[
https://issues.apache.org/jira/browse/HDFS-13110?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Rakesh R updated HDFS-13110:
----------------------------
Description:
This task is to address the following [~daryn]'s comments. Please refer
HDFS-10285 to see more detailed discussion.
*Comment-10)*
NamenodeProtocolTranslatorPB
Most of the api changes appear unnecessary.
IntraSPSNameNodeContext#getFileInfo swallows all IOEs, based on assumption that
any and all IOEs means FNF which probably isn’t the intention during rpc
exceptions.
*Comment-13)*
StoragePolicySatisfier
It appears to make back-to-back calls to hasLowRedundancyBlocks and
getFileInfo for every file. Haven’t fully groked the code, but if low
redundancy is not the common case, then it shouldn’t be called unless/until
needed. It looks like files that are under replicated are re-queued again?
was:
This task is to address [~daryn] comments.
*Comment No.10)*
NamenodeProtocolTranslatorPB
Most of the api changes appear unnecessary.
IntraSPSNameNodeContext#getFileInfo swallows all IOEs, based on assumption that
any and all IOEs means FNF which probably isn’t the intention during rpc
exceptions.
> [SPS]: Reduce the number of APIs in NamenodeProtocol used by external
> satisfier
> -------------------------------------------------------------------------------
>
> Key: HDFS-13110
> URL: https://issues.apache.org/jira/browse/HDFS-13110
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Reporter: Rakesh R
> Priority: Major
> Attachments: HDFS-13110-HDFS-10285-00.patch
>
>
> This task is to address the following [~daryn]'s comments. Please refer
> HDFS-10285 to see more detailed discussion.
> *Comment-10)*
> NamenodeProtocolTranslatorPB
> Most of the api changes appear unnecessary.
> IntraSPSNameNodeContext#getFileInfo swallows all IOEs, based on assumption
> that any and all IOEs means FNF which probably isn’t the intention during rpc
> exceptions.
> *Comment-13)*
> StoragePolicySatisfier
> It appears to make back-to-back calls to hasLowRedundancyBlocks and
> getFileInfo for every file. Haven’t fully groked the code, but if low
> redundancy is not the common case, then it shouldn’t be called unless/until
> needed. It looks like files that are under replicated are re-queued again?
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
