Memcached Clusters?

[sishen]

Bash Coder wrote:
> On 10/24/06, *Moritz Maisel* <maisel at sipgate.de 
> <mailto:maisel at sipgate.de>> wrote:
>
>
>     If there were 3 Servers X, Y and Z. A client uses cache Y. After
>     that Y goes
>     down. Now the client takes another cache. When B comes up again it
>     is again
>     used by the client and we get outdated data ...
>
>
> I would have thought that if Y goes down, the client goes to the 
> database(s).  At least that's what I'm planning.  If the client can't 
> connect it will go directly to the source.  I mean, that's what the 
> client would have to do anyway if it could connect to the sever but 
> the server didn't have the data cached.
Maybe you can see Tugela_Cache 
(http://meta.wikimedia.org/wiki/Tugela_Cache), which use bdb as its 
backend storage.  When you restart the server, you can reload the data 
from the database. I think this way have two shortcoming:
1. it  is time-consuming.  it use bdb cache for memory management, which 
is less effective as slab+hash.
2. When the server down, the client is just to choose to use another 
server, when you reload the data from db, the data is duplicate.
>
> I'm a new user, and I love the idea of memcached but I don't think it 
> was designed to be reliable.  It's not the Google filesystem.  That's 
> cool, as long as I don't architect my system as though it were.
keep the memcached simple and easy-maintenance, :)
>
> - Bash
>
> -- 
> http://bashcoder.com