hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Allen Wittenauer (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDFS-11300) HDFS Nameservices introduces DNS latency inter-dependencies between namenodes
Date Fri, 06 Jan 2017 23:38:58 GMT

    [ https://issues.apache.org/jira/browse/HDFS-11300?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15806162#comment-15806162
] 

Allen Wittenauer commented on HDFS-11300:
-----------------------------------------

bq. A solution to fix this issue is to fix the upfront resolution and do it in a lazy way.

Or, what most large sites do, deploy DNS caching servers on every node.  This greatly improves
performance in general.

> HDFS Nameservices introduces DNS latency inter-dependencies between namenodes
> -----------------------------------------------------------------------------
>
>                 Key: HDFS-11300
>                 URL: https://issues.apache.org/jira/browse/HDFS-11300
>             Project: Hadoop HDFS
>          Issue Type: Improvement
>          Components: hdfs, hdfs-client
>    Affects Versions: 2.6.0
>            Reporter: Rémy SAISSY
>
> When using HDFS Nameservices, a DNS outage or strong latency on  one of nameservices
will impact any DFSClient or other component which has this nameservice in its hdfs-site.xml.
Even though it doesn't use it.
> The issue is due to all nameservices being listed in hdfs-site.xml, and hadoop code trying
to resolve all of them upfront.
> To ilustrate the issue, here is the use case we have:
> * the 'prod' cluster with its dedicated DNS
> * the 'preprod' cluster with its dedicated DNS
> * both 'prod' and 'preprod'  hdfs-site.xml files have each other nameservices so that
they can use HDFS as follows:
> (from prod) hdfs dfs -ls hdfs://preprod/user/j.doe
> A DNS outage on the preprod cluster slow down production jobs being scheduled because
of the upfront resolution of the hdfs-site.xml nameservices entries even though a specific
job doesn't use it.
> A solution to fix this issue is to fix the upfront resolution and do it in a lazy way.



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

---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-help@hadoop.apache.org


Mime
View raw message