Open xhochy opened 3 years ago
For the sake of compatibility, I'm wondering if simplekv stores are still compatible or not. and how/if we can ensure this and manage the deprecation properly
For the sake of compatibility, I'm wondering if simplekv stores are still compatible or not. and how/if we can ensure this and manage the deprecation properly
We didn't change the API in any aspect, the move was solely a rename of the top-level namespace. Passing in an existing simplekv
store or a storefact
-URL will work without issues. Having simplekv
/ storefact
installed alongside minimalkv
also works.
Verified locally that https://github.com/data-engineering-collective/minimalkv/pull/24 fixes the documentation build issues. I'll retrigger the PR here once that has been released.
All tests passed except the one that is also failing on master. This is probably an issue with the latest pandas
nightly and unrelated to the simplekv
-> minimalkv
switch.
Figured out the issue we see in the tests confirming that it is totally unrelated to this PR:https://github.com/dask/dask/issues/7610
OK to merge?
Thank you @xhochy for the contribution and for looking into the broken nightly test. We are discussing a possible switch to minimalkv
internally, will get back to you.
@stephan-hesselmann-by Any update here or any information you would need? I did contact BY before starting minimalkv
but as I'm no longer part of it, I'm not sure I reached out to all relevant parties. Switching should work quite easily as this is designed with care so that you can mix & match simplekv
, minimalkv
and storefact
in one environment easily until the whole stack is migrated to minimalkv
.
If it helps with migration efforts we could think about implementing a proxy in kartothek (yet anotherget_store
) which would point to the "current" backend store lib. FWIW, I don't think this is really necessary as long as minimalkv doesn't change the class interfaces without announcement
Unfortunately I don't have a definitive answer yet. The issue for us is not only Kartothek but also other Blue Yonder artifacts that currently use simplekv
. That is also the reason for the indecisiveness.
@xhochy if we do decide to migrate to minimalkv
, would you be willing to grant Blue Yonder "co-ownerhsip"? We would want to avoid another simplekv
situation in the future and also make sure that the interface stays compatible with simplekv
as long as it is used by Blue Yonder.
Unfortunately I don't have a definitive answer yet. The issue for us is not only Kartothek but also other Blue Yonder artifacts that currently use
simplekv
. That is also the reason for the indecisiveness.
They can use simplekv
as a direct import and as a dependency, no need to change anything there. We already use minimalkv
together with Kartothek and have simplekv
and storefact
installed.
@xhochy if we do decide to migrate to
minimalkv
, would you be willing to grant Blue Yonder "co-ownerhsip"? We would want to avoid anothersimplekv
situation in the future and also make sure that the interface stays compatible withsimplekv
as long as it is used by Blue Yonder.
Happy to give access to people who contribute (see http://theapacheway.com/merit/), I don't want to hand over control to Blue Yonder just because they use it*. What would be your benefit / intention here exactly?
*You neither have full access to pandas
either?, or more concrete: I would also like to have access back to kartothek
because I use and advocate for it (also created it) but at the time of writing, I don't have that, putting me in the same situation. Thus for me this kind of argument is quite weird.
Note I gave @fjetter access to minimalkv
https://github.com/data-engineering-collective/minimalkv/pull/38 as he helped in the past and current time with it a bit. It may not be so relevant for you but that adds a QuantCo-independent maintainer to it.
simplekv
is de-factor not maintained and ownership is questionable. As we've seen with simplekv
, both ownership and question does also not necessarily connect with admin or write access.
minimalkv
is maintained and ownership is clearly defined. This PR removes therefore a high risk component which caused issues in the past with a better alternative.
The issue for us is not only Kartothek but also other Blue Yonder artifacts that currently use simplekv. That is also the reason for the indecisiveness.
Whether or not BY decides to "migrate" from simpelkv
to minimalkv
is irrelevant for this PR (although I can highly recommend). If you decide to use both, you are free to do so. The question for this PR would be more appropriately whether this change is breaking for the library kartothek
. As it stands right now, you can use both libraries but by default we'll install minimalkv
with this PR.
https://github.com/JDASoftwareGroup/kartothek/pull/376 implemented duck typing for the storage interface. as long as this stays, the choice for simplekv
or minimalkv
should not impact functionality
I agree with your points and also share some your frustrations with this situation. I will try to argue in favor of this PR but as you know I cannot guarantee success.
Are there any updates on this?
The original Maintainer of simplekv has agreed to add additional maintainers/hand over maintainership. @xhochy is one of the designated new maintainers, which should resolve the situation that lead to the fork.
I hope that once the formal changes are through, we can go back to just simplekv, this time with a bit more lively development!
I don't think that adding maintainers to the existing mbr/simplekv
repository fixes the root of the issue. We will still not be able to fully manage that without the help of mbr
.
Also, contrary to any mails, I have not yet received any rights over there.
As I understood, mbr would extend/transfer ownership and not keep a special role. You would have all the rights he and the other new maintainers have.
Description:
Since
simplekv
is unmaintained and we cannot add new maintainers, this PR moves to its successor projectminimalkv