[
https://issues.apache.org/jira/browse/HADOOP-7455?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13079058#comment-13079058
]
Suresh Srinivas commented on HADOOP-7455:
-----------------------------------------
@Justin
I am not sure if you need an HA state called "Neutral". It is standby (backup
node) in disconnected state. It is still a standby in my opinion. It is just
that until it catches up with the active, it cannot become active. Becoming
from standby to active means possible loss of data.
bq. In the doc attached for HDFS-1623, it says 'It is important that only the
active NN responds to the client'. It achieves the same, but has dependency on
the fencing done to the shared storage; when a nonâactive NN tries to write to
the fenced shared storage it will fail & the response to client will indicate
failure.
I am not sure where in doc you are referring to. Fencing is needed during
network partition. If not shared stored, you do not want clients/datanodes to
be talking to different nodes which each think is active.
@Florian
This is not specific to Linux-HA. It would be good to get your input on this
jira.
> Introduce HA Service Protocol Interface
> ---------------------------------------
>
> Key: HADOOP-7455
> URL: https://issues.apache.org/jira/browse/HADOOP-7455
> Project: Hadoop Common
> Issue Type: Sub-task
> Components: util
> Reporter: Suresh Srinivas
> Assignee: Suresh Srinivas
> Attachments: HADOOP-7455.1.patch, HADOOP-7455.2.patch,
> HADOOP-7455.3.patch, HADOOP-7455.4.patch, HDFS-7454.patch
>
>
> This jira introduces a protocol interface to be implemented by services that
> provide HA functionality.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
