etcd Questions (and Answers)

etcd-q-and-a.txt

Q: How do I remove a node from an etcd cluster?
A: We don't have a way to do that yet, but that functionality is slotted for the 0.3.0 series as part of the Cluster Management API.

Q: Can I count on a value I read from the slave always being a value that was at some point the current "true" value on the master?
A: Yes. Writes never go through the slaves.

Q: When reading from a slave that is in a minority partition, can I still read that (potentially old) value?
A: Yes.

Q: If each CoreOS machine has an etcd in the same cluster, doesn't that limit the number of CoreOS machines you can have?
A: We have a plan for that: https://trello.com/c/CPflQTUp/35-0-3-benched-followers.

Q: Why does updatedIndex have the same value as createdIndex?
A: Set operations do a wholesale overwrite. You must use Compare And Swap if you wish to *update* a value.

Q: Does etcd have endpoints exposing stats?
A: Yes, but they are not (yet) documented. 
     /v2/stats/self
     /v2/stats/leader
     /v2/stats/store

Q: Is there a reason why etcd doesn't listen on 0.0.0.0 by default?
A: By default there is no auth or security, it didn't seem very nice to do that.

Q: Is the config file format documented somewhere?
A: Not Yet. But if you know some go, you can look at these: 
     https://github.com/coreos/etcd/blob/master/server/config.go
     https://github.com/coreos/etcd/blob/master/server/config_test.go

Q: What happens if both -peers and -peers-file are given?
A: The peers file (if any) overrides peers.

Q: What format is the config file in?
A: TOML. See https://github.com/mojombo/toml.