[
https://issues.apache.org/jira/browse/HDFS-5304?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13789867#comment-13789867
]
Andrew Wang commented on HDFS-5304:
-----------------------------------
There's a mismatch if we still want to use {{boolean[]}} in the RPC format and
{{DatanodeInfo[]}} in {{LocatedBlock}}. It means the RPC layer has to map from
{{DI->booleans}} and back, which is problematic if you mistakenly add a cached
location that isn't also in the list of locations.
Maybe we can do something mildly complicated with both a {{repeated boolean}}
for cached locations that are in the {{locs}} array, and then a {{repeated
DatanodeInfoProto}} for anything that's not. The in-memory state of the
LocatedBlock will still be references when possible, and it's a nice future
proofing for if/when we have non-disk-backed caching.
> Expose if a block replica is cached in getFileBlockLocations
> ------------------------------------------------------------
>
> Key: HDFS-5304
> URL: https://issues.apache.org/jira/browse/HDFS-5304
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Components: namenode
> Affects Versions: HDFS-4949
> Reporter: Andrew Wang
> Assignee: Andrew Wang
> Attachments: hdfs-5304-1.patch
>
>
> We need to expose which replicas of a block are cached so applications can
> place their tasks for memory-locality.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
