Hello
(sorry for cross post)
I am reading the DistributedShell code and came across

 GetClusterMetricsRequest clusterMetricsReq Records.newRecord(GetClusterMetricsRequest.class);
 GetClusterMetricsResponse clusterMetricsResp applicationsManager.getClusterMetrics(clusterMetricsReq);

In the page: http://hadoop.apache.org/docs/current/api/org/apache/hadoop/yarn/api/ClientRMProtocol.html#getClusterMetrics%28org.apache.hadoop.yarn.api.protocolrecords.GetClusterMetricsRequest%29

I can't find any other function getClusterMetrics.

Why was the choice made to pass a GetClusterMetricsRequest object to
getClusterMetrics?
why couldn't the programmer invoke
applicationsManager.getClusterMetrics() and the method would reply
with a GetClusterMetricsResponse?

Regards
Sapsi
(P.S I havent used YARN or even written code with it, i'm just reading
some sources and my reading is incomplete, so apologies if this comes
across
as stupid)

Sapsi,

I am not positive on this but I think the reason for this is to future
proof the API.  If in the future we want to add in new optional
parameters, I believe that it is impossible in protocol buffers without
having that request object. I could be wrong I am not an expert on PB.

--Bobby Evans

On 10/9/12 11:11 PM, "Saptarshi Guha" <[EMAIL PROTECTED]> wrote:

>Hello
>(sorry for cross post)
>I am reading the DistributedShell code and came across
>
> GetClusterMetricsRequest clusterMetricsReq >Records.newRecord(GetClusterMetricsRequest.class);
> GetClusterMetricsResponse clusterMetricsResp >applicationsManager.getClusterMetrics(clusterMetricsReq);
>
>In the page:
>http://hadoop.apache.org/docs/current/api/org/apache/hadoop/yarn/api/Clien
>tRMProtocol.html#getClusterMetrics%28org.apache.hadoop.yarn.api.protocolre
>cords.GetClusterMetricsRequest%29
>
>I can't find any other function getClusterMetrics.
>
>Why was the choice made to pass a GetClusterMetricsRequest object to
>getClusterMetrics?
>why couldn't the programmer invoke
>applicationsManager.getClusterMetrics() and the method would reply
>with a GetClusterMetricsResponse?
>
>Regards
>Sapsi
>(P.S I havent used YARN or even written code with it, i'm just reading
>some sources and my reading is incomplete, so apologies if this comes
>across
>as stupid)

This is to help evolve the APIs easily. All apis have request and response objects, instead of passing in parameters directly. Some of the APIs clearly may not need this extra layer, but doing so helps uniformity.

Irrespective of that, following the original goal, GetClusterMetricsResponse itself could have all the fields, instead of one more nested ClusterMetrics object.

Thanks,
+Vinod

On Oct 9, 2012, at 9:11 PM, Saptarshi Guha wrote:

> Hello
> (sorry for cross post)
> I am reading the DistributedShell code and came across
>
> GetClusterMetricsRequest clusterMetricsReq > Records.newRecord(GetClusterMetricsRequest.class);
> GetClusterMetricsResponse clusterMetricsResp > applicationsManager.getClusterMetrics(clusterMetricsReq);
>
> In the page: http://hadoop.apache.org/docs/current/api/org/apache/hadoop/yarn/api/ClientRMProtocol.html#getClusterMetrics%28org.apache.hadoop.yarn.api.protocolrecords.GetClusterMetricsRequest%29
>
> I can't find any other function getClusterMetrics.
>
> Why was the choice made to pass a GetClusterMetricsRequest object to
> getClusterMetrics?
> why couldn't the programmer invoke
> applicationsManager.getClusterMetrics() and the method would reply
> with a GetClusterMetricsResponse?
>
> Regards
> Sapsi
> (P.S I havent used YARN or even written code with it, i'm just reading
> some sources and my reading is incomplete, so apologies if this comes
> across
> as stupid)