Travis not running

[John Sichi]

I just now noticed that Travis CI hasn't been running for JGraphT since Jan 13.  

When I check the status, it is giving me some error about needing to replenish OSS credits; I'm looking into it.

[John Sichi]

Here's more info on what happened:

https://blog.travis-ci.com/2020-11-02-travis-ci-new-billing

I've submitted a support ticket to see if we can get OSS credits.

[Dimitrios Michail]

It seems that the OSS credits are not replenished by default.

https://docs.travis-ci.com/user/billing-faq/#what-if-i-am-building-open-source

Do we have some specific reason for running on travis.com instead of travis.org? 

The good news is that we are not performing any MacOS builds, as they seem to 

change their billing policies on such builds.

--
You received this message because you are subscribed to the Google Groups "jgrapht-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email to jgrapht-dev...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/jgrapht-dev/b4bb7d62-30c5-4e16-ae7c-51df62e446fen%40googlegroups.com.

[John Sichi]

We used to run on travis-ci.org, but they made everyone migrate to travis-ci.com a while back, and travis-ci.org is being shut down soon.  Monetization!

If they don't get back to me soon about the OSS credits, I'll use some of our remaining GSoC funds to buy some paid credits as a temporary measure whlie we consider our options.

[Dimitrios Michail]

I am building the python-jgrapht stuff on travis-ci.org and there I actually need a MacOS, Windows and Linux builds!

Let us hope they truly support OSS, not just for marketing purposes. 

D.

To view this discussion on the web visit https://groups.google.com/d/msgid/jgrapht-dev/46bf1aa9-22a1-48d0-9bae-c20c24ff17f3n%40googlegroups.com.

[Szabolcs Besenyei]

Any particular reason why we are not leveraging Github Actions? Free for all public repositories, has matrix builds, supports all languages and many more.

[John Sichi]

If it can reliably do everything Travis has been doing for us (checking main branch commits as well as pull requests, and also publishing our site+javadocs), with good diagnostics when things go wrong, and someone wants to work on moving everything over, that would definitely be one of the options to consider.

[Dimitrios Michail]

Indeed. 

The reason is probably that it did not exist when we first wrote our pipelines. If it supports our use-cases and someone has the time to contribute such a transition...

I will definitely check if python-jgrapht builds can move there. They say that they give 2000 action minutes/month for free, and they seem to have all required environments.  

 

To view this discussion on the web visit https://groups.google.com/d/msgid/jgrapht-dev/ab4e3814-67fb-4c6c-b930-cff8eec2fa8fn%40googlegroups.com.

[Szabolcs Besenyei]

I created a Github workflow that I believe is equivalent to the Travis build that we currently have for pull requests.

Example execution is available at https://github.com/besza/jgrapht/pull/4/checks .

The YAML descriptor can be found at https://github.com/besza/jgrapht/blob/master/.github/workflows/PR-workflow.yaml .

A GH-hosted worker has 2 CPU cores, 7 GB RAM and 14 GB SSD RAM and it has the Zulu Community Distribution of OpenJDK.

I'm moving on to define a pipeline for the /master branch but I believe the PR pipeline can already be adapted as-is.

[John Sichi]

Thanks a bunch for taking this on!  I haven't gotten anything back from Travis support yet, so let's go ahead and get this set up for PR's immediately while you keep working on the master commits (note that a token will probably be needed there for deploying the site).  Can you test a failed build in your repo to make sure you get email notification?

For testing the PR support in JGraphT proper, is this the correct sequence?

1) I add the workflow by committing it directly in master

2) You pull latest and then create a new dummy PR

3) If everything is working properly, we can ask Dimitrios to merge from latest and then get a green checkmark for the webgraph PR again.

If so, I'll go ahead with step 1.

[Szabolcs Besenyei]

I got email notification for each failed build while I was trying to get it right. I'm not sure how that would translate to a real multi-user repository although it probably can be configured.

Your assumption about the rollout is correct, however, I forgot to mention that I had class-loading issues both on my local machine (AdoptOpenJDK 11 Openj9) and on the Github worker as well:
https://github.com/besza/jgrapht/runs/1840620338?check_suite_focus=true

The workaround seemed to be to set the "useSystemClassLoader" property to false for the Surefire plugin. I also bumped the version to 3.0.0-M5 for both the Failsafe and Surefire plugins, but I suspect this can be omitted.

[John Sichi]

OK, I added your PR workflow to master---can you give it a test?

[Szabolcs Besenyei]

Please also bump the "surefire-junit47" dependency to 3.0.0-M5.

[John Sichi]

Done.

[Szabolcs Besenyei]

Alright, here we go. The build is still going to fail because last time I checked there are some checkstyle violations that creeped in, however I think we get the gist that it is working.

[John Sichi]

OK, I pushed a fix for the checkstyle violations.  Can you sync that and verify that your PR goes green?

[Szabolcs Besenyei]

I did so myself. It is green now.

[John Sichi]

Awesome.  Dimitrios, can you sync the webgraph branch?

[Dimitrios Michail]

Just merged the changes from master. 

To view this discussion on the web visit https://groups.google.com/d/msgid/jgrapht-dev/8f545636-6a1d-49b8-88f4-e240271975ebn%40googlegroups.com.

[Szabolcs Besenyei]

I'm puzzled by the class-loading issue that keeps popping up, especially because it doesn't seem to be deterministic. (The dummy PR was green but the Webgraph PR is red)

John, could you please
  1. add a repository secret named `ACTIONS_RUNNER_DEBUG` with the value `TRUE`
  2. and restart the build?

I went down a rabbit hole of Stackoverflow and Github Issues thread but the problem is still not clear to me.

Any other ideas?

[John Sichi]

OK, I added the repository secret and hit re-run all jobs.

By the way, for permissions, can I change it to "Allow local actions only"?

[John Sichi]

First re-run, it failed with some connectivity error trying to fetch artifacts.

Second time, it succeeded and went green.

I'm re-running again to see what other faliures come up.

[John Sichi]

Hmm, green again.  Guess we can start using this for real and leave the secret set for diagnostics in case there are more non-deterministic failures?

[Szabolcs Besenyei]

> By the way, for permissions, can I change it to "Allow local actions only"?

We shouldn't. Currently I'm using official Github workflows that are defined in the "actions/*" namespace and one community workflow to deploy the Github Pages.

Once we finalize the pipelines we can go with the "Allow select actions". (More about this at https://docs.github.com/en/github/administering-a-repository/disabling-or-limiting-github-actions-for-a-repository)

Regarding the master build pipeline, I believe it is 95% percent there. It is similar to the PR pipeline with the following additions:

  1. Event trigger is "on push master"
  2. After a successful build, it installs Nodejs 14 LTS and Hercule v5.0.0
  3. Runs prepareDocs.sh
  4. Deploys the snapshot to Sonatype
  5. and finally, it pushes the docs/ folder to the gh-pages branch

Did I left something out?

I had a successful build, with the exclusion of step #4:
  * https://github.com/besza/jgrapht/actions/runs/545400465.
  * https://github.com/besza/jgrapht/commit/fd6efcd524ad11897431e8943c79fce87852b22f

PS: There is also an SVG badge for the README about the build status.

[John Sichi]

That pipeline looks correct to me.  Let me know when you're ready for me to test it on master.

[Szabolcs Besenyei]

I submitted the PR with instructions on how to setup some repository secrets. https://github.com/jgrapht/jgrapht/pull/1038

[John Sichi]

OK, I added the secrets, merged your PR, and then did a direct commit push for giving you credit.  So far I'm not seeing any action triggered.  Is it because the workflow is defined to trigger on "main" instead of "master"?

[Szabolcs Besenyei]

> Is it because the workflow is defined to trigger on "main" instead of "master"?

Yes. It should be master.

I keep switching them up because for new repositories it is called "main" since October.

[John Sichi]

I updated it to master, and now everything looks good.  So I went ahead and merged the WebGraph PR, and verified that the site docs and Sonatype snapshot got updated correctly.

Dimitrios, I guess we'll still need to figure out what to do about the Mac builds for python, since eventually the credits will run out there too.

Also, if you and Sebastiano can do a writeup for the WebGraph adapter, we can make some noise about that collaboration.