[
https://issues.apache.org/jira/browse/HDFS-4974?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13708864#comment-13708864
]
Suresh Srinivas commented on HDFS-4974:
---------------------------------------
bq. I don't believe DatanodeProtocol#blockReceivedAndDeleted is safe to retry,
due to things like modifying the pending replication count.
Current processing, I agree is not retry safe. This is not being done using
FSNamesystem. I will move this to FSNamesystem and add retry cache support.
I still need to double check if blockReport() is really idempotent. Especially
with the difference in processing firstBlockReport and the subsequent one.
bq. ClientProtocol#metaSave opens its output file for append, so retries would
cause duplication of data in the output file.
Given the entire metasave is done with writeLock() held, will this be an issue?
Currently one could have two metasaves issued by two clients right? Given that
I think it should be okay to mark it as idempotent?
bq. Is it correct that ClientProtocol#rollEdits is idempotent, but
NamenodeProtocol#rollEditLog is non-idempotent? The two methods are nearly
identical. They both call FSNamesystem#rollEditLog, which I believe is
idempotent.
I agree. We should mark NamenodeProtocol#rollEditLog as idempotent as well.
> Analyze and add annotations to Namenode and Datanode protocol methods to
> enable retry
> -------------------------------------------------------------------------------------
>
> Key: HDFS-4974
> URL: https://issues.apache.org/jira/browse/HDFS-4974
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Components: ha, namenode
> Reporter: Suresh Srinivas
>
> This jira is intended for:
> # Discussing current @Idempotent annotations in HDFS protocols and adding
> that annotation where it is missing.
> # Discuss how retry should be enabled for non-idempotent requests.
> I will post the analysis of current methods in subsequent comment.
--
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
