[jira] [Updated] (YARN-3399) Consider having a Default cluster ID

2015-08-03 Thread Brahma Reddy Battula (JIRA)

 [ 
https://issues.apache.org/jira/browse/YARN-3399?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Brahma Reddy Battula updated YARN-3399:
---
Assignee: (was: Brahma Reddy Battula)

> Consider having a Default cluster ID
> 
>
> Key: YARN-3399
> URL: https://issues.apache.org/jira/browse/YARN-3399
> Project: Hadoop YARN
>  Issue Type: Bug
>  Components: resourcemanager
>Reporter: Zhijie Shen
>
> In YARN-3040, timeline service will set the default cluster ID if users don't 
> provide one. RM HA's current behavior is a bit different when users don't 
> provide cluster ID. IllegalArgumentException will throw instead. Let's 
> continue the discussion if RM HA needs the default cluster ID or not here, 
> and what's the proper default cluster ID.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)


[jira] [Updated] (YARN-3399) Consider having a Default cluster ID

2015-03-26 Thread Vinod Kumar Vavilapalli (JIRA)

 [ 
https://issues.apache.org/jira/browse/YARN-3399?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Vinod Kumar Vavilapalli updated YARN-3399:
--
Summary: Consider having a Default cluster ID  (was: Default cluster ID for 
RM HA)

Editing title to be appropriate.

Others commented on YARN-3040. So I'll try to summarize the discussion from 
YARN-1029 and YARN-3040.
 - We should have a generic {{yarn.cluster-id}} and deprecate the current RM 
only configuration
 - We need to have a reasonable default cluster-id
-- This is needed for the Timeline service functionality - we want to 
gather insights per cluster
-- Forcing admins to set the ID explicitly is one more hurdle w.r.t 
configuration
-- For single node non-HA clusters, forcing the dev/admin to set it is 
unnecessary.
 - But there are concerns too
-- Default cluster-id can potentially cause hard-to-debug issues in HA mode.
 - Other constraints while picking a default cluster ID
-- Restarting RM on the same node shouldn't change the cluster-id

So, I propose that we set the default cluster-ID to be something like 
"default-$(RM-host-name)-cluster". This way
 - by default, single node clusters are good across RM restarts, unless you are 
running active/standby RMs on the same machine (dev environments)
 - HA RMs have to be setup explicitly to be part of the same cluster - thereby 
avoiding debuggability issues.
 - For real life use, in order to facilitate RM migrations, administrators will 
set their own cluster-id.

> Consider having a Default cluster ID
> 
>
> Key: YARN-3399
> URL: https://issues.apache.org/jira/browse/YARN-3399
> Project: Hadoop YARN
>  Issue Type: Bug
>  Components: resourcemanager
>Reporter: Zhijie Shen
>Assignee: Brahma Reddy Battula
>
> In YARN-3040, timeline service will set the default cluster ID if users don't 
> provide one. RM HA's current behavior is a bit different when users don't 
> provide cluster ID. IllegalArgumentException will throw instead. Let's 
> continue the discussion if RM HA needs the default cluster ID or not here, 
> and what's the proper default cluster ID.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)