cassandra
cassandra copied to clipboard
apache
CASSANDRA-18753: Provide cassandra_latest.yaml with suitable defaults for new users
The first commit renames cassandra.yaml to cassandra_latest.yaml to make it easier to see the differences, the next commit restores the main yaml, and the third one configures testing.
Given test-latest will test tries+oa we could remove all oa test specific jobs to spare CI cycles as they're redundant now?
Also shouldn't this ticket also enable all dtests, both python and jvm, with latest?
Given test-latest will test tries+oa we could remove all oa test specific jobs to spare CI cycles as they're redundant now?
Actually, it will not test the the big-oa format but the bti-da one.
As for dtests, I'm still working on making the relevant changes.
Your 'latest' does already run with SCM NONE which is the same as running with oa feature-wise. Do we see value in testing oa by itself?
What I am trying to explain is that latest will already exercise the same code paths as oa...
big-oa is a separate sstable format, and it needs to be tested. Doing only base junit tests with it is likely sufficient; we can skip dtests (python or in-jvm) on it.
As latest will dtest oa current features that sounds like a good compromise to me, agreed.
are we (and how) dtesting this config too ? (java and python dtests), separate ticket ?
^ https://github.com/apache/cassandra/pull/2896#issuecomment-1812084239
As for dtests, I'm still working on making the relevant changes.
@blambov I have gone through it and you see we have only minors. I would suggest a rebase + CI runs where I will check the new circle setup. Doing that trying to follow the config.yaml changes, which I did but only cursory, would be eye watering...
Well, the failures appear to be as expected.
We do have a few outstanding problems that we never tested for before.
CI workflows LGTM +1 on the overall approach and the PR. Disclaimer I have no deep knowledge of all the options that have been enabled but overall SGTM. Commenting in case you'd like sbdy with deep expertise in some area to doublecheck things.
I suggest sending an email to the ML to find out what you mentioned to me offline: Are we ok merging this and surfacing a number of failures?