[
https://issues.apache.org/jira/browse/HADOOP-13227?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Tsz Wo Nicholas Sze updated HADOOP-13227:
-----------------------------------------
Attachment: c13227_20160606.patch
> In RetryInfo#newRetryInfo, looks like failover, fail, and retry are mutual
> exclusive? ...
It is correct that the end result is mutually exclusive. However, we need to
loop all of the actions in order to determine which one to keep. Indeed, we
may combine the failover and fail in RetryInfo to a single action. Let me
change it.
> ... consider directly using ConcurrentLinkedQueue which utilizes an efficient
> non-block algorithm.
It is a good idea.
> In checkCalls, do you think we can avoid the poll+offer operations for a
> not-done-yet call?
I think it is hard to avoid. We don't want to read the non-head elements from
the queue since it is an O( n) operation, where n is the size of the queue.
Poll and offer indeed are cheap for linked queue. Let me know if you have an
idea to avoid poll+offer.
Here is a new patch:
c13227_20160606.patch
> AsyncCallHandler should use a event driven architecture to handle async calls
> -----------------------------------------------------------------------------
>
> Key: HADOOP-13227
> URL: https://issues.apache.org/jira/browse/HADOOP-13227
> Project: Hadoop Common
> Issue Type: Improvement
> Components: io, ipc
> Reporter: Tsz Wo Nicholas Sze
> Assignee: Tsz Wo Nicholas Sze
> Attachments: c13227_20160602.patch, c13227_20160606.patch
>
>
> This JIRA is to address [Jing's
> comments|https://issues.apache.org/jira/browse/HADOOP-13226?focusedCommentId=15308630&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-15308630]
> in HADOOP-13226.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
