cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Sylvain Lebresne (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-3216) A streamOutSession keeps sstables references forever if the remote end dies
Date Mon, 19 Sep 2011 06:54:08 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-3216?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13107652#comment-13107652
] 

Sylvain Lebresne commented on CASSANDRA-3216:
---------------------------------------------

It would, it's the same problem than with CASSANDRA-2433. But the patch the technique we've
use there to wait to reach twice the normal phi convict threshold to hopefully make sure this
won't happen. But we can increase the threshold even higher if we think that doubling doesn't
put us enough on the safe side.

> A streamOutSession keeps sstables references forever if the remote end dies
> ---------------------------------------------------------------------------
>
>                 Key: CASSANDRA-3216
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-3216
>             Project: Cassandra
>          Issue Type: Bug
>          Components: Core
>    Affects Versions: 1.0.0
>            Reporter: Sylvain Lebresne
>            Assignee: Sylvain Lebresne
>            Priority: Minor
>              Labels: streaming
>             Fix For: 1.0.0
>
>         Attachments: 3216.patch
>
>
> A streamOutSession acquire a reference on the sstable it will stream and release them
as soon as each sstable has been fully streamed. However, since a stream session has currently
no means to know when it failed, we'll keep references indefinitely (meaning until next restart)
if their is a failure. One way a stream session could very easily fail is if the remote end
dies. We must make sure we correctly release sstable references when that happens.
> Note that it won't be bulletproof, there is probably other means by which a streaming
could fail: a bug in the code throwing an exception, no space left on the receiving end, etc...
But those are unlikely enough that I propose to care only for the case of a node dying for
now and leave the bullet-proofing to CASSANDRA-3112. 

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira

        

Mime
View raw message