[
https://issues.apache.org/jira/browse/HDDS-896?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16732469#comment-16732469
]
Anu Engineer commented on HDDS-896:
-----------------------------------
[~nandakumar131]
+1, Thanks for a very good patch, Feel free to commit at will. I have some
minor nits. leaving those here since I wrote them down already.
# nit: ReplicaManager.java: Add some info traces when we start delete/replica
operations? It is kind of useful to have these statements in the log.
# nit: ReplicaManager.java: There are some classes that has a Java doc
comment that says -- Add JavaDoc. :)
# A future JIRA suggestion-- The time that we collect in
CloseContainerCommandHandler.java, it might be good to report back to SCM. That
way we can build a average time to close and std.dev etc for this operation.
Same comment for other commands in the DN.
# It is new year and I am just back from a long vacation. Hence a trick
question for you. Probably not something that we need to worry at all, I am
just thinking aloud. Feel free to ignore this comment completely. Is it
possible for the SCM to be confused and send conflicting commands to a DN. For
example, what happens if I send -- CloseContainer, ReplicateContainer and
DeleteContainer together? Is the data node guaranteed to process only one
command at a time? should we introduce the same kind of locking that we have in
SCM inside a datanode ? A locking is needed, if and only if we take the
position that we will allow completely parallel operations to proceed in
parallel, Then we need to worry about making a mistake and having these threads
pick up a command that conflicts. Let us talk some time. Please note that this
comment has no bearing at all on this patch.
> Handle over replicated containers in SCM
> ----------------------------------------
>
> Key: HDDS-896
> URL: https://issues.apache.org/jira/browse/HDDS-896
> Project: Hadoop Distributed Data Store
> Issue Type: Improvement
> Components: SCM
> Reporter: Nanda kumar
> Assignee: Nanda kumar
> Priority: Major
> Attachments: HDDS-896.000.patch, HDDS-896.001.patch
>
>
> When SCM detects that a container is over-replicated, it has to delete some
> replicas to bring the number of replicas to match the required value. If the
> container is in QUASI_CLOSED state, we should check the {{originNodeId}}
> field while choosing the replica to delete.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
