cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Peter Schuller (Issue Comment Edited) (JIRA)" <j...@apache.org>
Subject [jira] [Issue Comment Edited] (CASSANDRA-3897) StorageService.onAlive() only schedules hints for joined endpoints
Date Mon, 13 Feb 2012 01:09:00 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-3897?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13206600#comment-13206600
] 

Peter Schuller edited comment on CASSANDRA-3897 at 2/13/12 1:08 AM:
--------------------------------------------------------------------

Why would this be the case? They are supposed to receive writes; I see no reason why hints
should not be delivered. Hints is just a way to more quickly delivery writes in cases where
nodes are down (i.e., more quickly when they go up) and avoid AES need. I don't see why a
node actively bootstrapping into the ring should be "discriminated" against, in terms of seeing
as reliable delivery of writes as other nodes.

In other words, I don't buy your first sentence unless you explain why. I don't accept it
axiomatically :)

Obviously sending hints requires that hints are *there* first too, but the same argument applies.
If a node is supposed to see a certain writes and it's considered down - hint it.

Statistically I can see the argument that if a node is bootstrapping and down, it might be
practically more likely that the node is just going to be down for a longer period, and/or
that the node will completely re-bootstrap anyway (since normally a node is down because it's
being restarted, which would imply re-bootstrap if the node is bootstrapping).

                
      was (Author: scode):
    Why would this be the case? They are supposed to receive writes; I see no reason why hints
should not be delivered. Hints is just a way to more quickly delivery writes in cases where
nodes are down (i.e., more quickly when they go up) and avoid AES need. I don't see why a
node actively bootstrapping into the ring should be "discriminated" against, in terms of seeing
as reliable delivery of writes as other nodes.

In other words, I don't by your first sentence unless you explain why. I don't accept it axiomatically
:)

Obviously sending hints requires that hints are *there* first too, but the same argument applies.
If a node is supposed to see a certain writes and it's considered down - hint it.

Statistically I can see the argument that if a node is bootstrapping and down, it might be
practically more likely that the node is just going to be down for a longer period, and/or
that the node will completely re-bootstrap anyway (since normally a node is down because it's
being restarted, which would imply re-bootstrap if the node is bootstrapping).

                  
> StorageService.onAlive() only schedules hints for joined endpoints
> ------------------------------------------------------------------
>
>                 Key: CASSANDRA-3897
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-3897
>             Project: Cassandra
>          Issue Type: Sub-task
>          Components: Core
>            Reporter: Peter Schuller
>            Assignee: Peter Schuller
>            Priority: Minor
>
> It seems incorrect to not do hint delivery for nodes that are bootstrapping, as that
would cause sudden spikes in read repair need or inconsistent reads when a node joins the
ring. Particularly if the user is expecting to rely on the new hinted handoff code making
AES much less needed. It would be a POLA violation for bootstrapping nodes to be an exception
to that.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira

        

Mime
View raw message