[
https://issues.apache.org/jira/browse/HDFS-8210?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15217081#comment-15217081
]
Anu Engineer commented on HDFS-8210:
------------------------------------
[~cnauroth] Thanks for the patch. It looks excellent. There was one thing that
is not very clear to me.
# if Ozone front end requests a key with full path , say ->
getStorageContainerLocations("/Volume/Bucket/key.1") , the
StorageContainerManager might only know about a part of this key, That is it is
quite possible that SCM knows only about /Volume and which container it is in.
It would be nice to get that Prefix portion that you matched back in the
{{LocatedContainer}}. So that front end knows what was the validated part. For
example, if we are using Range Partition, and SCM matched /Volume/Bucket/Key.0
to a container location, then that is interesting information.
# Probably a subject matter for a later patch, we might need a delete RPC , so
that ozone front end can notify SCM that a particular entity in the ozone
namespace has been deleted, Since SCM maintains only ranges, it would allow SCM
to update the ranges. This is not strictly required since ozone front end
assumes that SCM is only giving a hint.
# nitpick: I see that your {{StorageContainerNameService#hasReadLock}} has a
wider interpretation than containerManager under ozone. We should probably sync
on this, if hasReadLock is true even if you have writelock is the methodology
that we are following, then I will fix in the containerManager code. Just
wanted to flag it so that you are aware.
> Ozone: Implement storage container manager
> -------------------------------------------
>
> Key: HDFS-8210
> URL: https://issues.apache.org/jira/browse/HDFS-8210
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Components: ozone
> Reporter: Jitendra Nath Pandey
> Assignee: Chris Nauroth
> Attachments: HDFS-8210-HDFS-7240.006.patch,
> HDFS-8210-HDFS-7240.1.patch, HDFS-8210-HDFS-7240.2.patch,
> HDFS-8210-HDFS-7240.3.patch, HDFS-8210-HDFS-7240.4.patch,
> HDFS-8210-HDFS-7240.5.patch
>
>
> The storage container manager collects datanode heartbeats, manages
> replication and exposes API to lookup containers. This jira implements
> storage container manager by re-using the block manager implementation in
> namenode. This jira provides initial implementation that works with
> datanodes. The additional protocols will be added in subsequent jiras.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
