[
https://issues.apache.org/jira/browse/HDFS-10898?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15543417#comment-15543417
]
Bob Hansen commented on HDFS-10898:
-----------------------------------
What would you think of keeping SEEK and READ at the TRACE level since
consumers will tend to do those gazillions of times?
We should move the async FileHandleImpl::PositionRead to DEBUG also.
While we're here, can we make the INFO level message for Connect be more
human-readable ("Connecting to <server name>:<service>"), perhaps with a DEBUG
level that includes debugging-appropriate data (this pointer, etc.). Have that
one only in the async version, so the consumer gets one INFO-level message per
connect. Not critical, but since whe're here...
> libhdfs++: Make logs more informative and consistent
> ----------------------------------------------------
>
> Key: HDFS-10898
> URL: https://issues.apache.org/jira/browse/HDFS-10898
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Components: hdfs-client
> Reporter: James Clampffer
> Assignee: James Clampffer
> Priority: Trivial
> Attachments: HDFS-10898.HDFS-8707.000.patch
>
>
> Most of the public C++ FileHandle/FileSystem operations have a LOG_TRACE
> level message about parameters passed in etc. However many methods use
> LOG_DEBUG and a couple use LOG_INFO.
> We most likely want FS operations that happen a lot (read/open/seek/stat) to
> stick to LOG_DEBUG consistently and only use LOG_INFO for things like
> FileSystem::Connect or RpcConnection:: that don't get called often and are
> important enough to warrant showing up in the log. LOG_TRACE can be reserved
> for things happening deeper inside public methods and methods that aren't
> part of the public API.
> Related improvements that could be brought into this to avoid opening a ton
> of small Jiras:
> -Print the "this" pointer address in the log message to make it easier to
> correlate objects when there's concurrent work being done. This has been
> very helpful in the past but often got stripped out before patches went in.
> People just need be aware that operator new may eventually place an object of
> the same type at the same address sometime in the future.
> -For objects owned by other objects, but created on the fly, include a
> pointer back to the parent/creator object if that pointer is already being
> tracked (See the nested stucts in BlockReaderImpl).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
