accumulo-notifications mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Josh Elser (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (ACCUMULO-3976) Lift "HBASE-11240 Print hdfs pipeline when hlog's sync is slow"
Date Fri, 28 Aug 2015 01:43:45 GMT

    [ https://issues.apache.org/jira/browse/ACCUMULO-3976?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14717892#comment-14717892
] 

Josh Elser commented on ACCUMULO-3976:
--------------------------------------

I think what I was thinking of was using HDFS-826 and detecting when the number of datanodes
in the pipeline are less than a configured value (e.g. a datanode in the pipeline is dead).
HBase has some code, also in FSHLog.java, which uses the {{DFSOutputStream.getNumCurrentReplicas}}
to detect when the number of replicas are less than what we expect, and teardown+recreate
the outputstream.

> Lift "HBASE-11240 Print hdfs pipeline when hlog's sync is slow"
> ---------------------------------------------------------------
>
>                 Key: ACCUMULO-3976
>                 URL: https://issues.apache.org/jira/browse/ACCUMULO-3976
>             Project: Accumulo
>          Issue Type: Improvement
>          Components: tserver
>            Reporter: Josh Elser
>            Assignee: Eric Newton
>             Fix For: 1.6.4, 1.7.1, 1.8.0
>
>
> HBASE-11240 is a really nice little change that has been extremely helpful in my debugging
sessions of HBase in the past. It provides a very clear message that can point at HDFS as
the reason for "slow ingest".
> I would suspect that we could easily lift this same concept into our code (around the
sync thread). I think this would be a very nice improvement to make.



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

Mime
View raw message