-
Notifications
You must be signed in to change notification settings - Fork 16
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
Goroutine creation in FastCommit() #267
Comments
@SaveTheRbtz Thanks for opening this issue and also PR #268.
I think each slab storage instance isn't reused for multiple transaction/script executions. Unless I'm mistaken or things changed:
For these reasons, I don't think there's much impact or ROI to be gained by adding a pool to each slab storage instance. Maybe we can evaluate adding a global worker pool of goroutines to atree. Thoughts? @SaveTheRbtz @turbolent @ramtinms |
Good point about slab storage lifetime (should've checked it before suggesting a solution). If cached deltas are common, then we should consider adding an injectable worker pool to atree. I'll take a look at the execution node profiles and come back here with more info. |
Good idea! Yes, Faye's points describing the current usage of atree in Cadence are still correct. But it might still make sense to have a worker pool here, either by defining it in atree for all users; or allowing it to be passed in by users, allowing it to be reused (better) |
Issue To Be Solved
Currently,
PersistentSlabStorage.FastCommit()
does create adhoc goroutines, which may add a substantial overhead on servers with a high number of cores.Suggested Solution
In the ideal world, each slab storage instance can have a goroutine pool of encoders ready to encode a set of keys. As a bonus, it would also be possible to reuse encoders, instead of creating a new one each time.
The text was updated successfully, but these errors were encountered: