Myracle GUI

[mar]

I have the tournament running right now and just now Reckless lost two times on time and a couple of games later Stockfish 18 failed to start (this time while I'm at the PC). So far, about 840 games have been played, and this "time forfeit" thing happened for the first time. I'd say that it's my PC's issue, but since it never happened with Arena, I thought that it could have to do with the GUI. I'm using my PC while the tournament is running, but I'm not doing anything heavy.

I'm using 1 thread per engine, 128 mb Hash, time control 30+0.3, no syzygy, no ponder, concurrency of 2. I'm not changing any engines settings other than overriding Hash and Threads.

Here are some info about init times from a few games:
engine init (Stockfish 18) took 2424 msec
engine init (PlentyChess 7.0.0) took 123 msec
engine init (Stockfish 18) took 2212 msec
engine init (Obsidian 16.0) took 814 msec
engine init (Reckless 0.9.0) took 90 msec
engine init (Stockfish dev-20260519-133731f3) took 2047 msec
engine init (Obsidian 16.0) took 190 msec
engine init (PlentyChess 7.0.0) took 108 msec
engine init (Stockfish 18) took 555 msec
engine init (Stockfish dev-20260519-133731f3) took 1964 msec

ok thanks, could you please upload (ideally compressed) <tourname>.anomaly.log in tournaments subfolder?
this looks like the engine wasn't getting the resources for a while and it's quite suspicious to me

concurrency of 2 on your hw should be very generous, so I wonder why this is happening. also the SF init of >2 seconds with a 128M hash
seems quite long. what's strange that sometimes it's only half a second - are you sure SF is configured to only use 128M?

I'll definitely increase engine startup limit from 10 seconds in the next build. 1 minute seems silly, so I'll probably go with 30 seconds instead or make it configurable. auto-detect limit will be kept at 10secs

btw Arena doesn't run multiple games in parallel, but concurrency of 2 isn't much

is it possible that some heavy processes (utilizing lots of CPU/RAM) were coincidentally running at the same time this happened?

[vencenda]

ok thanks, could you please upload (ideally compressed) <tourname>.anomaly.log in tournaments subfolder?

I don't know if I can upload files here, so I uploaded it here:
https://drive.proton.me/urls/P0D3JCHSD0#p82d0TjDAiqR

are you sure SF is configured to only use 128M?

Yes. In the new tournament I just started, I set no concurrency, 2 threads per engine and 256 MB hash, and also enabled logs.

is it possible that some heavy processes (utilizing lots of CPU/RAM) were coincidentally running at the same time this happened?

Not impossible, but I'm not really doing much other than the tournament. I have 5 or 6 browser tabs open, I had youtube playing at some point, nothing much. If something heavy is running, it has to be something system related I'm not aware of. I have an SSD if it's related to init times. No syzygy.

btw Arena doesn't run multiple games in parallel

I know and it is one of the reasons I like your GUI. I just meant that if it was something about my system, I'd expect it to have affected Arena too, maybe. Although I realize it's more complicated than this. On Arena, I was using 2 threads and 256 MB hash per engine.

[mar]

ok thanks, could you please upload (ideally compressed) <tourname>.anomaly.log in tournaments subfolder?

I don't know if I can upload files here, so I uploaded it here:
https://drive.proton.me/urls/P0D3JCHSD0#p82d0TjDAiqR

are you sure SF is configured to only use 128M?

Yes. In the new tournament I just started, I set no concurrency, 2 threads per engine and 256 MB hash, and also enabled logs.

is it possible that some heavy processes (utilizing lots of CPU/RAM) were coincidentally running at the same time this happened?

Not impossible, but I'm not really doing much other than the tournament. I have 5 or 6 browser tabs open, I had youtube playing at some point, nothing much. If something heavy is running, it has to be something system related I'm not aware of. I have an SSD if it's related to init times. No syzygy.

btw Arena doesn't run multiple games in parallel

I know and it is one of the reasons I like your GUI. I just meant that if it was something about my system, I'd expect it to have affected Arena too, maybe. Although I realize it's more complicated than this. On Arena, I was using 2 threads and 256 MB hash per engine.

thanks, so a bit of a misunderstanding - by concurrency I meant what's your core budget set to in settings/concurrency
I also found a bug in my UCI startup code where the engine only got 5 seconds to reply to isready, this will be fixed in the next build

back to your anomaly log, I see this:

Code: Select all

3615844560> go wtime 3212 btime 11361 winc 300 binc 300
3615847753< info depth 1 seldepth 1 multipv 1 score cp 48 nodes 36 time 2 nps 17918 hashfull 0 tbhits 0 pv f4g5
3615847925< info depth 2 seldepth 3 multipv 1 score cp 48 nodes 80 time 2 nps 34554 hashfull 0 tbhits 0 pv f4g5 e7g5 e3f1
3615848096< info depth 3 seldepth 4 multipv 1 score cp 48 nodes 129 time 2 nps 51542 hashfull 0 tbhits 0 pv f4g5 e7g5 e3f1 f5f4
3615848199< info depth 4 seldepth 5 multipv 1 score cp 48 nodes 183 time 2 nps 70013 hashfull 0 tbhits 0 pv f4g5 e7g5 e3f1 f5f4 e5e6
3615848324< info depth 5 seldepth 6 multipv 1 score cp 48 nodes 253 time 2 nps 92732 hashfull 0 tbhits 0 pv f4g5 e7g5 e3f1 f5f4 e5e6 g6g7
3615848449< info depth 6 seldepth 7 multipv 1 score cp 48 nodes 334 time 2 nps 116877 hashfull 0 tbhits 0 pv f4g5 e7g5 e3f1 f5f4 e5e6 g6g7 c1d2
3615850585< info depth 7 seldepth 11 multipv 1 score cp 48 nodes 556 time 4 nps 113462 hashfull 0 tbhits 0 pv f4g5 e7g5 e3f1 f5f4 e5e6 g6g7 c1d2 h6f5 d2c3 f5d4 e1d2
3616129425< info depth 8 seldepth 17 multipv 1 score cp 64 nodes 7740 time 283 nps 27273 hashfull 0 tbhits 0 pv d5d6 g5f4 e3d5 e7h4 c1f4
3616271569< info depth 9 seldepth 15 multipv 1 score cp 47 nodes 11858 time 425 nps 27842 hashfull 0 tbhits 0 pv d5d6 g5f4 e3d5 e7h4 c1f4 g6g7 a1d1 h6f7 e5e6 h4g3 h2g1 f7d6 f4g3
3619119403< info depth 10 seldepth 22 multipv 1 score cp 55 nodes 22055 time 3273 nps 6737 hashfull 0 tbhits 0 pv d5d6 g5f4 e3d5 e7h4 c1f4 g6g7 e5e6 h4f2 e1f2 g7a1 f2e3 g8g7
3619122045< bestmove d5d6

timestamp is in microseconds, also notice the extremely low nps from reckless, it was definitely competing for resources with something else
notice it took 2.8seconds to go from depth 9 to depth 10, that's completely unrealistic

[vencenda]

thanks, so a bit of a misunderstanding - by concurrency I meant what's your core budget set to in settings/concurrency

I have set core budget to 2 in settings/concurrency. In the first tournament, for which I raised the issue, I had enabled concurrency in tournament settings and each engine used 1 thread, so two games were running in parallel. In the second tournament I started now, I just unticked concurrency in tournament settings, and also set threads to 2 and hash to 256 MB. I haven't changed the settings/concurrency setting.

timestamp is in microseconds, also notice the extremely low nps from reckless, it was definitely competing for resources with something else
notice it took 2.8seconds to go from depth 9 to depth 10, that's completely unrealistic

Yes, this makes total sense. So maybe something was using resources momentarily at that point without me realizing. I don't know what it could be but it seems unrelated to the GUI. On Arena, I had "time flagging" disabled, so it's possible that things like this were happening there too and I was just missing them.

But about the init times, even though I assume that it could also be caused by something similar, 2 to 2.5 seconds for Stockfish under normal circumstances still seems too much when the limit is 10 seconds. I consider this a bigger issue (even if the problem is my PC), because an initialization failure stops the tournament entirely, and you may discover it many hours later. I wonder if it would be possible for the tournament to just continue with the next game if an engine failed to start and just show a warning next to the game like with time loss. Or maybe have it as an option. Or maybe let the user increase the init time limit themselves in case they face frequent initialization failures.

[Mark 1]

Hi...



It would be good to make this configurable. Some of the engines integrated through MessChess probably don’t start because they need more time to become ready.

Best regards
Mark 1

[vencenda]

(Sorry if I'm tiring you, I'm just trying to help.)

I found another bug, but unfortunately it doesn't happen every time and I can't reliably trigger it. But here it is:
When you have enabled "repeat reverse", if you pause the tournament and close the program between the two games of a pair, the opening doesn't always repeat. But it doesn't happen every time. In my first, 1000-game tournament, 28 out of 500 game pairs had different openings (I use a python script to find them and then check manually to confirm). In my case, it happened only when I had closed the program between the two games of the pair, but again, not in every such case.

Trying to reproduce it, I created a little tournament with just two engines and with 10 games played. In this case, pairs 2, 3 and 5 had different openings. I used UHO_2024_8mvs_+085_+094.pgn opening book. I have uploaded the debug file (tour4.debug) here:
https://drive.proton.me/urls/P0D3JCHSD0#p82d0TjDAiqR

[mar]

ok thanks, I'll have a look once I come home from work - somehow I don't get free food or rent...

by the way a pgn file would have been better, the debug log is very large and digging through it will be extremely painful


the opening pool is not saved to make tour files siginifcantly smaller, rather a random seed is saved and the pool is regenerated on load,
last time I checked this worked fine

did you by chance use any advanced tour editing features, namely adding/removing engines to existing tour? these change the opening
pool so new games will use different openings. this is by design and it was the only way to keep things simple and elegant,
a reasonable price to pay I think

further possibilities are changing the book midway (like changing opening book set) or changing the contents of the opening book

if you didn't do any of this (I assume you didn't) then sadly it's a bug that needs fixing, however I don't see how it would happen only now and then,
must investigate further...

EDIT: does it mean you closed the GUI more than 28 times while the tour was running?

[vencenda]

by the way a pgn file would have been better, the debug log is very large and digging through it will be extremely painful

I uploaded the PGN file in the same link. Btw I used very short time control to test, but it happened with longer time controls too.

did you by chance use any advanced tour editing features, namely adding/removing engines to existing tour? these change the opening
pool so new games will use different openings. this is by design and it was the only way to keep things simple and elegant,
a reasonable price to pay I think

further possibilities are changing the book midway (like changing opening book set) or changing the contents of the opening book

if you didn't do any of this (I assume you didn't) then sadly it's a bug that needs fixing, however I don't see how it would happen only now and then,
must investigate further...

I didn't do any of that. Yes, it was frustrating to try and trigger. It seems so random.

EDIT: does it mean you closed the GUI more than 28 times while the tour was running?

No, maybe 5 or 6 times. That tournament had 5 engines, so, 10 games per cycle, 20 games if we count pairs. So if I paused and closed the program once after say game 8, those 8 game pairs were affected. If I closed it after game 16, 4 game pairs were affected, and so on.

In the cases that the issue does appear, it affects all the pairs it can affect, meaning all the pairs from which only the first game has been played.

[mar]

I uploaded the PGN file in the same link. Btw I used very short time control to test, but it happened with larger time controls too.

I didn't do any of that. Yes, it was frustrating to try and trigger. It seems so random.

No, maybe 5 or 6 times. That tournament had 5 engines, so, 10 games per cycle, 20 games if we count pairs. So if I paused and closed the program once after say game 8, those 8 game pairs were affected. If I closed it after game 16, 4 game pairs were affected, and so on.

In the cases that the issue does appear, it affects all the pairs it can affect, meaning all the pairs from which only the first game has been played.

I see, thanks for the PGN and clarification, I'll try to trigger a repro automatically.
it'll postpone next build because I really want this annoying bug fixed...

as a bonus, next build will include (optional, default off) navigation bar below the board that will allow to use buttons
to navigate the move list, multiple people have asked for something like this already. while I was initially against it,
it was trivial enough to implement and since it can be toggled, no problem either way

[vencenda]

as a bonus, next build will include (optional, default off) navigation bar below the board that will allow to use buttons
to navigate the move list, multiple people have asked for something like this already. while I was initially against it,
it was trivial enough to implement and since it can be toggled, no problem either way

This is amazing, thanks!