Mailing-List: contact commits-help@cassandra.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@cassandra.apache.org
Date: Fri, 21 Feb 2014 03:41:31 +0000 (UTC)
From: "graham sanderson (JIRA)" <jira@apache.org>
To: commits@cassandra.apache.org
Message-ID: <JIRA.12693587.1391702611046.78985.1392954091271@arcas>
In-Reply-To: <JIRA.12693587.1391702611046@arcas>
References: <JIRA.12693587.1391702611046@arcas>
Subject: [jira] [Comment Edited] (CASSANDRA-6666) Avoid accumulating
 tombstones after partial hint replay
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit


    [ https://issues.apache.org/jira/browse/CASSANDRA-6666?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13907926#comment-13907926 ] 

graham sanderson edited comment on CASSANDRA-6666 at 2/21/14 3:40 AM:
----------------------------------------------------------------------

We have seen an issue where we get into the >100K tombstone state with the hints table also; It has happened more than once (worked around in 2.0.5 with manual compaction on all nodes of system.hints). 

One quick question... this patch will hopefully prevent us getting into this state in many cases, unless you have a huge number of hints for a node that is down for a very long time: since there is no auto-compaction, a large number of hints may have expired from TTL, thus preventing any further hint delivery.

This is OK if you bring the node back up as a fresh one, otherwise, I guess suggesting manual compaction is fine, or we could do a compact().get() on seeing a TombstoneOverwhelmingException? (just in this hint code path)


was (Author: graham sanderson):
We have seen an issue where we get into the >100K tombstone state with the hints table also; It has happened more than once (worked around in 2.0.5 with manual compaction on all nodes of system.hints). 

One quick question... this patch will hopefully prevent us getting into this state in many cases, unless you have a huge number of hints for a node that is down for a very long time: since there is no auto-compaction, a large number of hints may have expired from TTL, thus preventing any further hint delivery.

This is OK if you bring the node back up as a fresh one, otherwise, I guess suggesting manual compaction is fine, or we could do a compact().get() on seeing a TombstoneOverwhelmingException?

> Avoid accumulating tombstones after partial hint replay
> -------------------------------------------------------
>
>                 Key: CASSANDRA-6666
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-6666
>             Project: Cassandra
>          Issue Type: Bug
>          Components: Core
>            Reporter: Jonathan Ellis
>            Assignee: Jonathan Ellis
>            Priority: Minor
>              Labels: hintedhandoff
>             Fix For: 1.2.16, 2.0.6
>
>         Attachments: 6666.txt
>
>


--
This message was sent by Atlassian JIRA
(v6.1.5#6160)