Khepri

[Ed W]

Hi,
See about 43 mins in:
    https://braid.org/meeting-8

Kevin Jahns (the Yjs author) was also there and was impressed

I have been trying to find a few minutes free to work up my own implementation (there is also a
python implementation)

I don't think it works for every use case, but the lack of explicit tombstones seems interesting.
(arguably he uses nil values as a kind of tombstone). I've been eyeing the algorithm up for solving
an internal problem and when I saw your explicit version numbers for each node it looked extremely
similar to the shape needed to use this algorithm! I wonder if there is a possible scenario to make
an eventually-consistent data structure where you have lagging consistency over a (presumed) short
time interval.

Ed W


On 11/11/2021 10:15, Frank Muller wrote:
> Hi Ed,
>
> Any reference to Shelf (paper, Braid group link…)?
> Thanks
>
> /F.
>
>
>     Hi, I'm sorry, I joined the list too late to be able to grab message ids and reply specifically to
>     the parent post
>
>     I like the idea of what you have here! Very nice!
>
>
>     I wanted to bring to your attention something I saw recently on the Braid groups. There is a
>     rather
>     clever CDRT proposed called "shelf", details:
>
>         https://github.com/dglittle/shelf <https://github.com/dglittle/shelf>
>
>     It seems to me that with a little squinting, this basic architecture could be repurposed for an
>     eventual consistent version of Khepri
>
>
>     Now, I realise this is likely not of immediate interest to RabbitMQ as you are clearly looking for
>     an online only system. However, I wonder if it might solve some related interesting use cases?
>
>     Its elegant in that it's exceptionally simple, doesn't carry any tombstones and is eventually
>     consistent.
>
>     Thanks for listening
>
>     Ed W
>
>
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://erlang.org/pipermail/erlang-questions/attachments/20211111/b0cac803/attachment.htm>