Hi,
Thanks for pointing that out Yuqi. I'm going to work on merging 20158 and 20164
next.
Since there are no objections I am going to assume lazy consensus and merge
back to 5.0.
Ariel
On Thu, Feb 27, 2025, at 2:44 PM, Yuqi Yan wrote:
> Thanks Ariel for bringing this to the dev mail list.
>
Thanks Ariel for bringing this to the dev mail list.
I want to add few more notes for 20158 and 20164:
> The only problem with this approach is that the tree is not rebalanced
Actually 20158 won't build an imbalanced tree - to replace an interval in
the tree, the before and after must have the sam
I’ve encountered a handful of spinning platters, but not a lot.
I think we should generally optimize for the common case, not the
exception.
On Thu, Feb 27, 2025 at 9:51 AM Josh McKenzie wrote:
> This is a significant enough performance problem *in normal operations*
> I'd consider it a bug an
This is a significant enough performance problem *in normal operations* I'd
consider it a bug and thus eligible for back-porting. A couple other thoughts:
> CASSANDRA-20158 and CASSANDRA-20164 are several orders of magnitudes faster
> ... The only problem with this approach is that the tree is n
Hi,
I want to discuss what versions we should backport IntervalTree improvements to
specifically 19596 which I think is the lower risk option because it builds the
same trees as before. I think we should at least backport to 5.0.
IntervalTree performance has shown up as a problematic bottleneck
