Make Digest subclass lookup thread-safe in lru_cache

[cmd-ntrf]

r10k content_synchronizer uses Forge v3 API when installing a module: https://github.com/puppetlabs/r10k/blob/4.0.0/lib/r10k/module/forge.rb#L95 https://github.com/puppetlabs/r10k/blob/4.0.0/lib/r10k/module/forge.rb#L128 https://github.com/puppetlabs/r10k/blob/4.0.0/lib/r10k/module/forge.rb#L176

r10k uses a thread pool to install the modules, therefore the Forge v3 API calls have to be thread safe, and so does the LruCache class used to cache API responses.

If LruCache is left not thread safe, the same error as reported in r10k issue #979 happens.
puppetlabs/r10k#979

[CLAassistant]

All committers have signed the CLA.

[bastelfreak]

@cmd-ntrf thanks for the PR. Any chance you can add a test for this?

[cmd-ntrf]

It appears some tests have been written but commented:
https://github.com/puppetlabs/forge-ruby/blame/main/spec/unit/forge/lru_cache_spec.rb#L70
https://github.com/puppetlabs/forge-ruby/blame/main/spec/unit/forge/lru_cache_spec.rb#L105

@hsnodgrass is probably more qualified than I am to add test.

[hsnodgrass]

I'm not really sure how to unit test thread-safety in a deterministic manner. It looks like we could use https://github.com/yegor256/threads, but I have no experience with it.

[hsnodgrass]

I did find this merged PR for Puma which makes the same change: puma/puma#2744. Personally, I think we're ok to merge this change without a specific test. This is a known issue in Ruby that I wasn't aware of when I wrote the cache

[alexjfisher]

Is this why I've just hit the following in r10k? (4.0.0 running in an AWS lambda function...)

[2023-09-20 15:54:34 - ERROR] Module selinux_core failed to synchronize due to Digest::Base cannot be directly inherited in Ruby
[2023-09-20 15:54:34 - ERROR] Error during concurrent deploy of a module: Digest::Base cannot be directly inherited in Ruby

[cmd-ntrf]

Is this why I've just hit the following in r10k? (4.0.0 running in an AWS lambda function...)

yes. You can mitigate the problem by freezing puppet_forge gem version to 4.1.0 (i.e: gem install puppet_forge:4.1.0 r10k:4.0.0). The LruCache was introduced in 5.0.0.

[malikparvez]

LGTM