hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jing Zhao (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDFS-7412) Move RetryCache to NameNodeRpcServer
Date Fri, 21 Nov 2014 19:51:33 GMT

    [ https://issues.apache.org/jira/browse/HDFS-7412?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14221340#comment-14221340

Jing Zhao commented on HDFS-7412:

Looks like in our current implementation we have two patterns: 1) doing OperationCategory
check before retry cache check, or 2) checking retry cache before the OperationCategory check.
The difference here is that for 1) if the client side's retry happens during the NN failover,
the client has to failover to the other NN to get the result stored in retry cache, which
may cause issue if the client has already hit the max failover times. The current patch makes
sure all the calls follow pattern 2), which should be a good change.

After the change the FSNamesystem calls have to take an extra argument to indicate whether
the retry cache is available. Since this is only used to determine whether to record call/client
id in the edits, maybe we can let fseditlog check a globally visible marker or even a threadlocal
context, so that we can hide this logic away from the FSNamesystem APIs. But this can be future
work addressed by a separate jira.

> Move RetryCache to NameNodeRpcServer
> ------------------------------------
>                 Key: HDFS-7412
>                 URL: https://issues.apache.org/jira/browse/HDFS-7412
>             Project: Hadoop HDFS
>          Issue Type: Sub-task
>            Reporter: Haohui Mai
>            Assignee: Haohui Mai
>         Attachments: HDFS-7412.000.patch
> The concept of RetryCache belongs to the RPC layer.It would be nice to separate it from
the implementation of {{FSNameSystem}}.

This message was sent by Atlassian JIRA

View raw message