So, what I've been doing is using haproxy to set the local nodes as primary. Though, this created a problem with Cassandra when I did the same because it obfuscated application errors and created failover problems that the client would normally handle because the client only saw a single host. The Cass client has a configurable load balancing policy. I was hoping ZK had the same so I could get rid of haproxy.
On Wed, May 1, 2013 at 6:22 PM, Jordan Zimmerman <[email protected] > wrote: > If you can determine local clients, just limit your connection string to > those local clients. Of course, this limits failover to local servers, but > would be a decent workaround until there's better support in ZK. > > -JZ > > On May 1, 2013, at 6:05 PM, Owen Kim <[email protected]> wrote: > > > Thanks for the info. > > > > That feature would be useful. I can determine local nodes myself and feed > > them into the client. > > https://issues.apache.org/jira/browse/ZOOKEEPER-1698 > > > > > > On Wed, May 1, 2013 at 5:48 PM, Michi Mutsuzaki <[email protected] > >wrote: > > > >> C client has zoo_deterministic_conn_order() to make the connection > >> order deterministic. We can add a similar feature to Java client. > >> Owen, do you want to open a JIRA for this? > >> > >> --Michi > >> > >> On Wed, May 1, 2013 at 5:40 PM, Benjamin Reed <[email protected]> wrote: > >>> no it will connect randomly. we don't really have a way to specify > >>> preference or weight. it would be a nice feature to add though. > >>> > >>> ben > >>> > >>> > >>> On Wed, May 1, 2013 at 5:26 PM, Owen Kim <[email protected]> wrote: > >>> > >>>> Is the client configurable on which host it tries to connect to when I > >> give > >>>> it the host list? Will it perhaps try deterministically in order? > >>>> Specifically, I'd like it to favor ZK hosts in the same DC. > >>>> > >> > >
