You are viewing a plain text version of this content. The canonical link for it is here.

Posted to user@zookeeper.apache.org by mufc_fan <ra...@gmail.com> on 2014/08/13 10:46:15 UTC

Leader election in zookeeper

I have 15 machines connected through a network. out of 15, zookeeper servers
are running in three different machines and I have to select a leader for
this distributed systems. My doubts are

1. For the election, 3 are zookeeper servers here so whether they also
contest in this election.
2. How the election will be held. can u explain with each steps.
3. After electing a leader, how a node knows it is the leader.
4. Also can u explain how zookeeper co-ordinates this service simply.




--
View this message in context: http://zookeeper-user.578899.n2.nabble.com/Leader-election-in-zookeeper-tp7580147.html
Sent from the zookeeper-user mailing list archive at Nabble.com.

RE: Leader election in zookeeper

Posted by "Asta, Greg" <gr...@omnigon.com>.

Zookeeper has properties that make it good at these tasks.  It's designed specifically for these things.  Many other scenarios I've seen in the past were not really 'built for these activities' but instead - had these activities tacked onto something else that was maybe less than ideal.  Here's a few examples of where Zookeeper is "custom fit" for these activities:

- Automatic handling of disconnected clients (ephemeral nodes)
- Automatic failover and redundancy
- Callbacks that trigger real time on updates to watched nodes
- Data consistency/integrity on every server in the cluster - including 'counters' that are accurate across every node in the cluster

>From those basic features you can construct the more complicated functionality that people want like distributed locking and distributed "atomic" counters - in a fairly cheap and reliable fashion.

Now - you can build all those things yourself and you can make them work on some of the more traditional database tools as well but with varying degrees of labor.  For the users of Zookeeper - it's really avoiding that labor and headaches that drives them to use a product like this.

-Greg

-----Original Message-----
From: mufc_fan [mailto:rajeshkumarit8292@gmail.com] 
Sent: Thursday, August 14, 2014 12:02 AM
To: zookeeper-user@hadoop.apache.org
Subject: Re: Leader election in zookeeper

can u tell me why we go for zookeeper...Before zookeeper also companies used distributed environment in which services like leader election,lock and others are implemented and they works fine. why we go for zookeeper?



--
View this message in context: http://zookeeper-user.578899.n2.nabble.com/Leader-election-in-zookeeper-tp7580147p7580157.html
Sent from the zookeeper-user mailing list archive at Nabble.com.

Re: How to measure the load vs response time in zookeeper

Posted by Kuba Lekstan <ku...@gmail.com>.

Have you tried this: https://github.com/phunt/zk-smoketest?

Kuba.


2014-08-14 17:37 GMT+02:00 tbgu <tb...@cs.umass.edu>:

> Hi Folks:
>             I am trying to get the load vs response time graph about the
> use of zookeeper. I want to increase the number of requests sent per second
> and get the corresponding average response time for each request.  Is there
> any interface I can use to exactly know the arrival time of  each request
> response time? Thanks very much!
>
> Best
> Tianbo
>

How to measure the load vs response time in zookeeper

Posted by tbgu <tb...@cs.umass.edu>.

Hi Folks:
             I am trying to get the load vs response time graph about 
the use of zookeeper. I want to increase the number of requests sent per 
second and get the corresponding average response time for each 
request.  Is there any interface I can use to exactly know the arrival 
time of  each request response time? Thanks very much!

Best
Tianbo