[jira] [Commented] (CASSANDRA-7069) Prevent operator mistakes due to simultaneous bootstrap

["Anthony Grasso (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/CASSANDRA-7069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15948129#comment-15948129 ] 

Anthony Grasso commented on CASSANDRA-7069:
-------------------------------------------

{quote} Disccusing this offline with Jake, we decided it's still possible to violate consistent range movement even following the 2 minute rule, so leaving this as-is. If people don't care, they can simply disable consistent range movement. {quote}

[~brandon.williams] and [~tjake] just wondering how you both worked out it was possible to violate consistent range movement even after following the 2 minute rule? Is it possible for tokens assigned to a first bootstrapping node to then be reassigned to a second bootstrapping node? In which case does the second bootstrapping node stream from the first. If so, can see how bootstrapping in parallel is an issue.

> Prevent operator mistakes due to simultaneous bootstrap
> -------------------------------------------------------
>
>                 Key: CASSANDRA-7069
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-7069
>             Project: Cassandra
>          Issue Type: New Feature
>            Reporter: Brandon Williams
>            Assignee: Brandon Williams
>            Priority: Minor
>             Fix For: 2.1.1, 2.2.0 beta 1
>
>         Attachments: 7069.txt
>
>
> Cassandra has always had the '2 minute rule' between beginning topology changes to ensure the range announcement is known to all nodes before the next one begins.  Trying to bootstrap a bunch of nodes simultaneously is a common mistake and seems to be on the rise as of late.
> We can prevent users from shooting themselves in the foot this way by looking for other joining nodes in the shadow round, then comparing their generation against our own and if there isn't a large enough difference, bail out or sleep until it is large enough.



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)