Configure all nodes of a distributed system

Question

I want to update the configuration of a distributed system, i.e. I want every node to retrieve an updated configuration.

Now I can imagine several ways of doing this:

• There could be a central configuration database.
  • Advantage: Even completely new nodes know where to get their initial configuration from.
  • Disadvantage: The database is a single point of failure.
• Every node has its own config, and all nodes are updated from the outside.
  • Advantage: No single point of failure.
  • Disadvantages: Long waiting times; new nodes need a way to get their initial configuration.

If I use a central tool for the second approach, or something such as a gossip protocol, doesn't matter. Basically, the general idea is to either go for authority (pull-based) or for autonomy (push-based).

How do distributed systems solve this usually? What options are there? Any hints, ideas, …? What are best or bad practices here?

Answer 1

Golo,

i've always favored the second approach though its easier said than done! basically, you could have a look at what cassandra DB does as they do gossiping in the background. This methodology is not strict as a node may have an update configuration than another one however if you dont care for some seconds then its ok.