HN Explorer

Ask HN: Hostname Conventions for Distributed Systems?

3 points by rellik ↗ HN
I've gone through a variety of hostnaming strategies over the years, and the one I use now is <service><##> (eg zookeeper01).

With distributed systems in particular, but any system in general, boxes can come and go, and I hate having holes in my names (eg zookeeper01, 02, 04, 05, 06). That, and by choosing the number of leading zeroes, you're guessing at the max number of servers this service will be served by.

How do people manage this? I'm thinking something like <service>-<random_token> (eg zookeeper-fi2) would make sense and be more scalable, but I'm interested in what strategies other people use.

(I'd create a poll but I don't have the 200 karma) =/

3 comments

[ 3.9 ms ] story [ 17.6 ms ] thread
merinid
I like to think of everything as ephemeral. You can always write software for a new box to "fill in" holes when systems are replaced. I like this because it helps me when I assign roles or node based redundancy to my boxes.
rellik
Yeah, the tokenized strategy is kinda going after the ephemeral nature of the boxes, which is why I like it. Do you use the numbered approach?
merinid
Yes. Numbered approach, which I like because I can pack data in. So for instance: dbserver_01_n01. Which is the first node of the first db server. I have servers that are mostly used in read only and others which are write only. But their roles switch. We do all of the writing (not users), so it makes sense to split the namespacing this way.