Level / classic-level

An abstract-level database backed by LevelDB.
MIT License
58 stars 11 forks source link

Breaking: bump `abstract-level` to 2.0.0 #52

Open vweevers opened 1 year ago

vweevers commented 1 year ago

TODO:

To be rebased (not fully squashed) before merge.

ishfx commented 6 months ago

Hey, when do you plan to merge/release v2? Looking forward to it, thanks!

jacoscaz commented 5 months ago

@vweevers would you like some help with benchmarking?

vweevers commented 5 months ago

Help is always welcome, thank you! Though FYI, that's not why I haven't released this yet. Just haven't had the time, and I don't consider benchmarks to be a blocker (because e.g. promises are unlikely to make a big difference here) so I planned to just skip them when I got to it.

The only thing that slightly worries me is that db.get() now creates a snapshot synchronously. I don't know if that's a fast operation for LevelDB.

jacoscaz commented 5 months ago

@vweevers no pressure meant or implied. I depend on level packages through quadstore and I'd be happy to help an upstream dependency if I can find the time.

Also, my experience with promise-based iteration is exactly the opposite; so far I've always found it to make a rather big difference, particularly with asynchronous sources capable of reading and returning multiple items in batches and particularly when structuring chains of iterators. That said, I also recognize that I'm not as experienced as you are. I'll have a look next week!

vweevers commented 5 months ago

I'm not that experienced with promises themselves, I meant more that disk and the C<>JS barrier are the bigger bottlenecks here. But it's always good to challenge such assumptions. You're probably right that there's an overhead to e.g. async iterators.

For people that were already using promises, I do hope that abstract-level v2 and classic-level v2 are faster because they no longer have to translate callbacks into promises, allowing V8 to optimize it.