[
https://issues.apache.org/jira/browse/HADOOP-13831?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15749632#comment-15749632
]
Gaurav Kanade commented on HADOOP-13831:
----------------------------------------
Thanks [[email protected]] ! So I have run all relevant tests during package
compilation (number of tests passing before and after the patch is the same).
In addition, I have created a custom jar with the same patch on my hdp cluster
and did basic sanity testing. This works. The specific case of testing the
functionality of the patch pertains to detecting server failure condition which
is hard to simulate without long running test, hence as such this is not
particularly tested. I believe we have had similar patches in the past such as
HADOOP-12334 and HADOOP-11693 which followed a similar template. If there is
any additional testing that needs to / can be done please give me some idea and
I will explore it. The patch is fairly straightforward - a simple change in
conditional check - test for http status code 503 instead of relying on storage
error code - and was recommended by azure storage team.
Now adding [~dchickabasapa], [~onpduo] [~liuml07] and [~nitinverma] for review
and further input
> Correct check for error code to detect Azure Storage Throttling and provide
> retries
> -----------------------------------------------------------------------------------
>
> Key: HADOOP-13831
> URL: https://issues.apache.org/jira/browse/HADOOP-13831
> Project: Hadoop Common
> Issue Type: Bug
> Components: fs/azure
> Affects Versions: 2.7.3
> Reporter: Gaurav Kanade
> Assignee: Gaurav Kanade
> Attachments: HADOOP-13831.001.patch
>
>
> Azure Storage throttling affects HBase operations such as archiving old
> WALS and others. In such cases the storage driver needs to detect and handle
> the exception. We put in this logic to do the retries however the condition
> to check for the exception is not always met due to inconsistency in which
> the manner the error code is passed back. Instead the retry logic should
> check for http status code (503) which is more reliable and consistent check
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
