Ideally the downlevel tests would be a matrix of supported versions.

[shankarseal]

          Ideally the downlevel tests would be a matrix of supported versions.

Originally posted by @mtfriesen in #3421 (comment)

[saxena-anurag]

Since the selfhosted runners are limited, should we instead use github runners to run kernel regression / downlevel tests? The only drawback will be that we will not get kernel dumps though, but will allow us to scale.

[saxena-anurag]

Along with the matrix of down-level versions, we can also add permutations of tests for each version:

1. both extensions and native modules from older version -- tested in selfhosted runners.
2. only extensions from older version -- tested in GitHub runner.
3. only native modules from older version -- tested in GitHub runner.