Me and Marianne finished running all the tests for Paxos V2 with the bug
fixes from CASSANDRA-18086 and the results for the 100 partitions are
definitely better than V1. For the larger partitions (1000, 1), the
results for V2 are comparable and overall V2 did not have any performance
regression
Me and Marianne are also still chasing a performance issue with Paxos v2
when compared with v1. We
see way more contention on v2 for a LOCAL_SERIALIZABLE workload that
writes/reads to only 100
partitions (v2 performs better for higher partition counts). We're still
investigating what's going
on.
S
f they dont
> realize that's what they want to happen.
>
>
>
> On Tue, Aug 10, 2021 at 12:53 PM Matt Fleming
> wrote:
>
> > Hi,
> >
> > With the way that hint transfer currently works during decommission,
> > it's possible to leave hints on the
Hi,
With the way that hint transfer currently works during decommission,
it's possible to leave hints on the disk of the decommissioning node if
those hints are generated after the decommission begins.
I'd like to discuss automatically closing the native transport port
before hints are transferre
