[ 
https://issues.apache.org/jira/browse/HDFS-11888?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16137173#comment-16137173
 ] 

Chen Liang commented on HDFS-11888:
-----------------------------------

Thanks [~xyao] for taking care of this! Looks pretty good to me overall. Also 
got some minor comments.

# {{BlockManagerImpl#allocateBlock:353}} LOG.info to LOG.debug maybe? because 
seems every single block allocation will trigger one log line.
# SCMContainerInfo -> SCMContainerStateInfo?
# I found this field "used" of a container a bit ambiguous. Because I think 
there is no guarantee that this allocated block will actually be used, i.e. 
client may fail somehow after allocation, right? So to me maybe a better name 
can be "allocated"?

And one clarification question of mine:
I saw there are two types of objects, pipeline and container. What is the 
difference of these two terms here?

> Ozone: SCM: use container state machine for open containers allocated for 
> key/blcoks 
> -------------------------------------------------------------------------------------
>
>                 Key: HDFS-11888
>                 URL: https://issues.apache.org/jira/browse/HDFS-11888
>             Project: Hadoop HDFS
>          Issue Type: Sub-task
>          Components: ozone
>    Affects Versions: HDFS-7240
>            Reporter: Xiaoyu Yao
>            Assignee: Xiaoyu Yao
>         Attachments: HDFS-11888-HDFS-7240.001.patch
>
>
> SCM BlockManager provision a pool of containers upon block creation request 
> that can't be satisfied with current open containers in the pool. However, 
> only one container is returned with the creationFlag to the client. The other 
> container provisioned in the same batch will not have this flag. Client can't 
> assume to use these containers that has not been created on SCM datanodes, 
> This ticket is opened to fix the issue by persist the createContainerNeeded 
> flag for the provisioned containers. The flag will be eventually cleared by 
> processing container report from datanode when the container report handler 
> is fully implemented on SCM. 
> For now, we will use a default batch size of 1 for 
> ozone.scm.container.provision_batch_size so that the container will be 
> created on demand upon the first block allocation into the container. 



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to