[#580] chore: move deploy/kubernetes to a standalone workflow

[kaijchen]

What changes were proposed in this pull request?

1. Move kubernetes profile from parallel.yml to a standalone workflow.
2. Speed up kubernetes workflow by skipping java tests.

Why are the changes needed?

Sub-task of #580

1. Cleanup regular CI parallel.yml.
2. Speed up kubernetes CI.

Does this PR introduce any user-facing change?

No.

How was this patch tested?

CI.

[codecov-commenter]

Codecov Report

Merging #578 (17cb26a) into master (6692516) will decrease coverage by 12.33%.
The diff coverage is n/a.

@@              Coverage Diff              @@
##             master     #578       +/-   ##
=============================================
- Coverage     61.13%   48.80%   -12.33%     
=============================================
  Files           213       14      -199     
  Lines         12328     1971    -10357     
  Branches       1041        0     -1041     
=============================================
- Hits           7537      962     -6575     
+ Misses         4382      949     -3433     
+ Partials        409       60      -349     

Impacted Files	Coverage Δ
...org/apache/uniffle/server/LocalStorageChecker.java
.../java/org/apache/uniffle/common/util/RssUtils.java
...org/apache/uniffle/common/metrics/GRPCMetrics.java
...niffle/client/response/CompressedShuffleBlock.java
...niffle/common/exception/FileNotFoundException.java
...niffle/coordinator/strategy/storage/RankValue.java
...k/shuffle/writer/WrappedByteArrayOutputStream.java
...java/org/apache/uniffle/common/util/ExitUtils.java
...apache/uniffle/storage/common/ShuffleFileInfo.java
...niffle/common/rpc/MonitoringServerInterceptor.java
... and 189 more

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[advancedxy]

Seems like other module is still compiled . it is possible to run build-operator.sh and test-operator.sh directly.
You can use mvn package -P kubernetes -am -pl deploy/kubernetes to just build the operator package.
And there's no need to skipTests here.

[advancedxy]

Anyway, many thanks for the work. I thought about it a while ago.

[jerqi]

Seems like other module is still compiled . it is possible to run build-operator.sh and test-operator.sh directly. You can use mvn package -P kubernetes -am -pl deploy/kubernetes to just build the operator package. And there's no need to skipTests here.

We need other jars to run K8S integration tests, so we should compile them.

[advancedxy]

Seems like other module is still compiled . it is possible to run build-operator.sh and test-operator.sh directly. You can use mvn package -P kubernetes -am -pl deploy/kubernetes to just build the operator package. And there's no need to skipTests here.

We need other jars to run K8S integration tests, so we should compile them.

Currently, there's no jars needed to run K8S tests? Even if we enable integration tests, we may relies the rss-server images to run integration tests?

[jerqi]

Seems like other module is still compiled . it is possible to run build-operator.sh and test-operator.sh directly. You can use mvn package -P kubernetes -am -pl deploy/kubernetes to just build the operator package. And there's no need to skipTests here.

We need other jars to run K8S integration tests, so we should compile them.

Currently, there's no jars needed to run K8S tests? Even if we enable integration tests, we may relies the rss-server images to run integration tests?

We should use latest jars to produce image.

[advancedxy]

We should use latest jars to produce image.

you mean deploy/kubernetes/docker/Dockerfile? I think it's not used in CI.

[jerqi]

We should use latest jars to produce image.

you mean deploy/kubernetes/docker/Dockerfile? I think it's not used in CI.

Yes. why don't we use it in IT?

[advancedxy]

We should use latest jars to produce image.

you mean deploy/kubernetes/docker/Dockerfile? I think it's not used in CI.

Yes. why don't we use it in IT?

We just don't docker build this Dockerfile yet.

[jerqi]

We should use latest jars to produce image.

you mean deploy/kubernetes/docker/Dockerfile? I think it's not used in CI.

Yes. why don't we use it in IT?

We just don't docker build this Dockerfile yet.

I mean that we will use it if we have IT tests.

[advancedxy]

We should use latest jars to produce image.

you mean deploy/kubernetes/docker/Dockerfile? I think it's not used in CI.

Yes. why don't we use it in IT?

We just don't docker build this Dockerfile yet.

I mean that we will use it if we have IT tests.

Ah, yes. But I think in most cases, we will just reuse the already existed docker image or put the shuffler server docker image
build process parallel with rss-controller/rss-webhook build process.

When we start integration tests, the ci workflow will need some refactor work to support it.

[kaijchen]

@jerqi @advancedxy Let's just create a new CI workflow for deploy and k8s tests later.
We can add -DskipUTs -DskipITs in the new workflow, and remove it from the kubernetes profile.

The parallel.yml should be used just for integration with different compute engines.

[advancedxy]

@jerqi @advancedxy Let's just create a new CI workflow for deploy and k8s tests later. We can add -DskipUTs -DskipITs in the new workflow, and remove it from the kubernetes profile.

The parallel.yml should be used just for integration with different compute engines.

fair point..

[jerqi]

We should use latest jars to produce image.

you mean deploy/kubernetes/docker/Dockerfile? I think it's not used in CI.

Yes. why don't we use it in IT?

We just don't docker build this Dockerfile yet.

I mean that we will use it if we have IT tests.

Ah, yes. But I think in most cases, we will just reuse the already existed docker image or put the shuffler server docker image build process parallel with rss-controller/rss-webhook build process.

When we start integration tests, the ci workflow will need some refactor work to support it.

It's ok.

[kaijchen]

@jerqi @advancedxy I've added a standalone workflow. How do you think?

[jerqi]

It's ok for me.

[advancedxy]

@jerqi @advancedxy I've added a standalone workflow. How do you think?

Sounds good to me.

[advancedxy]

for this case, I think it should be more related to the k8s profile?

For k8s, it doesn't need to run maven with uts..
however there could be other deploy options.

[advancedxy]

Unless you call it k8s_deploy.yml ?

[kaijchen]

There might be deploy/YARN in the future. Let's keep it as is now?
If you wish to not run it with maven, feel free to change it later.

[advancedxy]

There might be deploy/YARN in the future. Let's keep it as is now? If you wish to not run it with maven, feel free to change it later.

I like the idea that we can simply call maven to build all the stuff. I am not intended to run without mvn.
I'am thinking about this case.

# for k8s
mvn -Pk8s -DskipUTs -DskipITs

# for yarn, it should be without -DskipUTs.
mvn -Pyarn

[advancedxy]

maybe you should use this trick to put maven options in the matrix side. https://stackoverflow.com/questions/66025220/paired-values-in-github-actions-matrix

[kaijchen]

# for yarn, it should be without -DskipUTs.
mvn -Pyarn

I don't want to support YARN in this PR.
YARN will be supported when it's introduced (probably never).

[kaijchen]

I just want to do minimal changes to move k8s test out of parallel.yml.
And speed up k8s CI by skipping UT and IT.

[advancedxy]

I'm ok to move the -DskipUTs ... into the deploy.yml side.

[kaijchen]

Thanks @advancedxy and @jerqi for the discussion and review.