You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@cassandra.apache.org by "Jonathan Halliday (JIRA)" <ji...@apache.org> on 2014/08/22 16:12:12 UTC
[jira] [Reopened] (CASSANDRA-7810) tombstones gc'd before being
locally applied
[ https://issues.apache.org/jira/browse/CASSANDRA-7810?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Jonathan Halliday reopened CASSANDRA-7810:
------------------------------------------
sorry guys, still seeing this intermittently even after an env cleanup. I suspect it's linked to the merging of sstables. If I flush the data and tombstones together to a single sstable and compact then it's fine. If I flush the data and the tombstones separately such that I have two sstables, then compact them, it goes wrong. Any chance someone could try that modified process and see if it's reproducible? thx.
> tombstones gc'd before being locally applied
> --------------------------------------------
>
> Key: CASSANDRA-7810
> URL: https://issues.apache.org/jira/browse/CASSANDRA-7810
> Project: Cassandra
> Issue Type: Bug
> Environment: 2.1.0.rc6
> Reporter: Jonathan Halliday
> Assignee: Marcus Eriksson
> Fix For: 2.1.0
>
> Attachments: range_tombstone_test.py
>
>
> # single node environment
> CREATE KEYSPACE test WITH replication = {'class': 'SimpleStrategy', 'replication_factor': 1 };
> use test;
> create table foo (a int, b int, primary key(a,b));
> alter table foo with gc_grace_seconds = 0;
> insert into foo (a,b) values (1,2);
> select * from foo;
> -- one row returned. so far, so good.
> delete from foo where a=1 and b=2;
> select * from foo;
> -- 0 rows. still rainbows and kittens.
> bin/nodetool flush;
> bin/nodetool compact;
> select * from foo;
> a | b
> ---+---
> 1 | 2
> (1 rows)
> gahhh.
> looks like the tombstones were considered obsolete and thrown away before being applied to the compaction? gc_grace just means the interval after which they won't be available to remote nodes repair - they should still apply locally regardless (and do correctly in 2.0.9)
--
This message was sent by Atlassian JIRA
(v6.2#6252)