Re: Election vs consensus
Roberto Ostinelli <[email protected]> Fri, 7 Jan 2022 23:57:15 +0100
| Newsgroups | gmane.comp.lang.erlang.general |
|---|---|
| Message-ID | <[email protected]> |
Thank you Anthony. I need to spend more time on the raft.pdf (I’ve only skimmed through it before). Raft, afaik, once a leader is elected all it does is to send a new log (i.e. a write in a kv store) to all the followers and wait for their ack. What I’m trying to understand is what raft (a consensus algorithm) offers on top of a simple bully (a leader algorithm) with a similar ack mechanism, for example, in the context of a kv store. Some explanations probably are in the raft.pdf so that could be a good start. Thank you. Any other pointers welcome! Thanks, r. > On 7 Jan 2022, at 23:24, Anthony Howe <[email protected]> wrote: > > On 2022-01-07 17:14, Roberto Ostinelli wrote: >> Dear list, >> This is not directly related to erlang but for obvious reasons it’s pretty tightly related. >> >> I would like to understand why election algorithms such as bully are afaik deemed not enough to build consistent systems, hence the existence of consensus algorithms such as paxos/raft. More specifically in the creation of simple key/value stores. >> >> I would have imagined that with leader election you can pipe all the read/write operations through the leader, hence implement consistency. However i.e. raft came in to fill this kind of scenario, and I would like to understand why it is needed. >> >> Does someone have pointers on things to read? > > Some of the Leader Election / RAFT related material I collected when I was > looking into the subject. > > https://raft.github.io/raft.pdf > > https://www.elastic.co/blog/found-leader-election-in-general > > https://www.tigera.io/blog/using-etcd-for-elections/ > > -- > Anthony C Howe SnertSoft > [email protected] Twitter: SirWumpus BarricadeMX & Milters > http://snert.com/ http://nanozen.snert.com/ http://snertsoft.com/