[
https://issues.apache.org/jira/browse/YARN-231?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13503363#comment-13503363
]
Bikas Saha commented on YARN-231:
---------------------------------
At this point I have tried to make the logic fail safe keeping in mind imminent
HA work where load/store operations need to make sure that partial operations
dont happen. Errors could either be because of store error or loss of master
status. Loss of master status may actually manifest as a store error in many
cases. Hence I have chosen to bail out on every error in the store. Of course,
this may change to allow partial reads if such optimizations are deemed safe.
Does that help clarify?
Thats also partly responsible for why I am currently throwing Exception
everywhere. I am yet not sure about all different exceptions that might be
validly thrown by different store implementations. In the near term, I do
expect that these might narrow down at least to IOException for a pure store
error (e.g. NN is unavailable) and HAException that would clearly show that
loss of master status has occurred. Until some preliminary HA work provides
more inputs to this end, I thought I would leave the Exceptions generic for
now. All interfaces are marked Evolving for this reason.
My next patch should address the remaining feedback and any other new ones that
you might have.
> Add persistent store implementation for RMStateStore
> ----------------------------------------------------
>
> Key: YARN-231
> URL: https://issues.apache.org/jira/browse/YARN-231
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: resourcemanager
> Reporter: Bikas Saha
> Assignee: Bikas Saha
> Attachments: YARN-231.1.patch
>
>
> Add stores that write RM state data to ZooKeeper and FileSystem
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
