-
-
Notifications
You must be signed in to change notification settings - Fork 5.4k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
v1.9 rc1 significant slowdown in garbage collection #49120
Comments
Is this a mac? If so this is a known issue that's fixed in RC2. |
Might be #48923 |
No, it is on Linux unfortunately. Also issue remains on nightly 703b3f8. |
The number of loaded packages is small, so I don't think #48923 could be affecting this benchmark. FWIW, we are doing a lot more collections on 1.9:
|
Also:
|
(as the label indicates) a bisection would be extremely helpful here. Edit: I am doing one. |
It looks like we used to be willing to let the collections grow exponentially in size, while now we cap them at some fixed rate |
Bisected to #44215. The title of that PR and the behaviour here seem to be at odds with each other... |
A local profile shows a bunch of time spent doing the in_image check so I'm not so sure it's not just that. Did the bisection already show a big regression there? |
There might have been two separate regressions. But IIRC #44215 was one of the big ones. And just checking |
I didn't really notice any improvement here with the pkgimage bisection GC PR. |
The GC PRs should only make this slightly better because they make the GC faster, but the issue here is why the hell are we doing so many small GCs |
I'm a bit worried about how much code was deleted in #44215. Was all that really irrelevant? |
Before #44215, the
After:
Notice how before the |
Looking into this I see a bug where we weren't increasing the interval at all on these sweeps, and that I can fix, but since the interval was based on live bytes after sweeping we weren't increasing it enough. You can see that on 1.8 we almost double it per iteration, and even after fixing it we weren't getting that, so I'm looking a bit into what's going on. |
Can we close this? It's basically fixed. If fixing #49205 is simple then maybe we add it in a patch release |
I have a significant slowdown for some code in Julia 1.9 rc1 which I could track down to garbage collection during filling of arrays. There seems to be something fishy with v1.9 rc1.
Below is some reduced code to reproduce the issue. The garbage collection is here significantly slower for Julia 1.9 compared to previous versions.
Total runtime for ´julia test.jl' on my machine is
I saw some GC issues in #48923 , not sure if this is related. If so, let me know and I will close this.
The text was updated successfully, but these errors were encountered: