Hmm, I think "config" is too vague here. I've seen numerous incidents caused by NGINX config changes that would have been far worse if we had instant 100% rollouts, but I'm not sure this is the kind of config change the author is referring to.
> Run 3 of everything
What is a "thing" in this example? Is it an instance of a service? A database? Something else?
If you're using Kubernetes and have autoscaling, what does "3 of everything" even mean? It sounds like the author is basically talking about databases..?
> What is a "thing" in this example? Is it an instance of a service? A database? Something else?
No, you just run three of everything. While only databases and similarly distributed data storages really need exactly three (or sometimes, just an odd number larger than two) for quorum and network partition detection, there is a good possibility that you'd have three availability zones because of your databases. And since you already have three availability zones, it's only logical to put every service in every zone at least once, so running "three of everything" becomes a rule of the thumb, easy to remember, easy to implement.
> If you're using Kubernetes and have autoscaling, what does "3 of everything" even mean? It sounds like the author is basically talking about databases..?
If you use autoscaling, you still need to start somewhere. And this somewhere is "three of everything".
Hmm, I think "config" is too vague here. I've seen numerous incidents caused by NGINX config changes that would have been far worse if we had instant 100% rollouts, but I'm not sure this is the kind of config change the author is referring to.
> Run 3 of everything
What is a "thing" in this example? Is it an instance of a service? A database? Something else?
If you're using Kubernetes and have autoscaling, what does "3 of everything" even mean? It sounds like the author is basically talking about databases..?