Thomas Weise 2012-06-22, 22:10
Bikas Saha 2012-06-22, 22:29
Thomas Weise 2012-06-22, 22:37
Arun C Murthy 2012-06-25, 00:46
Finger, Jay 2012-06-28, 17:21
-Re: Hadoop 2.0 / YARN - getting full node report in application master
Here's the javadoc for getUpdatedNodes(). It says that the return value is
the delta of the updated nodes since the last response. So the API gives
one the difference between the updated nodes since the last time one
One could augment the doc to clarify that nothing will be returned in the
first call to allocate(). Also, it will help everyone if the wording of
the doc could be made more clear. So I would gladly change that based on
suggestions. Or you could do it yourself.
* Get the list of updated <code>NodeReport</code>s. Updates
* changes in health, availability etc of the nodes.
* @return The delta of updated nodes since the last response
On 6/28/12 10:21 AM, "Finger, Jay" <[EMAIL PROTECTED]> wrote:
>AMResponse.getUpdatedNodes returns nodes that have been updated
>_since_when_ ? The doc comments leave too much to the imagination in that
>A reasonable assumption could be that the first time you call it that you
>get a full snapshot.
>Presumably the second (and later) calls return updates since the previous
>call, but even that isn't clear.
>On 6/22/12 3:29 PM, "Bikas Saha" <[EMAIL PROTECTED]> wrote:
>>AMResponse.getUpdatedNodes() would give you only nodes that have been
>>updated. Say they are lost or become healthy again. So you won't get
>>cluster information from that.
>>On 6/22/12 3:10 PM, "Thomas Weise" <[EMAIL PROTECTED]> wrote:
>>>I'm looking for a way to initially obtain a full report of cluster
>>>nodes in the application master. That would be the equivalent of
>>>ClientRMService.getClusterNodes (which seems to be intended for use by
>>>the client only?)
>>>If my understanding is correct AMResponse.getUpdatedNodes would give
>>>me deltas that can be used subsequently to maintain the list?