When the box goes down, all of the clients are smart enough to realize it and they will take it out of rotation. If you want to put another one in its place, then I'd say you should regenerate your config minus the old machine and including the new one, then push it to your client machines...
On Fri, Feb 19, 2010 at 6:25 PM, Nelz <[email protected]> wrote: > A-ha! I knew I'd need to flesh this out a bit. :-D > > The problem is when one of the memcached boxes goes down... Using IP > addresses is not feasible because a new box coming up won't have the > same IP address. > > - Nelz > > On Fri, Feb 19, 2010 at 15:03, Adam Lee <[email protected]> wrote: > > I'm trying to understand your problem. Can't you just have one > > configuration file that's generated in one place and rsynced out to all > > client machines? If it needs to be updated, then regenerate, push it out > > again and call a reload function... > > > > On Fri, Feb 19, 2010 at 5:55 PM, Nelz <[email protected]> wrote: > >> > >> Hey all, > >> > >> We are using the spymemcached client with consistent hashing. In my > >> understanding, because we are using consistent hashing we need to > >> utilize the same exact strings on all clients for identifying the > >> physical nodes in a logical memcached cluster. > >> > >> Does anyone have a good solution for handling the dynamic internal IPs > >> on EC2 instances? > >> > >> We're going to try dynamic DNS and use hostnames, but wondered if > >> anyone has tried other solutions in large-scale implementations. > >> > >> Let me know if my question is as clear as mud and could use some more > >> background. > >> > >> - Nelz > > > > > > > > -- > > awl > > > -- awl
