Cache bucket locations

[henrikskog]

Fixes #14

Run with RUST_BACKTRACE=1 cargo test test_load_all_buckets -- --nocapture. This requires that you have an account authenticated with some s3 buckets though. I'll remove the tests if this gets approved. It's just the easiest way I have found to do manual testing.

• Creates a file in $DEFAULT_STU_DIR/cache.json that is currently only used for the bucket locations
• Fetches bucket locations from this cache if it gets a bucket that has been fetched before
• Uses moka for the cache. I write the whole cache to file each time put is ran. Super inefficient, but does not really matter maybe? The cache is loaded from file at startup.
• Factored out the check if a bucket is in a region to a seperate function to not bloat the other function too much

Very very rough and needs a lot of cleanup with logging and error handling probably. I thought it might be smart to hear your thoughts about the approach here before fixing up all the small things with the code. What do you think?

[henrikskog]

Marked as draft as this is not something that is ready to merge.

[lusingander]

Thank you for addressing this!

At this stage, I believe implementing the bare minimum is sufficient.

• A library like moka doesn't seem necessary because there are no repeated references or updates while the application is running (at least for now).
• I don't think multiple updates are much of a problem, but it seems better to read the entire cache first and save the entire cache at the end.
  • There are several possible places to do this, and any would be fine for now:
  • Read it before creating the client and pass it to new
  • Read it in new
  • Read it in load_all_buckets
  • Write it after load_all_buckets
  • Write it in load_all_buckets
  • (However, it seems better to be consistent about whether you do it inside or outside load_all_buckets.)
• The cache file location looks fine.
• Refactoring the function seems like a good idea.

[henrikskog]

A library like moka doesn't seem necessary because there are no repeated references or updates while the application is running (at least for now).

I removed the moka dep and created the cache manually instead

I don't think multiple updates are much of a problem, but it seems better to read the entire cache first and save the entire cache at the end.

I added the location check to the load_bucket method as well. Now the cache is written at the end of both load_bucket and load_buckets. IMO it would be clean to do it at the end of get_bucket_region at the expense of writing multiple times to file. This way the cache logic is only a concern of that spesific function.

(However, it seems better to be consistent about whether you do it inside or outside load_all_buckets.)

Hmm I'm now reading in the new method and writing inside the load-bucket methods. I think it's nice to keep it as an internal concern in Client.rs instead of doing it in app.rs?

[henrikskog]

Thanks for the review. Knowing when to panick and when to propagate errors is a bit tricky I think.

[lusingander]

Looks good, thanks!