[
http://issues.apache.org/jira/browse/HADOOP-237?page=comments#action_12420694 ]
David Bowen commented on HADOOP-237:
------------------------------------
This code is not using the metrics API as intended, in that it calls the update
method after each metric modification. The API is record-oriented, so update
copies the whole record to the client library.
I don't think that this will cause significant, observable problems with the
metric data, but it could be a significant performance issue.
The preferred model would be to replace per-metric methods like
void mapInput(long numBytes)
void mapOutput(long numBytes)
with something like
void mapIO(long numBytesInput, long numBytesOutput)
and have this only call the update method once.
> Standard set of Performance Metrics for Hadoop
> ----------------------------------------------
>
> Key: HADOOP-237
> URL: http://issues.apache.org/jira/browse/HADOOP-237
> Project: Hadoop
> Type: Improvement
> Components: metrics
> Versions: 0.3.0
> Environment: All
> Reporter: Milind Bhandarkar
> Assignee: Milind Bhandarkar
> Attachments: hadoop-metrics.patch
>
> I am starting to use Hadoop's shiny new Metrics API to publish performance
> (and other) Metrics of running jobs and other daemons.
> Which performance metrics are people interested in seeing ? If possible,
> please group them according to modules, such as map-reduce, dfs,
> general-cluster-related etc. I will follow this process:
> 1. collect this list
> 2. assess feasibility of obtaining metric
> 3. assign context/record/metrics names
> 4. seek approval for names
> 5. instrument the code.
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see:
http://www.atlassian.com/software/jira
