Skip to content
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

Optimize resolver with threadlocal caches #156

Merged
merged 6 commits into from
Oct 28, 2024

Conversation

yamadapc
Copy link
Contributor

@yamadapc yamadapc commented Oct 27, 2024

On #10 we have removed dashmap from this project dependencies.

The reason for that is we were observing an alarming amount of crashes
related to memory bugs on that crate on developer machines. We have
confirmed that those bugs were in the dashmap crate and they stopped
happening after it was removed.

However, the solution put in place made resolution around 3-5x slower
than what it was before with dashmap caches. That is due to contention
on the RwLock hashmap used for resolver internal caches.

This change moves those caches to use a thread-local map, which achieves
close performance to the dashmap solution, around 30% slower.

The entire resolution task can be completed very quickly on a large
project (few seconds), hence we prefer to have a more stable
implementation for now.

Motivation

  • Improve performance of the resolver

Changes

Move resolver internal caches to use thread-local storage.

Checklist

  • Existing or new tests cover this change

On #10 we have removed `dashmap` from this project dependencies.

The reason for that is we were observing an alarming amount of crashes
related to memory bugs on that crate on developer machines. We have
confirmed that those bugs were in the `dashmap` crate and they stopped
happening after it was removed.

However, the solution put in place made resolution around 3-5x slower
than what it was before with `dashmap` caches. That is due to contention
on the `RwLock` hashmap used for resolver internal caches.

This change moves those caches to use a thread-local map, which achieves
close performance to the `dashmap` solution, around 30% slower.

The entire resolution task can be completed very quickly on a large
project (few seconds), hence we prefer to have a more stable
implementation for now.
@yamadapc yamadapc requested a review from a team October 27, 2024 23:34
let read = cache.read().unwrap();
let cached = read.get(&ret).cloned();
drop(read);
let cached = cache.get(&ret);
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is a lot nicer now 👍

@yamadapc yamadapc enabled auto-merge (squash) October 28, 2024 00:31
@yamadapc yamadapc disabled auto-merge October 28, 2024 01:24
@yamadapc yamadapc merged commit d635462 into main Oct 28, 2024
17 checks passed
@yamadapc yamadapc deleted the optimize-resolver-with-threadlocal-caches branch October 28, 2024 02:05
Nikola-3 pushed a commit that referenced this pull request Oct 29, 2024
@yamadapc yamadapc mentioned this pull request Nov 10, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants