Amending the CEP with the proposed addendum seems to me like a reasonable compromise to de-escalate this matter and move forward, addressing potential concerns without any prejudice to the original goals of the CEP.
Em sex., 15 de out. de 2021 às 15:11, Jonathan Ellis <[email protected]> escreveu: > Hi all, > > We have had several discussions today as to how to move forward on CEP-15, > given that the first vote was vetoed by myself and Mick. From my side the > concern has been that the distributed transactions design space inherently > requires tradeoffs; Accord represents one set of those tradeoffs but I want > to make sure that what we do now makes it easier to add other > implementations representing other points in that design space, rather than > tightly coupling us to just one option as we have been to date. > > I do think Accord is an interesting proposal that advances the state of the > art in material ways. As Mick has pointed out, my veto was not intended to > block it as such, but to make sure that we spend the time necessary to > understand how it might fit into a longer term roadmap beyond the scope of > CEP-15 itself. > > It was my assumption that we could afford to continue such discussions to > get further clarity while the Accord team continues to improve their > prototype. However, I've learned today via some background discussions that > not approving the CEP in fact blocks the team behind it from fully > committing to this work. > > That's unfortunate, and I suspect frustrating. To unblock things, I think > we can move forward if we can add the following language to the CEP, under > "Long Term". (Some degree of pluggability is already implied by the goal > of replacing LWT, but this is worth making explicit.) > > *This work shall be developed in a modular manner, to allow for coexistence > with other consensus protocols or transaction managers. This will allow us > to evolve Accord without precluding alternative solutions, as future work > expands Cassandra's transactional capabilities beyond the goals of this > CEP. Initially, supporting the Paxos-based LWT and Accord side by side is > also an example of such modularity and optionality.* > > (For completeness, I note that explicitly adding pluggability as a > requirement means that it is no longer necessary to also add a LOCAL_SERIAL > option to Accord itself, although that is of course still an option.) > > -- > Jonathan Ellis > co-founder, http://www.datastax.com > @spyced >
