[
https://issues.apache.org/jira/browse/HADOOP-13321?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15428670#comment-15428670
]
Chris Nauroth commented on HADOOP-13321:
----------------------------------------
[[email protected]], are you suggesting that we don't in fact deprecate the
APIs? I'd be comfortable with that if we mitigate in other ways, such as clear
warnings in JavaDocs about the potential inefficiencies.
bq. If the FS client does some very short term caching, even a fraction of a
second, the penalty of two back-to-back getFileStatus() checks would become
zero ... that may be the way to go.
Linking to HADOOP-12876, which tracks an implementation of this idea in Azure
Data Lake. I'd like to explore refactoring that out to Hadoop Common for any
file system to use.
> Deprecate FileSystem APIs that promote inefficient call patterns.
> -----------------------------------------------------------------
>
> Key: HADOOP-13321
> URL: https://issues.apache.org/jira/browse/HADOOP-13321
> Project: Hadoop Common
> Issue Type: Improvement
> Components: fs
> Reporter: Chris Nauroth
> Assignee: Mingliang Liu
> Attachments: HADOOP-13321.000.patch
>
>
> {{FileSystem}} contains several methods that act as convenience wrappers over
> calling {{getFileStatus}} and retrieving a single property of the returned
> {{FileStatus}}. These methods have a habit of fostering inefficient call
> patterns in applications, resulting in multiple redundant {{getFileStatus}}
> calls. For HDFS, this translates into wasteful NameNode RPC traffic. For
> file systems backed by cloud object stores, this translates into wasteful
> HTTP traffic. This issue proposes to deprecate these methods and instead
> encourage applications to call {{getFileStatus}} and then reuse the same
> {{FileStatus}} instance as needed.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
