cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jonathan Ellis (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-5051) Allow automatic cleanup after gc_grace
Date Fri, 01 Mar 2013 15:07:13 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-5051?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13590604#comment-13590604
] 

Jonathan Ellis commented on CASSANDRA-5051:
-------------------------------------------

v2 attached that makes cleanup-during-compaction always-on.  (v2 also switches from comparing
based on table name to checking for LocalStrateg, and moves indexColumns list internal to
rmIdxRenewCounter.)

this highlights a problem, though -- some tests now fail, because StorageService.getLocalRanges
will return an empty list until StorageService initializes it.  (I think this is either via
loading the stored ring through initServer, or via gossip filling things in.  Not really sure
how gossip fills in my own token if we don't load from the system table.  Bit of a mess here.)

So (a) obviously this is a bit fragile for the tests.  But is there potential for us to also
throw away data that we shouldn't if we get behind on gossip somehow?  At the very least I
think we need to include pending ranges for the local node.

Nit: ISTM we ought to be able to move LCR's Iterable<OnDiskAtom> into AbstractCompactedRow,
but I'm not actually sure how to make generics happy w/ the PR subclass returning Iterator<Column>.
                
> Allow automatic cleanup after gc_grace
> --------------------------------------
>
>                 Key: CASSANDRA-5051
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-5051
>             Project: Cassandra
>          Issue Type: New Feature
>          Components: Core
>            Reporter: Brandon Williams
>            Assignee: Vijay
>              Labels: vnodes
>             Fix For: 2.0
>
>         Attachments: 0001-CASSANDRA-5051.patch, 5051-v2.txt
>
>
> When using vnodes, after adding a new node you have to run cleanup on all the machines,
because you don't know which are affected and chances are it was most if not all of them.
 As an alternative to this intensive process, we could allow cleanup during compaction if
the data is older than gc_grace (or perhaps some other time period since people tend to use
gc_grace hacks to get rid of tombstones.)

--
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