Performance in multilingual project with site content mount fallbacks

support
1

We have a multilingual website project that uses translation by directory and site content mounts. This allows to have “complete” languages sites, taking the default site content as fallback:

#config/_default/module.yaml
mounts:
  - source: content/zh
    target: content
    sites:
      matrix:
        languages:
          - zh
  - source: content/en
    target: content
    sites:
      matrix:
        languages:
          - "*"

This works well, as users do not “traverse” site boundaries anymore (i.e. one navigates the Chinese website, and gets English content as fallback if Chinese is missing, while always staying on the Chinese site, so theme and other links are presented in Chinese if available).

Regarding translations management, we set a threshold of translated content needed before enabling newly added languages. But even with some guidelines, for many languages, the English fallback takes a large portion of the cake.

So the question is more about performance:
on such heavy website projects (say a few thousand pages, a few dozen languages, and maybe even a second dimensions like versions), how to optimize when most of the content comes from the same sources (i.e. English content is used at >90% for other language sites)?

Some rough stats on an old laptop (still faster than GitHub CI :smile:)

EN other langs
Pages 500 500
Paginator pages 100 100
Non-page files 500 0
Static files 10 10
Processed images 1000 0
Aliases 100 100
Cleaned 0 0
Number of langs Time ms
1 (only English) 1507
2 2508
5 5540
10 15327
20 25362

Where are possible improvements the most useful? Config, optimisation of theme, content structure? Any tips and tricks?
TIA!

2

This may not be an option for you, but you could have a look at the sites.complements option, e.g:

mounts:
  - source: content/zh
    target: content
    sites:
      matrix:
        languages:
          - ar
  - source: content/en
    target: content
    sites:
      matrix:
        languages:
          - "en"
        complements:
          languages:
           - "*"

The above is totally untested, but the gist of it is that you only render the en pages once. All the RelPermalink etc. linking is handled. The downside is obvious, but may be an OK compromise.