ZEPPELIN-5543 Use TravisCI for testing on Linux ARM64

[martin-g]

What is this PR for?

Build and test on Linux ARM64 at TravisCI.
It uses Partner Queue Solution at Equinix Metal ARM64 nodes, so the jobs do not consume build credits!

The PR is based on the latest version of .travis.yml from branch-0.9 and the Miniconda3 with Python 3.8 with the packages listed in testing/env_python_3.8_with_R.yml

What type of PR is it?

Improvement

Todos

• - Confirm that the build passes after the merge of ZEPPELIN-5542 Use headless Firefox instead of PhantomJS #4237. Currently most of the jobs fail due to the usage of PhantomJS
• - Debug an issue in the PySpark jobs :

testSharedInterpreter(org.apache.zeppelin.livy.LivyInterpreterIT) Time elapsed: 26.146 sec <<< FAILURE!
java.lang.AssertionError: %text No module named matplotlib.pyplot
Traceback (most recent call last):
ImportError: No module named matplotlib.pyplot
expected:<SUCCESS> but was:<ERROR>
at org.junit.Assert.fail(Assert.java:88)
at org.junit.Assert.failNotEquals(Assert.java:834)
at org.junit.Assert.assertEquals(Assert.java:118)
at org.apache.zeppelin.livy.LivyInterpreterIT.testSharedInterpreter(LivyInterpreterIT.java:807)
INFO [2021-10-05 09:37:15,548] ({Thread-0} Logging.scala[info]:39) - Shutting down cluster pool.
Results :
Failed tests: LivyInterpreterIT.testSharedInterpreter:807 %text No module named matplotlib.pyplot
Traceback (most recent call last):
ImportError: No module named matplotlib.pyplot
expected:<SUCCESS> but was:<ERROR>

What is the Jira issue?

• https://issues.apache.org/jira/browse/ZEPPELIN-5543

How should this be tested?

• Make sure that the build at TravisCI passes!

Questions:

• Does the licenses files need update? - NO
• Is there breaking changes for older versions? - NO
• Does this needs documentation? - Update the How to contribute documents to mention that Linux ARM64 is being tested

[zjffdu]

#4237 is merged, @martin-g Could you rebase this PR ?

[martin-g]

@zjffdu I've rebased the branch but I don't see the TravisCI check. Should it be enabled manually via GitHub UI ?!

[zjffdu]

@martin-g Have you tried use travis CI for your forked zeppelin repo first ?
https://docs.travis-ci.com/user/tutorial/#to-get-started-with-travis-ci-using-github

[martin-g]

Yes! It works fine on my fork!
Let me investigate!

[martin-g]

Apache Infra team enabled it: https://issues.apache.org/jira/browse/INFRA-22411

[martin-g]

I've started working few times on this PR but the build is very unstable on Travis.
Part of the reason is also that I don't quite get the build setup. As a Maven project I expect that mvn clean install should do everything that is needed. But it fails because some Maven modules cannot find other modules. One has to mvn install -DskipTests first and then execute mvn test in a second pass.

Last Thursday there was a meeting with Apache Infra where we discussed how we could use Linux ARM64 build nodes for Github Actions CI. The idea is to use self-hosted Linux ARM64 runner + custom Github Action runner with security enhancements to prevent abuses from contributors. Also see https://cwiki.apache.org/confluence/display/INFRA/GitHub+-+self-hosted+runners
This way Zeppelin will continue using just Github Actions as a CI.
I'll let you know once I have a working POC!

[zjffdu]

Thanks @martin-g

[martin-g]

@zjffdu I've documented the approach at https://martin-grigorov.medium.com/github-actions-arm64-runner-on-oracle-cloud-a77cdf7a325a
If you are interested in this solution then you will need to contact Apache Infra for the token needed by ./config.sh and to setup the security/approvals. Everything else could be done by any member of the Zeppelin team!

[martin-g]

@zjffdu Did you have a chance to take a look at the GitHub Actions self-hosted runners on Oracle Cloud approach ?

[zjffdu]

Sorry for late response @martin-g , this approach LGTM, have you tried to build Zeppelin in oracle cloud?

[martin-g]

I haven't tried yet!
Before I do it I want to explain something that was a stopper for another project:

When creating an account at Oracle Cloud (OCI) they will ask you for a credit card details. It won't be charged if you use the Free Tier (and this is enough for testing on ARM64)! But if you want to share this OCI account with the rest of the Zeppelin team then this might be problematic, because someone may add a paid resource to the account and OCI will start charging your card!

If the above is a problem for Zeppelin too then I could offer you a VM donated by my employer (OpenLab Testing) that could be used the same way!

[martin-g]

I've setup 4 self-hosted runners on my ARM64 VM at Oracle Cloud.
With https://github.com/martin-g/zeppelin/pull/1/files I've started using them.
But there is a problem with Miniconda: https://github.com/martin-g/zeppelin/runs/4507098156?check_suite_focus=true

Error: No installed conda 'base' enviroment found at 

I've asked for help at conda-incubator/setup-miniconda#203 (comment)

[zjffdu]

Thanks for your update @martin-g

[zjffdu]

I haven't tried yet! Before I do it I want to explain something that was a stopper for another project:

When creating an account at Oracle Cloud (OCI) they will ask you for a credit card details. It won't be charged if you use the Free Tier (and this is enough for testing on ARM64)! But if you want to share this OCI account with the rest of the Zeppelin team then this might be problematic, because someone may add a paid resource to the account and OCI will start charging your card!

If the above is a problem for Zeppelin too then I could offer you a VM donated by my employer (OpenLab Testing) that could be used the same way!

This is indeed a problem, it would be great if your employer can help with that.

[martin-g]

OK! I will request a VM for Zeppelin project!

In the meantime I have started testing at OracleCloud with Github Actions self-hosted runner and I've found some issues with setup-miniconda action but I was able to work them around!
The latest state could be seen at https://github.com/martin-g/zeppelin/runs/4560861885?check_suite_focus=true:

Tests run: 2, Failures: 0, Errors: 2, Skipped: 0, Time elapsed: 0.452 sec <<< FAILURE! - in org.apache.zeppelin.notebook.repo.MongoNotebookRepoTest
testGetNotePath(org.apache.zeppelin.notebook.repo.MongoNotebookRepoTest) Time elapsed: 0.198 sec <<< ERROR!
java.lang.IllegalArgumentException: this version does not support 32Bit: PRODUCTION:Linux:B32

Embedded Mongo supports Linux ARM64 since recently: flapdoodle-oss/de.flapdoodle.embed.mongo#343

I will open new PRs for the new improvements!

[zjffdu]

@martin-g Are you still working on it?

[martin-g]

I was waiting for d8d18f5 to be merged.
I will resume/restart my work on this in the coming days!

[martin-g]

Closing this PR because TravisCI won't be used.
I will work on self-hosted runner for Github Actions.