kamille
kamille
> I attempted to address this issue through OpenDAL's API design but ultimately found it unfeasible, so I implemented a workaround using task notifications instead. The concurrent writing occurs within...
It looks good to me. I just helped to sort out codes, mainly move the compaction client to runner crate. I am checking it in my local, let's merge it...
Thanks @LeslieKid
I am working on a poc #12947 for checking the improvement of this idea.
I found it actually now a trivial change. The challenge is that, we should refactor codes to support vectorize append firstly. (otherwise, we need to add a branch about which...
This is the number get from my local: ``` ┏━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓ ┃ Query ┃ main ┃ vectorize-append-value ┃ Change ┃ ┡━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩ │ QQuery 0 │ 0.69ms │ 0.67ms │ no change...
> Makes sense to vectorize appends at the same time, I think this might have even more impact than just omitting null checks. Removing the null check, but adding another...
finished in #12996
For simplificity, I think we can just setup a standalone horaedb in test which acts as both `remote compaction caller` and `remote compaction server`?
🤔 As reviewing #12697 , seems we can still continue to improve partial skipping? Now we can modify threshold to get performance improvement, but it may be a bit tricky?...