[RubenVerborgh]

Background

Prior to #1116, two simultaneous PATCH requests could corrupt files. The solution in #1116 was built with the design goals:

• to function in a multi-worker environment
• to not require external services (like Redis)

The file-based locking system—although meeting the above two goals—did cause some annoyances (notably #1372).

What this PR does

This PR switches the file-based locking to in-memory locking.

Closes #1372
Obsoletes #1464

Advantages

• no more gratuitous writes to disk
• uses an external library that is well-tested
• still no dependencies on external services

Disadvantages

• does not function with multiple workers or threads
  • …but neither do a a lot of other parts of NSS, so multi-worker support is future work anyway

What this PR does not

• read/write locking (Replace locking with proper in-memory RW locking #1460)
  • so still can't watch videos with multiple people (unless you use HTTP ranges)
  • we would need to implement https://en.wikipedia.org/wiki/Readers%E2%80%93writer_lock for that
  • …but I'm not touching that myself until NSS has a layered system where reads and writes happen in a single place, because they are now all over the handlers, which would make building and testing a nightmare

[michielbdejong]

Nice! Is it covered by a unit test?

@bourgeoa what do you think? Especially given your remark in https://github.com/solid/node-solid-server/pull/1464/files#diff-42068aebd9eb6d9326293ef88bb61b3dR32 - why did you discardasync-lock originally?

I don't have a strong preference either way, as long as it works :) Eventually I want to implement a proper caching layer with write-through so we will no longer need locking (writes to the cache will be synchronous).

[RubenVerborgh]

Nice! Is it covered by a unit test?

Yes, all the LDP logic covers it.

[bourgeoa]

@michielbdejong
I did find this solution and thought it could be used as a read cache for resources with text/turtle contentType, if the idea was wellcomed.

[RubenVerborgh]

It's probably too early for a cache, before we have the single read/write layer. Plus, we still need to solve the locking then with regard to disk writing.

[RubenVerborgh]

I think your SSD will be very happy, @timbl.