reef-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Julia (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (REEF-1453) GroupCommunication should create a new observer for each client
Date Fri, 17 Jun 2016 19:54:05 GMT

    [ https://issues.apache.org/jira/browse/REEF-1453?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15336805#comment-15336805
] 

Julia commented on REEF-1453:
-----------------------------

For 4, I mean if we are to use driver to control sending event to tasks to close them, do
we really care about if *TransportServer.ProcessClient() loop and *TransportClient.ResponseLoop()
cannot read message and throw exceptions? What we want to make sure is to stop those threads
after receiving the close event from driver.  

> GroupCommunication should create a new observer for each client
> ---------------------------------------------------------------
>
>                 Key: REEF-1453
>                 URL: https://issues.apache.org/jira/browse/REEF-1453
>             Project: REEF
>          Issue Type: Improvement
>          Components: REEF.NET
>            Reporter: Andrew Chung
>            Assignee: Andrew Chung
>              Labels: FT
>
> {{GroupCommunication}} currently only has one universal observer that handles connections
for all incoming clients. This is inconvenient because when a client fails or disconnects,
there is no easy way to propagate the failure/completion signal of the *specific* failed client
up to the universal observer.
> Provided that each {{CommunicationGroupClient}} knows all its downstream nodes and can
reverse-lookup the endpoints, the {{CommunicationGroupClient}} should only create an observer
for each endpoint instead of using the universal observer.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message