[
https://issues.apache.org/jira/browse/HDDS-981?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16744763#comment-16744763
]
Lokesh Jain commented on HDDS-981:
----------------------------------
Other issues not addressed in the patch -
# Synchronization might be required in
ContainerStateManager#getMatchingContainer to avoid multiple threads using the
same container.
# Need to automatically allocate containers to have configured number of
containers with atleast some free space. For a busy cluster it is possible to
have many clients receiving blocks on the same set of containers. It is
essential to have a strategy where ClosedContainerExceptions can be minimized
during writes.
> Block allocation should involve pipeline selection and then container
> selection
> -------------------------------------------------------------------------------
>
> Key: HDDS-981
> URL: https://issues.apache.org/jira/browse/HDDS-981
> Project: Hadoop Distributed Data Store
> Issue Type: Bug
> Components: SCM
> Reporter: Lokesh Jain
> Assignee: Lokesh Jain
> Priority: Major
> Attachments: HDDS-981.001.patch
>
>
> Currently SCM maintains a list of preallocated containers and allocates
> blocks from these containers. This approach does not work well with dynamics
> of the cluster where new nodes are being added and pipelines are destroyed.
> New containers are not created until all the preallocated containers are
> exhausted.
> The Jira aims to establish a criteria in block allocation where first a
> pipeline is selected amongst the available pipelines and then a container is
> selected in that pipeline. In order to handle the dynamics of the cluster a
> fixed interval pipeline creator job can be launched which creates pipelines
> in the system.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
