hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Enis Soztutar (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-17556) The client will not invalidate stale region caches
Date Fri, 27 Jan 2017 23:39:24 GMT

    [ https://issues.apache.org/jira/browse/HBASE-17556?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15843664#comment-15843664
] 

Enis Soztutar commented on HBASE-17556:
---------------------------------------

Some of the fixes are here: 
{code}
commit 94d576025fd158dae03dd6a6b50a62483b0f04e0
Author: Mikhail Antonov <antonov@apache.org>
Date:   Wed Mar 16 16:32:11 2016 -0700

    HBASE-15390 Unnecessary MetaCache evictions cause elevated number of requests to meta

commit 43698b3fcb6b6759c5acae3b889e6b280c2205bd
Author: Mikhail Antonov <antonov@apache.org>
Date:   Mon Mar 7 17:33:43 2016 -0800

    HBASE-15354 Use same criteria for clearing meta cache for all operations (Ashu Pachauri)
commit d747188f2cb7d198393a35f4dde543a798fed3c8
Author: Josh Elser <elserj@apache.org>
Date:   Mon Feb 8 14:25:37 2016 -0500

    HBASE-15232 Handle region location cache mgmt in AsyncProcess for multi()'s
{code}

Maybe you can take a look at the branch-1 or master code to see whether the issue is still
there. A unit test that can reproduce would be really useful if you have the time. 

> The client will not invalidate stale region caches
> --------------------------------------------------
>
>                 Key: HBASE-17556
>                 URL: https://issues.apache.org/jira/browse/HBASE-17556
>             Project: HBase
>          Issue Type: Bug
>          Components: Client
>    Affects Versions: 1.0.0, 2.0.0, 0.98.24
>            Reporter: Marcin Januszkiewicz
>            Priority: Critical
>
> We noticed in our application, that sometimes when we interact with a table an operation
will fail with an exception, an all operations that happen on the same region will also fail
until the application is restarted.
> It seems that when a merge or split happens on a region that is already in the clients
cache, and the client is configured to retry operations, then there is no way for the client
to detect this. In RpcRetryingCaller#callWithRetries if a call fails with RegionNotServingException
then the cache will be cleared only if the retry parameter is equal to 1. This means the call
will fail but the following calls will succeed.
> RpcRetryingCaller#callWithoutRetries contains the comment "It would be nice to clear
the location cache here". Additionally, the stale cache will cause this call to fail, even
though the data is available.
> See also HBASE-12534



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

Mime
View raw message