[jira] [Commented] (CASSANDRA-8591) Tunable bootstrapping

["Sylvain Lebresne (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/CASSANDRA-8591?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14273410#comment-14273410 ] 

Sylvain Lebresne commented on CASSANDRA-8591:
---------------------------------------------

The thing is, tunable consistency doesn't mean that C* is free to randomly fail the consistency guarantees that you ask. If we were to allow this, that is what would happen: some queries woud fail the consistency level you ask it to do. I have a hard time seeing that as an acceptable behavior, unless maybe (maybe) the option is very clearly labeled as {{I_willingly_authorise_cassandra_to_give_up_all_consistency_guarantee_from_now_on_and_until_I_ve_run_a_full_repair}}. Open to see other opinions, but _a priori_ I feel reticent about the concept.

> Tunable bootstrapping
> ---------------------
>
>                 Key: CASSANDRA-8591
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-8591
>             Project: Cassandra
>          Issue Type: Improvement
>            Reporter: Donald Smith
>            Priority: Minor
>             Fix For: 3.0
>
>
> Often bootstrapping fails due to errors like "unable to find sufficient sources for streaming range". But cassandra is supposed to be fault tolerant, and it's supposed to have tunable consistency.  
> If it can't find sources for some ranges, it should allow bootstrapping to continue and should print out a report about what ranges were missing.   Allow the bootstrap to be tunable, under control of parameters ("allow up to 100 failures", for example).
> For many apps, it's far better to bootstrap what's available then to fail flat.
> Same with rebuilds.
> We were doing maintenance on some disks, and when we started cassandra back up, some nodes ran out of disk space, due to operator miscalculation. Thereafter, we've been unable to bootstrap new nodes, due to "unable to find sufficient sources for streaming range."  But bootstrapping with partial success would be far better than being unable to bootstrap at all, and cheaper than a repair. Our consistency requirements aren't high but we prefer as much consistency as cassandra can give us.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)