (Min: 4..Max: 100), Idle: -3, Retired: 0, Suspended: 0, CPU(Avg): 1, CPU: 0 Finished in 00:01:21.6815195 After: Worker: 4, (Min: 4..Max: 100), Idle: -2, Retired: 1, Suspended: 0, CPU(Avg): 20, CPU: 100 ------------------------ I ran this in Berlin and it appears to work fine. But I made changes to System.Threading, both to achieve near 100% CPU usage as well as applying some of the patches submitted by those reporting to QC..   Then there is the issue that in Win64 this takes over 37 seconds, RSP-24570 is for CBuilder but its the same in Berlin and Rio 10.3.3