[jira] [Commented] (RATIS-270) Replication ALL requests should not be replied from retry cache if they are delayed.

["Mukul Kumar Singh (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/RATIS-270?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16576145#comment-16576145 ] 

Mukul Kumar Singh commented on RATIS-270:
-----------------------------------------

Thanks for working on this [~szetszwo]. The patch looks good to me, Some minor comments as following.

1) RaftServerImpl:1069, lets rename updateCache to isReplyDelayed, I feel that this will help with the comments as well, also lets add a note that for delayed replies, the retryCache will be updated as part of RaftClientReply#getReply
2) RetryCacheTests.java:205, lets add a small note inside the fail statement.
3) Inside RetryCacheTests.java, when the first set of request fail, can we retry with another round of same requests being sent, and making sure that they are blocking on the failed node and not reading from the retry cache of the leader.

> Replication ALL requests should not be replied from retry cache if they are delayed.
> ------------------------------------------------------------------------------------
>
>                 Key: RATIS-270
>                 URL: https://issues.apache.org/jira/browse/RATIS-270
>             Project: Ratis
>          Issue Type: Bug
>          Components: server
>            Reporter: Mukul Kumar Singh
>            Assignee: Tsz Wo Nicholas Sze
>            Priority: Major
>              Labels: ozone
>         Attachments: r270_20180803.patch
>
>
> Retry requests are answered from the retry cache when requests have Replication_ALL semantics. This leads to a case, where the client retries for a response which is stuck in the delayed replies queue. This new retry is now answered from the retry cache even though the request has not been completed on all the nodes.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)