Paxos implementation in doozer

[Prat]

How exactly is paxos implemented in doozer? Any change made on a server will go through the consensus algorithm? First time a change is made on a server, it sends the prepare command to all the servers and the value gets chosen. The next time a change is made on the server, it sends out a prepare statement to all the servers, but in this case the acceptors return the previous value accepted, and this value will be sent again by the proposer. So when will a new value ever be chosen? What exactly happens after a value is chosen?

[Phil Whelan]

I think doozerd is no longer under active development. You are better to move attention to the etcd and consul.io projects which both use the raft protocol instead of paxos. 

Sent from my iPhone

On Dec 3, 2014, at 11:37 PM, Prat <dprat...@gmail.com> wrote:

How exactly is paxos implemented in doozer? Any change made on a server will go through the consensus algorithm? First time a change is made on a server, it sends the prepare command to all the servers and the value gets chosen. The next time a change is made on the server, it sends out a prepare statement to all the servers, but in this case the acceptors return the previous value accepted, and this value will be sent again by the proposer. So when will a new value ever be chosen? What exactly happens after a value is chosen?

--
You received this message because you are subscribed to the Google Groups "doozer" group.
To unsubscribe from this group and stop receiving emails from it, send an email to doozer+un...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Prat]

I know its no longer active. But I would like to know how its designed. Since there is no proper document for doozer, I am finding it difficult to understand.

Why is it no longer active? Does that mean the existing code is not stable? We can still use it right? And I observed the memory keeps growing inconsistently. 

Is there any way to know how memory is used by doozer? How does it change as the number of members and slaves grow?

[Phil Whelan]

We found it to be very flakey. Especially the available clients, which we spent a lot of time trying to fix. I haven't encountered anyone successfully running a cluster, but they might be out there. 

Sent from my iPhone