[
https://issues.apache.org/jira/browse/HDFS-5951?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13907653#comment-13907653
]
Suresh Srinivas commented on HDFS-5951:
---------------------------------------
I think the scope of this jira is probably misunderstood. The proposal is not
to do away the monitoring systems. Frequently I see many issues that can be
flagged from HDFS itself. To name a few:
# Configuration issues
#* Using /tmp for storage
#* For a given size of the cluster getting ipc handler count wrong, number of
datanode transceivers wrong, and ulimit for daemons wrong etc.
#* JVM heap size misconfiguration for the size of the cluster and for the
number of the objects etc.
# Flag issues that need to be addressed, which sometimes is missed even with
monitoring in place, where alerts are categorized incorrectly or were ignored.
#* Checkpoints not happening (I know instances where missing this has resulted
in startup times of clusters over 18 hours!)
#* Growth in editlog size or editlog.
#* Corruption in fsimage and editlog checkpointing silently ignored.
Some of these are covered in best practices documents that vendors put out or
in hadoop operations related tech talks. Some of them can be covered in this
WebUI where issues described can be flagged, with information on why it needs
to be addressed and how to address it.
> Provide diagnosis information in the Web UI
> -------------------------------------------
>
> Key: HDFS-5951
> URL: https://issues.apache.org/jira/browse/HDFS-5951
> Project: Hadoop HDFS
> Issue Type: Bug
> Reporter: Haohui Mai
> Assignee: Haohui Mai
> Attachments: HDFS-5951.000.patch, diagnosis-failure.png,
> diagnosis-succeed.png
>
>
> HDFS should provide operation statistics in its UI. it can go one step
> further by leveraging the information to diagnose common problems.
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
