Mailing-List: contact issues-help@hbase.apache.org; run by ezmlm
Precedence: bulk
Date: Sat, 20 May 2017 14:23:04 +0000 (UTC)
From: "Yu Li (JIRA)" <jira@apache.org>
To: issues@hbase.apache.org
Message-ID: <JIRA.13073531.1495219687000.255857.1495290184480@Atlassian.JIRA>
In-Reply-To: <JIRA.13073531.1495219687000@Atlassian.JIRA>
References: <JIRA.13073531.1495219687000@Atlassian.JIRA> <JIRA.13073531.1495219687900@jira-lw-us.apache.org>
Subject: [jira] [Commented] (HBASE-18083) Make large/small file clean thread
 number configurable in HFileCleaner
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit
archived-at: Sat, 20 May 2017 14:23:09 -0000


    [ https://issues.apache.org/jira/browse/HBASE-18083?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16018477#comment-16018477 ] 

Yu Li commented on HBASE-18083:
-------------------------------

bq. Last time, we found tcp nagle issue with 0.98 based releases, which caused about 120 ms delay for continuous deletes
Any JIRA opened for this one?

bq. How fast is the cleaner thread finish one run
Pretty fast in our case, and no obvious change in NN metrics after we opened 25 threads for each (large and small file deleting)

> Make large/small file clean thread number configurable in HFileCleaner
> ----------------------------------------------------------------------
>
>                 Key: HBASE-18083
>                 URL: https://issues.apache.org/jira/browse/HBASE-18083
>             Project: HBase
>          Issue Type: Bug
>            Reporter: Yu Li
>            Assignee: Yu Li
>
> Currently we have only one thread for both large and small file cleaning, but when write pressure is huge we might need more cleaner threads, so we need to make the thread number configurable.
> We observed more than 1.8PB data in archive directory online due to business access rate change, and this proposal is one of the necessary changes required.
> Default value of the configurations would still be left to 1 to keep low pressure to NN for normal case.


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