Re:Circuit: A Minimal Distributed Programmatic OS written in Go
by alexpw on Saturday April 18, 2015 @10:36AM
(#9399575)
Afaict, the circuit server runs on every host, and is aware of the other nodes, coordinating with peers without a single master. So if one host fails, the other circuits can still function, broadcasting their underlying services.
"Clients dial into [any] circuit server in order to interact with the entire system. All [circuit] servers are equal citizens in every respect and, in particular, any one can be used as a choice for dial-in."
However, the example diagrams I've seen, include services which aren't masterless or even multi-master, so the app/system the circuit cluster supports will surely fail in some way.
Re:Circuit: A Minimal Distributed Programmatic OS written in Go
by al2o3cr on Saturday April 18, 2015 @04:34PM
(#9400907)
"Clients dial into [any] circuit server in order to interact with the entire system. All [circuit] servers are equal citizens in every respect and, in particular, any one can be used as a choice for dial-in."
This sounds like it will cause exciting behavior under a network partition...
When I read the document, start a process in circuit is equal to start a process on a individual host. So how circuit make sure this process which run on a individual host could not fail?
But if the fail mean the whole circuit cluster could not fail, as I now, CoreOS, Mesos, Kubernetes also could not fail.