On 2023/07/08 08:51:48 Roman Shaposhnik wrote:
On Sat, Jul 8, 2023 at 3:08 AM Joseph M Hellerstein
<he...@berkeley.edu> wrote:
>
Agreed! That said, I'm still curious about things like Neon, simply b/c
they do seem to care about ML use case as one of their differentiator
for thin-cloning: https://neon.tech/docs/introduction/branching#machine-learning
But I guess I just need to ask Heikki ;-)
And he just did, by a private mail. Hi everyone! :-)
> The challenge is that, to my knowledge, none of them support arbitrary C
> UDFs. So I'd explore both business and technical solutions to that. On the
> business side, one could try to convince vendors to anoint MADlib as an
> approved extension. That would likely require some customer advocacy. On
> the technical side, the goal would be to get MADlib running using only
> approved UDF frameworks. I can see various approaches there that seem
> feasible.
That's a good point actually! A bit of BD activity would really help...
That is on our roadmap for Neon, we want to let users to "bring their
own extension". It's still work-in-progress though; we need to design
the API and UI for users to manage their custom extensions, and figure
out how we can inject them into the docker images we run.
If I remember correctly, MADLib doesn't do anything low-level with
storage, so it should work fine on Neon. We do already bundle a bunch of
extensions in our standard docker images, I'll check what it would take
to include MADlib too.
--
Heikki Linnakangas
Neon (https://neon.tech)
