You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@zookeeper.apache.org by "kishore gopalakrishna (JIRA)" <ji...@apache.org> on 2010/11/08 21:19:08 UTC

[jira] Created: (ZOOKEEPER-923) TaskManagement Using Zookeeper Recipe

TaskManagement Using Zookeeper Recipe
-------------------------------------

                 Key: ZOOKEEPER-923
                 URL: https://issues.apache.org/jira/browse/ZOOKEEPER-923
             Project: Zookeeper
          Issue Type: Task
          Components: recipes
         Environment: Any
            Reporter: kishore gopalakrishna


A typical use case in distributed system is " There are T tasks and P processes running but only T processes must be active always [ P > T ]  and remaining P-T processes acting as stand by and be ready to take up a Task with one or more active processes fail".  

Zookeeper provides an excellent service which can be used to co ordinate among P processes and using the mechanism of locking we can ensure that there is always T processes active.  Without a central co-ordinating service generally there will be 2T processes[ i.e atleast one back up for each process].  With Zookeeper we can decide P based on the failure rate.

The assumption here are
1. At any time we have P > T. P can be chosen appropriately based on failure rate.
2. The tasks are stateless. That is any process P_i that takes up a task T_j does not know the state of the process P_k which previously processed T_j.  This is not entirely true and there are ways to over come this draw back on a case by case basis.


This was developed for a different project S4 which is also open sourced http://s4.io/. The communication layer and task management layer is completely independent of S4 and can be used in any application.





-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.