FANNG1 commented on PR #11060:
URL: https://github.com/apache/gravitino/pull/11060#issuecomment-4456530942

   Thanks for adding the matrix runners and documenting how to re-verify the 
Lance compatibility ranges. I think the CI strategy here needs to be called out 
explicitly before we merge this.
   
   My concern is that the full compatibility matrix should probably not become 
part of the default PR validation path. These tests are expensive, depend on 
external package ecosystems, and can become noisy when PyPI / Maven resolution 
changes upstream. That makes them a poor fit for the main per-PR signal.
   
   What seems more sustainable is a two-layer approach:
   1. Keep one pinned, known-good Lance combination in the normal CI path as 
the regression guard.
   2. Keep the full multi-version matrix as a separate automation path: 
scheduled workflow, release-time validation, and manual dispatch when we need 
to re-verify or extend the compatibility claims.
   
   That would still keep the matrix exercised continuously enough to avoid 
script rot, but without making every PR pay the cost or absorb upstream 
ecosystem flakiness. If we go this direction, I also think the matrix runner 
should avoid floating dependencies by default (for example, not resolving 
`lance-namespace` as unpinned `latest`).
   
   Can we align on that execution model in this PR? Right now the tasks and 
docs are being added, but the intended automation boundary is still unclear.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to