[
https://issues.apache.org/jira/browse/HDFS-7702?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14297745#comment-14297745
]
Ray commented on HDFS-7702:
---------------------------
Is the transfer granularity blockpool only?
This is the last step(Clean Up), after metadata flip done, the target namenode
will only transfer one blockpool id to datanodes, then datanode won't have to
report the same block to both namenodes. So from namenode to namenode, transfer
namespace sub-tree, from namenode to datanode only transfer blockpool id and
block id.
but then this statement:
it will mark delete the involved sub-tree from its own namespace
leads me to believe that it's sub-trees in the namespace.
Ok, this one is from the previous step(Metadata Flip), it won't transfer
anything. The metadata exist on the source namenode, so just mark it delete,
not really delete, just remove reference, in case we have to roll back.
Could you please clarify this statement:
all read and write operation regarding the same namespace sub-tree is
forwarding to the target namenode.
Who does the forwarding, the client or the source NN?
The source NN will give info whether metadata on other NN and the client will
do the forwarding, I haven't mention the overflowing table yet, client will get
metadata from namenode first, the metadata include some data structure point
out some path in other namenodes. Then client will connect to all other
namenodes (if have, might only have previous step don't have to do this one )
in parallel, and get all metadata merged together.
Please feel free to tell me if anything still not clear.
Thank you for your first response!
> Move metadata across namenode - Effort to a real distributed namenode
> ---------------------------------------------------------------------
>
> Key: HDFS-7702
> URL: https://issues.apache.org/jira/browse/HDFS-7702
> Project: Hadoop HDFS
> Issue Type: New Feature
> Reporter: Ray
> Assignee: Ray
>
> Implement a tool can show in memory namespace tree structure with
> weight(size) and a API can move metadata across different namenode. The
> purpose is moving data efficiently and faster, without moving blocks on
> datanode.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
