[ 
https://issues.apache.org/jira/browse/HADOOP-13031?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15414339#comment-15414339
 ] 

Mingliang Liu commented on HADOOP-13031:
----------------------------------------

[~mingma], in this JIRA, I plan to separate the distance-specific rack-aware 
read bytes logic from {{FileSystemStorageStatistics}} to a new HDFS-specific 
class {{DFSRackAwareStorageStatistics}}, as commented in [HADOOP-13032].

We need to preserve the thread-local mechanism for doing this, tracked by 
[HADOOP-13435]. I submitted a patch for [HADOOP-13435], and any 
comments/reviews are very welcomed. After that is committed, I think the 
[HADOOP-13032] should be almost done as I also attached the initial patch. So 
after those two blockers, I will consolidate the effort of consuming the 
StorageStatistics with this JIRA. As you suggested, MR should probe before 
dumping the rack distance-aware read bytes so that undefined counters will not 
display.

> Rack-aware read bytes stats should be managed by HFDS specific 
> StorageStatistics
> --------------------------------------------------------------------------------
>
>                 Key: HADOOP-13031
>                 URL: https://issues.apache.org/jira/browse/HADOOP-13031
>             Project: Hadoop Common
>          Issue Type: Sub-task
>          Components: fs
>            Reporter: Mingliang Liu
>            Assignee: Mingliang Liu
>
> [HADOOP-13065] added a new interface for retrieving FS and FC Statistics. 
> This jira is to refactor the code that maintains rack-aware read metrics to 
> use the newly added StorageStatistics. Specially,
> # Rack-aware read bytes metrics is mostly specific to HDFS. For example, 
> local file system doesn't need that. We consider to move it from base 
> FileSystemStorageStatistics to a dedicated HDFS specific StorageStatistics 
> sub-class.
> # We would have to develop an optimized thread-local mechanism to do this, to 
> avoid causing a performance regression in HDFS stream performance.
> Optionally, it would be better to simply move this to HDFS's existing 
> per-stream {{ReadStatistics}} for now. As [HDFS-9579] states, ReadStatistics 
> metrics are only accessible via {{DFSClient}} or {{DFSInputStream}}. Not 
> something that application framework such as MR and Tez can get to.



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

---------------------------------------------------------------------
To unsubscribe, e-mail: common-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: common-issues-h...@hadoop.apache.org

Reply via email to