This PR currently just logs failures. However, this is probably not really ideal
since this effectively means that if the write() call fails (due to a panic
occurring while the lock was previously held), the server essentially continues
operation without any project cache.

This isn't great since that would hinder performance until the server is
restarted. It would be better if the cache was reset if the lock has been
poisoned.