[
https://issues.apache.org/jira/browse/HDFS-12809?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16252025#comment-16252025
]
Virajith Jalaparti commented on HDFS-12809:
-------------------------------------------
Hi [~arpitagarwal], This issue is specified to the Provided storage abstraction
(HDFS-9806) and is not related to HDFS-11419. In HDFS-9806, when
{{getBlockLocations}} is called on a file that has replicas with StorageType
{{PROVIDED}}, datanodes (with {{PROVIDED}} storagetype) are chosen at random as
locations for these replicas. It doesn't use the {{BlockPlacementPolicy}} for
this purpose.
> [READ] Fix the randomized selection of locations in {{ProvidedBlocksBuilder}}.
> ------------------------------------------------------------------------------
>
> Key: HDFS-12809
> URL: https://issues.apache.org/jira/browse/HDFS-12809
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Reporter: Virajith Jalaparti
>
> Calling {{getBlockLocations}} on files that have a PROVIDED replica, results
> in the datanode locations being selected at random. Currently, this
> randomization uses the datanode uuids to pick a node at random
> ({{ProvidedDescriptor#choose}}, {{ProvidedDescriptor#chooseRandom}}).
> Depending on the distribution of the datanode UUIDs, this can lead to large
> number of iterations (which may not terminate) before a location is chosen.
> This JIRA aims to replace this with a more efficient randomization strategy.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
