Song select performance really bad after spectating in a multi lobby

[bytenekos]

Type

Performance

Bug description

Stutters when pressing f2 or causing big jumps in beatmap selection (as in going from one end of the beatmap carousel to another) after going into spectator in a multi lobby and exiting

Screenshots or videos

Toolbox.2024.07.25.-.22.39.02.32.mp4

https://discord.com/channels/188630481301012481/1097318920991559880/1266115025031004181
https://discord.com/channels/188630481301012481/1097318920991559880/1266112819791925391

Version

2024.726.0-lazer

Logs

compressed-logs.zip

[Ladwick]

I was about to make an issue for this too, I didn't know exactly it was after spectating but I noticed most times after a multi I would get horrid performance at song select, specially when hitting the random button.

[peppy]

I can't reproduce anything here, and the graphs you are showing aren't really crippling slowdowns either.

I've checked for leaks from multiplayer spectator and cannot find any. I've profiled performance locally and can't get a slowdown large enough to investigate further.

[bytenekos]

I've done a bit more testing and I find it interesting how after exiting spectator and scrolling through beatmaps the gc count (in the ctrl + f2 menu) goes up drastically. No idea if that's at all relevant or if there's maybe something else that's interesting but that's what I noticed at least

Video before entering spectator (fresh start)

Osu.2024.07.30.-.23.23.38.20.mp4

Video after entering spectator (same session)

Osu.2024.07.30.-.23.25.00.21.mp4

Was also getting these kinds of messages when trying to exit the multi lobby:

Some logs with performance-xyz.log stuff:

logs.zip

If you have anything that you'd like me to try out to get more info/have any questions about please lmk because I just don't know what specific info I should provide in cases like these

[peppy]

Once you're in the bad state, can you create a memory dump from task manager (right click on the osu! process) then email it to me? You'll probably need to zip and upload to google drive as it may be quite big.

[bytenekos]

Should be sent now 👌
I've attached the issue number in the subject line

[bytenekos]

One observation I've noticed is that if you don't have a lot of maps, the stuttering subsides after a few "rounds" around the beatmap carousel and the stutter isn't that bad to begin with.
If you have a lot of beatmaps, performance slows down to a crawl (as mentioned in the original issue message).

Osu.2024.08.27.-.12.56.39.03.webm

[peppy]

There's a fix which will be included in the next release. Please check whether it fixes your issue. If not, please leave a reply here.

[bytenekos]

Tried out the current master (a09c6e2) and the issue is still there :(
Exact same behavior as my last comment

[Joehuu]

Can confirm, reopening.

[bdach]

Were you both testing in debug? I'm not sure that is representative of anything if yes.

[bytenekos]

Tested with release and no difference at all. Release, debug and the current release (817.0) have the same behavior.

[peppy]

I've been trying to repro this with a huge beatmap database and still can't...

Confirmation that everyone that is reproing this is on windows?

[bytenekos]

I'm on windows 11 23h2

[Joehuu]

Also initially tested on Windows 11, but I can also reproduce it on macOS.

I've been trying to repro this with a huge beatmap database and still can't...

That doesn't matter, I think. I only have 744 matches (beatmap difficulties), and can only reproduce if I have the files. Nuking the files, I can't seem to reproduce.

[peppy]

@Joehuu and it's enough to just enter the multiplayer listing once and exit out?

[Joehuu]

No, you have to be in MultiSpectatorScreen and wait until the MultiSpectatorPlayerLoaders push to MultiSpectatorPlayers (i.e. when you see gameplay in spectator). Also want to point out SoloSpectatorPlayer is unaffected by this.

[peppy]

Thanks

[peppy]

Able to reproduce. Definitely GC related.

Cycling carousel normally:

Cycling carousel after multiplayer spectator:

Comparing total genX sizes and object counts shows a large but not crippling difference (68.56mb -> 101.90mb managed / 553k -> 824k live objects). First thought was audio adjustments (each screen has muting stuff) so I attempted disabling that but it's still performing horribly.

Will require some further investigation.

[peppy]

@Joehuu can you try updating your .NET and reproducing on the latest HEAD? I can't seem to repro anymore...

[peppy]

Never mind, fix incoming.