[
https://issues.apache.org/jira/browse/HDFS-9228?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14991785#comment-14991785
]
Bob Hansen commented on HDFS-9228:
----------------------------------
By my reading of the Java code
(hadoop-common/src/main/java/org/apache/hadoop/ipc/Client.java), the key
requirement is simply setting the retry count on the PB message. If the server
has already received and processed a message with that client ID : call ID, it
will respond with a success (and/or the appropriate data).
Does this seem fundamentally sound? Do we need to do any processing on the
response reading side?
Since the state of the request is stored in the RpcConnection object, I will
put the retry count there. To keep things together, I think this requires
moving the connect/reconnect and retry logic from the RpcEngine to the
RpcConnection.
> libhdfs++ should respect NN retry configuration settings
> --------------------------------------------------------
>
> Key: HDFS-9228
> URL: https://issues.apache.org/jira/browse/HDFS-9228
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Components: hdfs-client
> Reporter: Bob Hansen
>
> Handle the use case of temporary network or NN hiccups and have a
> configurable number of retries for NN operations.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
