updating readme with free memory requirement [revised]

[rorosaurus]

There is a non-obvious requirement for memory to be available for the crypto libraries on the client. The readme's System Requirements section didn't reflect this. On low-memory systems this crashes the client with an extremely verbose but not terribly useful error.
See #1883 (comment)

[pde]

I agree we should mention that in the readme, but the RAM isn't a requirement of the client; it's a requirement of the installation approach taken by letsencrypt-auto.

If there are client packages available for your OS, those will work on low-memory systems. And if your OS packages a recent copy of python-cryptography but not the letsencrypt client, you may be able to do pip install letsencrypt letsencrypt-apache and have that work.

Can you update the PR to reflect those nuances in some way?

[rorosaurus]

Excellent points. How does this new wording sound? Does that capture the details adequately?

Can you confirm I correctly stated this last bit?

if your OS packages a recent copy of python-cryptography, you
may be able to run letsencrypt without the letsencrypt-auto wrapper,
avoiding the compilation.

[pde]

Hi @rorosaurus, I've made a PR to get that wording to be clearer; however it starts to seem like too much detail on this specific issue for the main README. Should we perhaps describe these workarounds in docs/using.rst and just link to that section from the top level README?

[rorosaurus]

Yeah, I was thinking the same. It now seems too specific for the main README.
My original reason for wanting to mention the memory complications here was just to spread awareness. Maybe a brief message, then a link to that section in docs/using.rst would be perfect?

[seanthewebber]

@pde @rorosaurus Can we consider using letsencrypt instead of letsencrypt-auto an official solution or is this a temporary workaround?

[pde]

@seanthewebber yes that's definitely the plan; in the long run we aspire to retiring letsencrypt-auto in favour of native packages.

[jsha]

Thanks for the PR! I've wrapped the advice into #4305, which also updates some more of the documentation.