Hi, Yes. However this will only work as long as long as you don't loose too many VMs at the same time - a quorum (at least half) of the servers still need to have their data (snapshot files and transaction logs). As long as this is true, they can elect an up-to-date server among themself and this leader then will get all other servers up to date.
Alex On Fri, Dec 19, 2014 at 7:46 AM, Ben Hsu <[email protected]> wrote: > > Hello Alex, > > Thats interesting. Does that mean if I lose a VM, I can spin a new VM with > the same DNS name and myid file, attach it to the cluster, and ZK will fill > in the data bits? > > On Thu, Dec 18, 2014 at 5:41 PM, Alexander Shraer <[email protected]> > wrote: > > > > Hi, > > > > The id uniquely identifies a server during its lifetime. IP and port can > be > > change during runtime but the id remains. > > > > Also theoretically its possible for multiple servers to have the same IP > > and port, although probably no one will define it like that :) > > > > Alex > > > > On Thu, Dec 18, 2014 at 12:47 PM, Ben Hsu <[email protected]> > > wrote: > > > > > > Hello > > > > > > A friend at work asked me why the myid file is needed. Can't zookeeper > > > divine which node a request is coming from based on the source IP and > > port? > > > > > >
