Hi Peng
> S1 and S2 will take turns to become leader for ever. This will make the raft
> cluster make no/little progress for serving its clients.
That is correct. Note that with most academic consensus algorithms,
the goal is to prove that the algorithm will continue to make
progress. Raft, as you described, will do that. How fast or slow the
progress will be, is usually not considered.
In practice of course you'd want to avoid such flip-flopping. Ongaro's
thesis - which is the last and fullest publication describing Raft -
actually briefly acknowledges this problem and suggests addition of a
pre-vote step to resolve the issue. It does not however describe such
a step in more detail.
I have written a paper that describes a pre-vote algorithm as well as
some other edge cases to arrive at a more practical Raft algorithm:
http://openlife.cc/blogs/2015/september/4-modifications-raft-consensus
henrik
> --
> You received this message because you are subscribed to the Google Groups
> "raft-dev" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to
raft-dev+u...@googlegroups.com.
> For more options, visit
https://groups.google.com/d/optout.
--
henri...@avoinelama.fi
+358-40-5697354 skype: henrik.ingo irc: hingo
www.openlife.cc
My LinkedIn profile:
http://fi.linkedin.com/pub/henrik-ingo/3/232/8a7