to answer myself if anybody else will need this too :)
in the case described the right way will be not to use consistent hashing 
but simple modulo. If application can tolerate cache inconsistency easily - 
option with consistent hashing is good. In case of failure just remove 
failed server and most keys are unaffected. If cache consistency is somehow 
priority - modulo distribution is better as replacement server will take 
full load of failed one.

On Thursday, May 22, 2014 12:22:18 PM UTC-7, Denis Samoylov wrote:
>
> hi,
> sorry for libmemcached question, i did not find dedicated group for it and 
> found question about it in this group. But if not appropriate - pls delete.
>
> we use chain of pecl memcached->libmemcached and use 
> option OPT_LIBKETAMA_COMPATIBLE. So e.g. we have three servers 
> "appcache1,appcache2,appcache3" and server #2 (appcache2) goes down and we 
> want to replace it with a new one: appcache4. Even if we feed pecl 
> memcached preserving order: appcache1,appcache4,appcache3 keys will not 
> stay with the same server :(. Is there any way to make buckets that I will 
> be able to assign servers and change of one server will not affect 
> consistent hashing?
>
> Thank you!
>

-- 

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

Reply via email to