hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Josh Elser (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-17534) SecureBulkLoadClient squashes DoNotRetryIOExceptions from the server
Date Mon, 01 May 2017 16:46:04 GMT

    [ https://issues.apache.org/jira/browse/HBASE-17534?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15991069#comment-15991069
] 

Josh Elser commented on HBASE-17534:
------------------------------------

bq. It looks like the same issue is present in master, could you make a version of the patch
for that branch?

Actually, I don't think this is an issue for master. Since the secure bulk load implementation
is now using the standard client RPC model in Master (e.g. RegionServerCallable, RpcRetryingFactory)
instead of the coprocessor RPC implementation, this works correctly.

Let me just work on rebasing the patch for the 1.x lines.

> SecureBulkLoadClient squashes DoNotRetryIOExceptions from the server
> --------------------------------------------------------------------
>
>                 Key: HBASE-17534
>                 URL: https://issues.apache.org/jira/browse/HBASE-17534
>             Project: HBase
>          Issue Type: Bug
>          Components: Client
>            Reporter: Josh Elser
>            Assignee: Josh Elser
>             Fix For: 2.0.0, 1.4.0, 1.2.6, 1.3.2, 1.1.11
>
>         Attachments: HBASE-17534.001.branch-1.patch, HBASE-17534.002.branch-1.patch,
HBASE-17534.003.branch-1.patch
>
>
> While writing some tests against 1.x, I noticed that what should have been a DoNotRetryIOException
sent to the client from a RegionServer was getting retried until it reached the hbase client
retries limit.
> Upon inspection, I found that the SecureBulkLoadClient was wrapping all Exceptions from
the RPC as an IOException. I believe this is creating a case where the RPC system doesn't
notice that there's a DNRIOException wrapped beneath it, thinking it's a transient error.
> This results in clients having to wait for the retry limit to be reached before they
get acknowledgement that something failed.



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

Mime
View raw message