Cache last published event #87
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
strohel
force-pushed
the
last-value-cache
branch
from
d5aed6b to
8aee172
Compare
|
I've pushed 4 commits to this, @goodhoko (and others) PTAL. It makes sense to review them one-by-one (and check their messages) if you're curious. It would be also interesting to run |
strohel
force-pushed
the
last-value-cache
branch
from
b336f5f to
789611c
Compare
…ro-sized-type The benchmark results are therefore not comparable before/after (they are 10% to 30% slower on my laptop).
and let subscribers receive it upon subscription. Closes #86
This fixes the performance regression for me, but we need to think whether that's worth the dependency tradeoff.
With a small trick (that unfortunately leaks to a public type) we were able to implement a kind of specialization [1] with perfect behavior: `Event`s don't pay the price for caching, and it is not possible to subscribe_and_receive_latest() to bare `Event`s. Also, it means that we didn't have to add bounds to `Event` [2]. [1] I thought the orphan rule would prevent us from adding the blanked implementation, but not! [2] I've sneaked in the removal of `'static`, but that's mostly aesthetic: `std::any::Any` requires it anyway.
...now that we have it in the dependency tree anyway. Allows us to remove the `RwLock` which in turn allows us to deduplicate subscribe_recipient() vs. subscribe_and_receive_latest(). Let's think about the possible situations of "publishing the first message" vs. "just subscribing and getting latest".
strohel
force-pushed
the
last-value-cache
branch
from
789611c to
148227a
Compare
|
@goodhoko @bschwind alright I've:
|
goodhoko
commented
Aug 8, 2024
There was a problem hiding this comment.
Nice work, @strohel! Thanks for taking this over, it's a pleasure see you work on something that excites you!
The benchmark results look good! If I read them right, this PR no longer causes noticeable performance regression. This is equally true for using dashmap and intelligent locking (i.e. upgrade()).
There's a significant (and expected) difference in performance of the cacheable event when there are multiple publishers between Dashmap and intelligent locking. IMO that's totally ok!
I gave this one more thorough read and it looks good, though I have couple last remarks.
|
@bschwind @goodhoko (sorry I meant to submit this right after the benchmark results Benchmark Observations
Conclusion
|
|
bschwind
approved these changes
Aug 9, 2024
goodhoko
commented
Aug 9, 2024
mbernat
reviewed
Aug 9, 2024
|
This was successfully tested in https://github.com/tonarino/portal/pull/3746. Should be good to merge! |
strohel
reviewed
Aug 13, 2024
Co-authored-by: Matěj Laitl <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
and let subscribers receive it upon subscription.
Optionally, we could provide a separate marker trait for cached events and persist only those. I couldn't figure out how to dynamically query for the type (trait) in
publish. Somehowdowncast_refis not implemented forEvent(despiteEventbeingAny). I'd appreciate some help with that.This causes performance regression for the pub-sub system. Here's output of
cargo bench --bench pub_subrunning on this branch (comparing it to the previous run onmain):Details
That's likely caused by the contention on
SystemHandle::event_subscribersRwLock. Previously it only needed exclusive access when subscribing and shared access when publishing. Now it needs exclusive access when subscribing and publishing.Closes #86
This is a semver-breaking change (due to the new bounds on Event).