cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Eric Evans (JIRA)" <>
Subject [jira] Commented: (CASSANDRA-385) intellibootstrap
Date Tue, 06 Oct 2009 23:02:31 GMT


Eric Evans commented on CASSANDRA-385:

This looks good to me, but I do have one question/concern.

Introduced in 0006-get-token-on-bootstrap-that-gives-us-half-of-the-keys.txt, when a node
bootstraps and hasn't been assigned a token, it selects one via StorageService.getBootstrapTokenFrom
which uses Thrift to retrieve it as a property. 

My concern is that this is the first time we've used Thrift for anything that wasn't client
oriented. Do we want to start now, or does it make sense to preserve that separation?

If nothing else, sticking with this approach will mean that work is needed to improve how
 the Thrift service is bound to interfaces, (currently configured via ThriftAddress), because
there is only one option that is guaranteed safe with this patch applied (

> intellibootstrap
> ----------------
>                 Key: CASSANDRA-385
>                 URL:
>             Project: Cassandra
>          Issue Type: Bug
>          Components: Core
>            Reporter: Jonathan Ellis
>            Assignee: Jonathan Ellis
>             Fix For: 0.5
>         Attachments: 0001-CASSANDRA-385-clean-up-loadinfo-and-SLB.txt, 0002-make-LoadInfo-a-double-instead-of-going-from-long-s.txt,
0003-move-javadoc-into-mbean.-inline-methods-where-wrappin.txt, 0004-use-Strings-instead-of-Endpoints-in-jmx-methods.-merge.txt,
0005-add-getLoadMap-jmx-method-add-load-info-to-nodeprobe.txt, 0006-get-token-on-bootstrap-that-gives-us-half-of-the-keys.txt
> ideally bootstrap mode should determine its new position on the ring by examining the
Load of the existing nodes in the cluster.  (currently Load is just disk space used but making
this pluggable is another ticket.)  having found the highest-load-node-that-is-not-participating-in-bootstrap,
it should ask that node for a Token which would move half the keys over to the new node.
> This is easily computed since we have a periodic sampling of keys in memory of all the
keys on disk, and even SSTable.getIndexedKeys that merges all such keys.  So pick the midpoint,
and turn it into a token (these are decorated keys so that is always possible).

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

View raw message