[
https://issues.apache.org/jira/browse/HDFS-1623?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13070611#comment-13070611
]
Ted Dunning commented on HDFS-1623:
-----------------------------------
{quote}
I guess it helps to quantify how "small" this window needs to be and whether
current ZK is able to provide the notification fast enough. if not, maybe
implementing the ZAB protocol as part of the namenode/backup-nodenode
communication ? (in that case it would be nice if ZK exports its protocol as a
library)
{quote}
The problem isn't on the Zookeeper side, really.
The problem is that you have to be able to detect failures quickly, but not
capriciously. Pretty much the only valid way to do this is with heartbeats of
some kind which means that the time between heartbeats is the shortest response
time possible. Moreover, the monitored program has to be sure that it can meet
the real-time constraint imposed by the heartbeat rate. With the NameNode and
possible GC, this is really, really hard to do for short heartbeats periods.
There is perhaps some mileage to be had to synchronizing any fencing with the
heartbeat so that you at least get rid of that uncertainty.
Overall, this is a hard problem which is the reason for protocols like
two-phase commit.
> High Availability Framework for HDFS NN
> ---------------------------------------
>
> Key: HDFS-1623
> URL: https://issues.apache.org/jira/browse/HDFS-1623
> Project: Hadoop HDFS
> Issue Type: New Feature
> Reporter: Sanjay Radia
> Assignee: Sanjay Radia
> Attachments: HDFS-High-Availability.pdf, NameNode HA_v2.pdf, NameNode
> HA_v2_1.pdf, Namenode HA Framework.pdf
>
>
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
