simpleflake

I've used something very similar in the past, called SimpleFlake[0], which is essentially a 64 bit version with the same principles. I've used it in Lisp, C, C++, Clojure, Python, and Rust. It's conceptually simple, and fits in a 64bit int, which is natively available in a lot of databases. [0] SimpleFlake - https://github.com/SawdustSoftware/simpleflake/blob/f2b51f76...

Hey, this is pretty insightful and not insulting at all - I always love hearing about the experiences of others. The RNG being used is not the random.random() Mersenne Twister one, it's the OS-provided cryptographically secure one: https://github.com/SawdustSoftware/simpleflake/blob/master/s... But in any case, the reason you need NTPd sync is not to seed the RNG, but because the IDs themselves are prefixed with millisecond timestamps. This is to ensure that you're not using the same sub-keyspace for that millisecond over and over again from multiple processes. If you're seeing that many collisions, obviously, it's time to change methods like you mention. Data being dropped out of existence sounds like an application code issue, and not an inherent problem with your IDing scheme. That sounds like a place where ops should get a big honking