Remove restore cache keys fallbacks and solve CI segfault problem

[microstudi]

🎩 What? Why?

This is (yet) another attempt to discard a possible issue due cached gems.
It seems that is possible that we are retrieving backups from an older version of ruby as we use many fallbacks for the action/cache in Github workflows. This PR is to see if this may be the cause.

📌 Related Issues

• Related to #?
• Fixes #?

📋 Subtasks

• Add CHANGELOG upgrade notes, if required
• If there's a new public field, add it to GraphQL API
• Add documentation regarding the feature
• Add/modify seeds
• Add tests
• Another subtask

📷 Screenshots (optional)

[microstudi]

One test still failing due a segfault 😢, the other seems to be a flaky.

Usually segfaults come from a C extension gems. According to heroku this command should return all the C extensions used in the gemfile (assuming you have your gems in a subfolder of the current directory):

bundle show --paths | ruby -e "STDIN.each_line {|dep| puts dep.split('/').last if File.directory?(File.join(dep.chomp, 'ext')) }"

which gives the list:

bcrypt-3.1.13
bindex-0.8.1
bootsnap-1.4.6
byebug-11.1.3
charlock_holmes-0.7.7
concurrent-ruby-1.1.7
ffi-1.13.1
html_tokenizer-0.0.7
jaro_winkler-1.5.4
msgpack-1.3.3
nio4r-2.5.3
nokogiri-1.10.10
pg-1.1.4
puma-4.3.5
rb-fsevent-0.10.4
redcarpet-3.5.0
sassc-2.3.0
seven_zip_ruby-1.3.0
thread_safe-0.3.6
unf_ext-0.0.7.7
websocket-driver-0.7.3

PR from Sep 10:

From these, only the next were upgraded in #6513

nio4r (2.5.2) => nio4r (2.5.3)
nokogiri (1.10.9) => nokogiri (1.10.10)
concurrent-ruby (1.1.6) => concurrent-ruby (1.1.7)
seven_zip_ruby (1.2.5) => seven_zip_ruby (1.3.0)
websocket-driver (0.7.2) => websocket-driver (0.7.3)

In #6509 only this was upgraded:

seven_zip_ruby (~> 1.2, >= 1.2.2) => seven_zip_ruby (~> 1.3)

[microstudi]

Looking at several logs with segfaults,
I haven't found these gems to be implicated:

seven_zip_ruby
websocket-driver

and:

nio4r
nokogiri
concurrent-ruby

Where present in all the traces.
However, specifically in the Process memory map dumped trace, only these gems are present:

nokogiri
nio4r

[microstudi]

I haven't found information about nokogiri segfault that could be the same as we are having here.
However I did find something about the gem nio4r in socketry/nio4r#251

It seems that version 2.5.3 is giving intermittent segfaults as pointed here socketry/nio4r#251 (comment)

[microstudi]

Bingo!

[mrcasals]

Excellent job @microstudi! Super nice debugging! 👏 👏 Thanks for putting the explanations and the train of your thoughts, I found it super easy to understand what's going on!

Should we add the cache keys again changing their names so we don't get old cache versions? Maybe we could prepend the Ruby version:

          key: ${{ runner.OS }}-ruby266deps-${{ hashFiles('Gemfile.lock') }}
          restore-keys: |
            ${{ runner.OS }}-ruby266deps-${{ env.cache-name }}-
            ${{ runner.OS }}-ruby266deps-
            ${{ runner.OS }}-

[microstudi]

@mrcasals do we need to specify fallback keys? I mean, the one created is going to be used anyway. In fact, for what I've read here: https://github.com/ruby/setup-ruby#caching-bundle-install-automatically it seems that we could get rid of all cache statement as is included in an option of the setup-ruby action.

We could configure setup-ruby as:

    - uses: ruby/setup-ruby@v1
      with:
        bundler-cache: true

[mrcasals]

Oh, I didn't know about that option! Then yeeah, I think it's worth using it. We should update the action version too to make sure the option is available! 😄

[microstudi]

All right, I'll make another commit see if the builtin cache works, this way we pass the test suite again, let's make sure there's no segfaults.

[tramuntanal]

Excel·lent analysis @microstudi 👏 👏
We were myops only focusing on upgrading Ruby

[tramuntanal]

Let's see if with the next commit the corresponding buid succeeds at first attempt 🤞
I'm sure it will

[microstudi]

@mrcasals It seems that the built in bundler works smoothly.

Here is the first run:


An here the second:

I'll change all the workflows then.

[Leusev]

Wow!! Excellent catch here and great job @microstudi 😄
Thanks for all doc & explanation links too! It was easier to follow your thoughts
It seems all test have been passed, so looks good for me 👍
Thanks a lot!

[tramuntanal]

Great!