Hacker News Reader

I'm trying to figure out the odds that you intended this as sarcasm.

In the real world, it doesn't work like that.

Sarcasm indeed.

Anyone who correctly implements a payment service wouldn't make it go down for 7 hours.

Right... I'm sure that's what happens.

I would actually love to know if you have done this in prod and what it looked like.

I've seen people try and do it. It involves each service replicating copies of the data from every service it's dependent on (via CDC streams on kafka/whatever) in its own datastores and reading directly from its copy of data rather than actually calling the source-of-truth service.

There are some instances where that kind of thing is necessary, but...insane amount of clear downsides so your problem statement for doing it had better be pretty compelling.

This is typically referred to event carried state transfer for anyone curious.

The premise being high service availability and low latency to serve requests (no need to talk to source of truth) with the tradeoffs of higher storage / processing costs and from my experience higher complexity.

This is brilliant :)