Long story short, it always ends up with Lamport's clocks (or some variation of them). Some are easy, some are overly overcomplicated. Some implementation are just reinventing Lamport's clocks, because authors didn't know about them before.

Yes. Now do it on a production level distributed system with thousands of request per second with network latency and network failure and every race condition you can think of. Orders of magnitude more difficult.