cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jason Brown (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (CASSANDRA-5128) Stream hints on decommission
Date Fri, 18 Jan 2013 14:40:13 GMT

     [ https://issues.apache.org/jira/browse/CASSANDRA-5128?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Jason Brown updated CASSANDRA-5128:
-----------------------------------

    Attachment: 5128-v3.patch

Moved the logic of finding an alive host into a new method, streamHints(), which is called
by unbootstrap(). 

Also found a minor bug in Jonathan's patch:

{code}List<InetAddress> candidates = new ArrayList<InetAddress>(StorageService.instance.getTokenMetadata().cloneAfterAllLeft().getAllEndpoints());{code}

TM.cloneAfterAllLeft() calls cloneOnlyTokenMap(), which sets up the tokenToEndpointMap field
but not the endpointToHostIdMap. Then getAllEndpoints() is called on new TM, which refers
to an empty endpointToHostIdMap map. Thus, in cloneOnlyTokenMap() I've added setting up the
endpointToHostIdMap. I think it should be reasonably harmless.
                
> Stream hints on decommission
> ----------------------------
>
>                 Key: CASSANDRA-5128
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-5128
>             Project: Cassandra
>          Issue Type: Improvement
>          Components: Core
>    Affects Versions: 1.2.1
>            Reporter: Jason Brown
>            Assignee: Jason Brown
>              Labels: decommission, hints
>         Attachments: 0001-CASSANDRA-5128-stream-hints-on-decommission.patch, 5128-v2.txt,
5128-v3.patch
>
>
> Looks like decommissioning a node (nodetool decommission) will stream all the non-system
table data out to it's appropriate peers (StorageService.unbootstrap()), but hints will disappear
with the node. Let's send those hints to another peer (preferably in the same rack, and hopefully
at least the same datacenter) them to be replayed.
> The use case here is auto-scaling vnode clusters in ec2. When auto-scaling down, I'd
want to call decommission on the node to leave the ring (and be terminated), and still have
all of it's artifacts (data and hints) survive.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Mime
View raw message