You are viewing a plain text version of this content. The canonical link for it is here.
Posted to common-commits@hadoop.apache.org by Apache Wiki <wi...@apache.org> on 2011/04/01 12:55:16 UTC

[Hadoop Wiki] Update of "RMAvailability" by VinodKumarVavilapalli

Dear Wiki user,

You have subscribed to a wiki page or wiki category on "Hadoop Wiki" for change notification.

The "RMAvailability" page has been changed by VinodKumarVavilapalli.
The comment on this change is: Putting some comments.
http://wiki.apache.org/hadoop/RMAvailability?action=diff&rev1=2&rev2=3

--------------------------------------------------

    i. Application Id, state of the application (RUNNING/PENDING etc), Application Master Information – This involves persisting the latest state, application master information
    i. The containers that have been allocated per application – This means we have to persist every container allocation in the system
    i. The Hostname to NodeID map and the capacity of each node – This means we have to persist once when a NodeManager registers and remove it when a NodeManager is expired
+         -- '''If we make NodeID as hashes of hostname:port, then we don't need that map. Capacity of the nodes can be obtained from the heartbeat, no? -vinodkv'''
  
  Note that we are not persisting the nodemanager to container map. This map can be reconstructed with the application to containers map. Each container in the system has information on which nodemanager it belongs to. So it becomes easy to create the nodemanager to container map. 
+         -- '''May be instead of trusting the AM, we should obtain the container map from the nodemanager when it registers back. -vinodkv'''
  
  Also, the available/used information can be derived from (i) to (iii).