[Post Ray 2.3 Release] Update Ray versions to Ray 2.3.0

[cadedaniel]

Inspired by #822

I think I got all references to 2.2.0:

$ ag '2\.2\.0' | grep -v 'third_party' | grep -v 'go.sum' | grep -v 'go.mod'
CHANGELOG.md:17:* Critical stability improvements to the [Ray Autoscaler integration](https://ray-project.github.io/kuberay/guidance/autoscaler/). (To benefit from these improvements, use KubeRay >=0.4.0 and Ray >=2.2.0.)

[architkulkarni]

@cadedaniel should we expect the tests to be passing at this point, or will they only pass after 2.3.0 is actually released?

[architkulkarni]

Ah looks like the 2.3.0 images already exist, so I guess the tests should pass.

[DmitriGekhtman]

Let's also drop the stage for Ray 2.0.0 -- current strategy is to guarantee compatibility with the last three 2.x and one 1.x

[Jeffwan]

agree. too many ci jobs now.

[cadedaniel]

Done

[DmitriGekhtman]

Ah looks like the 2.3.0 images already exist, so I guess the tests should pass.

It looks like some issues with tests of GCS HA functionality, which are historically flakey.

[kevin85421]

Currently, "Compatibility Test - 2.3.0" and "Compatibility Test - Nightly" will always fail due to #852 (OOM). I will try to run the tests on my devbox to make sure the compatibility tests pass.

[kevin85421]

Update:

I ran the compatibility tests on my devbox, and I still got the same error. The Ray fault tolerance in release 2.3 is incompatible with KubeRay. I will work on it immediately.

[kevin85421]

This issue needs to be fixed by the Ray community. I opened an issue to track the progress #927. Happy to pair programming to help reproduce the bug.

@cadedaniel

[kevin85421]

@cadedaniel #929 is merged. Would you mind rebasing the master branch?

[cadedaniel]

Rebased and removed 2.0.0 compat test. This should be ready for merge.

[cadedaniel]

@kevin85421 hey can you take on fixing this test? I looked at the failure log but it's not clear to me what is the problem

[kevin85421]

LGTM. I reran the tests and they passed. The exit code is 137, and it may be related to the insufficient computing resources in the GitHub Actions free runner.