[
https://issues.apache.org/jira/browse/CASSANDRA-17164?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17490198#comment-17490198
]
Benedict Elliott Smith commented on CASSANDRA-17164:
----------------------------------------------------
Quick update: the burn tests unfortunately shook out some remaining teething
issues with simulation of Paxos v2, that I am still working through. I have
incorporated [~bdeggleston]'s suggestions, with one minor change to the
refactor of {{sizeOfConsensusQuorum}}.
I can see [~blambov] is still making good progress on a very comprehensive
piece of documentation with semi-formal proofs. I haven't had time to properly
digest it but it looks great.
One comment I would perhaps make, more generally than specifically to modify
this documentation, is that I think it would be quite possible to use
Multi-Paxos without burdensome state maintenance (e.g. by using token ranges
for leader election, perhaps with per-key leaders in some rare circumstances).
This would have trade-offs, however, in that the fast-read optimisation would
be excluded and we would have heterogenous read performance between regions.
This is probably a desirable trade-off particularly for write-heavy workloads
and the _main_ reason we haven't done this is because we wanted to pursue
_Accord_ and its superior semantics, as multi-key transactions are highly
desirable, as are global 1RT reads (and global 1RT writes under realistic
scenarios).
It was my belief that pursuing Multi-Paxos in the near term would be a
mis-investment of project resources. I think it is entirely possible we may
loop back to provide Multi-Paxos eventually, as I do not think it is a hugely
challenging proposition in itself and there may be some workloads that prefer
it even once _Accord_ is delivered. The main difficulty would be the additional
logic for routing operations to the correct command leader (perhaps involving
client integration) and robust failure detection (currently much too poor in
Cassandra). If somebody else wants to pursue this, I would be more than happy
to help shepherd it, as it is certainly achievable.
> CEP-14: Paxos Improvements
> --------------------------
>
> Key: CASSANDRA-17164
> URL: https://issues.apache.org/jira/browse/CASSANDRA-17164
> Project: Cassandra
> Issue Type: Improvement
> Components: Consistency/Coordination, Consistency/Repair
> Reporter: Benedict Elliott Smith
> Assignee: Benedict Elliott Smith
> Priority: Normal
> Fix For: 4.1
>
>
> This ticket encompasses work for [CEP-14|
> https://cwiki.apache.org/confluence/display/CASSANDRA/CEP-14%3A+Paxos+Improvements].
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org
