[
https://issues.apache.org/jira/browse/HDFS-11564?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15951471#comment-15951471
]
Xiaoyu Yao commented on HDFS-11564:
-----------------------------------
Thanks [~anu] for working on this. The patch looks good to me overall. I just
have few questions below:
SCMCommonPolicy.java
Line 150: Can you add javadoc for the purpose of this public method
getResultSet()?
Maybe we can consolidate chooseDatanodes() and getResultSet()? It seems the
former consider both numNodes and size and the later consider only numNodes.
SCMContainerPlacementCapacity.java
line 96-99: do we attempt to handle the case where we can't find enough ndoe
with required capacity
different from SCMCommonPolicy? If yes, we will need to catch the exception
like UNABLE_FIND_NODES_WITH_SPACE
thrown from super.chooseDatanodes() in line 95.
SCMContainerPlacementRandom.java
line 70:73: similar to above.
NIT: Define separate class for Remaining/ScmUsed/Capacity is bit
over-engineering. Is it possible to just use LongMetric.
SCMException.java
Line 99: NIT: NODE_NOT_FOUND_IN_NODEPOOL -> FAILED_TO_FIND_NODE_IN_POOL to be
consistent
Line 101: NIT: UNABLE_FIND_NODES_WITH_SPACE -> FAILED_TO_FIND_NODES_WITH_SPACE
to be consistent
Line 103: NIT: CAPACITY_IS_ZERO: should we define INVALID_CAPACITY which covers
both 0 and negative cases.
> Ozone: SCM: Add Comparable Metric Support
> -----------------------------------------
>
> Key: HDFS-11564
> URL: https://issues.apache.org/jira/browse/HDFS-11564
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Components: ozone
> Affects Versions: HDFS-7240
> Reporter: Anu Engineer
> Assignee: Anu Engineer
> Attachments: HDFS-11564-HDFS-7240.001.patch,
> HDFS-11564-HDFS-7240.002.patch, HDFS-11564-HDFS-7240.003.patch
>
>
> While picking the datanode or the pool to place a container, SCM needs to be
> able to compare pools and datanodes. This jira proposes to add a Comparable
> Metric interface from which DatanodeCapacity and DatanodeIOLoad metrics can
> be derived. This will also convert SCM Container Placement Capacity policy to
> use these metrics.
> Adding these metrics makes all placement policy operate on the basis of these
> metrics and it makes it easy to simulate cluster imbalances as well as reason
> about code. All new SCM metrics which have the property of total ordering
> should be derived from this interface.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
