Home | About | Sematext search-lucene.com search-hadoop.com
 Search Hadoop and all its subprojects:

Switch to Plain View
Zookeeper >> mail # user >> Use zookeeper as Load Blancer

Neo Qing 2013-05-15, 05:26
Camille Fournier 2013-05-15, 15:55
Jordan Zimmerman 2013-05-15, 16:01
Neo Qing 2013-05-16, 01:28
Neo Qing 2013-05-16, 01:26
Copy link to this message
Re: Use zookeeper as Load Blancer
Hello Neo Qing,

I am not the best expert, but I will try to help.
I understand that you want to have a ZooKeeper client connected to a
cluster and you wonder if there will be fail over and load balancing for
the requests coming from that client. If I didn't get the question right,
please correct me.
If I am not wrong, if the ZooKeeper client loses its connection it will try
to reconnect to any of the servers that are indicated when you create the
ZooKeeper object, so there will be fail over. Of course, that will mean a
new connection. Some failures in the servers can be handled without the
client losing its connection, e.g. when a server other than the one you are
connected to goes down but there is still a quorum in the cluster.
Load Balancing for ZooKeeper write requests coming from clients wouldn't
help because all request need to go through the leader. Load Balancing of
read request could be achieved if the ZooKeeper clients connections are
balanced between the servers of the cluster, but not for a single client

I hope that helps. Anybody, please correct me if I got anything wrong.

Germán Blanco.

On Wed, May 15, 2013 at 7:26 AM, Neo Qing <[EMAIL PROTECTED]> wrote:

> Hi,There
>   I'm new to zookeeper, currently I'm working on a project, I have a
> cluster of server, client need to connection server (TCP/IP), and we need
> to keep the connection long live.
>   To do the fail over and load balance , I was wondering whether zookeeper
> can handle this case, especially for long live connection.
>   thanks in advance.
> Neo
Bahman Kalali 2013-05-15, 16:26
Neo Qing 2013-05-16, 01:25
Neo Qing 2013-05-16, 01:23
Jordan Zimmerman 2013-05-16, 01:27