That’s correct. The solution would be either:

1. Centralize authentication and nothing else
2. Replicate password hashes & salts to all instances

#1 would absolutely be the most preferable. There are two possible snags with option #1: sdfsdf

• You’re introducing a small element of centralization into a decentralized platform
• The admins of the centralized authentication server must be competent and trustworthy

Cost wouldn’t really be a factor for this solution and could easily be sustained from donations, you’re not going to be getting a huge amount of traffic from authentication requests.

I included option #2 for the sake of completeness. This would work, but it isn’t the best idea from a security standpoint. The risk can be greatly mitigated with good password requirements and the use of a strong password hashing algorithm like Argon2/bcrypt/PBKDF2/etc in combination with salting the hash. A quick look at Lemmy’s code shows that they’re already hashing passwords with bcrypt, so that requirement is met, but it doesn’t look like they’re explicitly salting it. That doesn’t really matter too much in this scenario with that algorithm though, since it’s going to be salted automatically anyway. Lemmy’s code also shows that it’s using bcrypt’s default cost value (10 rounds), so it would take thousands (to millions) of years to crack the hash if you have even the most basic password requirements in place. If you add the option to put MFA in front of that, you’ve almost removed the risk entirely, as it won’t matter in the very unlikely event that the password actually is cracked, because it’s useless without access to the second authentication factor.

So yeah, there are a couple of ways to do it, and each have their downsides/tradeoffs, but the level of difficulty/effort to do it is not very high in either case.