hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Íñigo Goiri (JIRA) <j...@apache.org>
Subject [jira] [Commented] (HDFS-12778) [READ] Report multiple locations for PROVIDED blocks
Date Tue, 14 Nov 2017 18:28:00 GMT

    [ https://issues.apache.org/jira/browse/HDFS-12778?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16251896#comment-16251896

Íñigo Goiri commented on HDFS-12778:

Can we add explicit unit tests for corner cases like:
# Only one DN with PROVIDED
# Less DNs with PROVIDED than default replication
# More blocks than DNs (trying to fill the excluded nodes)

Should we make the block locations deterministic to some degree?
I can see two mappers trying to accessing the same block and in that way some caching could
be done.

> [READ] Report multiple locations for PROVIDED blocks
> ----------------------------------------------------
>                 Key: HDFS-12778
>                 URL: https://issues.apache.org/jira/browse/HDFS-12778
>             Project: Hadoop HDFS
>          Issue Type: Sub-task
>            Reporter: Virajith Jalaparti
>            Assignee: Virajith Jalaparti
>         Attachments: HDFS-12778-HDFS-9806.001.patch
> On {{getBlockLocations}}, only one Datanode is returned as the location for all PROVIDED
blocks. This can hurt the performance of applications which typically 3 locations per block.
We need to return multiple Datanodes for each PROVIDED block for better application performance/resilience.

This message was sent by Atlassian JIRA

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

View raw message