Re:Strict-serializability, but at what cost, for what purpose?
by sitkack on Friday August 05, 2022 @03:57AM
(#32353041)
> Ever try to make an atomic change to two git repositories?
I know your intent wasn't to discuss this. But you could
1) combine them
2) use tags or git hashes as your URI
3) create a third repo that uses git submodules to combine them by said git hash
Trying to maintain global state by having both gitrepos be compatible at the same wall time isn't as important as it seems. There are lots of ways to meaningfully deal with this.
I think the article is talking about distributed systems where it's costly to guarantee ordering among multiple entity groups (shards for example), and there are "never" transactions across those entity groups. For example, the database hosting your website and the database backing my website can't talk to each other, so it doesn't really matter that transactions that we execute against our database don't appear in a defined order; we don't even know that the other exists.
I do agree with the sentiment that if those two databases are at the same company, eventually you'll be burned by not being able to do transactions across them. Ever try to make an atomic change to two git repositories? Not possible, but often quite annoying. This is doubly annoying when you start having internal sync processes because two teams used two different database to ultimately build, what looks like to the user, one app.
My philosophy is that if strict serializability ever makes it impossible to run my service, I'm already so rich that I can just sell the service and let someone else deal with the problem. So far, no dice on that.