hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Konstantin Shvachko (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDFS-1172) Blocks in newly completed files are considered under-replicated too quickly
Date Mon, 21 Mar 2011 02:15:06 GMT

    [ https://issues.apache.org/jira/browse/HDFS-1172?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13009004#comment-13009004
] 

Konstantin Shvachko commented on HDFS-1172:
-------------------------------------------

I think this should be controlled by "dfs.namenode.replication.interval". It is currently
set to 3 secs. If DNs do not keep up with reporting blocks it should be increased.
Putting blocks to pendingReplication feels like a trick, although it slows down replication
of the last block.
I think the right solution would be to add logic to processing of a failed pipeline. When
this happens the client asks for a new generation stamp. At this point NN can make a note
that this block will not have enough replicas. This will distinguish between blocks that have
not been reported yet, and those that will never be reported. This is much more work.
In practice I think tuning up the "replication.interval" parameter should be sufficient.

> Blocks in newly completed files are considered under-replicated too quickly
> ---------------------------------------------------------------------------
>
>                 Key: HDFS-1172
>                 URL: https://issues.apache.org/jira/browse/HDFS-1172
>             Project: Hadoop HDFS
>          Issue Type: Bug
>          Components: name-node
>    Affects Versions: 0.21.0
>            Reporter: Todd Lipcon
>         Attachments: HDFS-1172.patch
>
>
> I've seen this for a long time, and imagine it's a known issue, but couldn't find an
existing JIRA. It often happens that we see the NN schedule replication on the last block
of files very quickly after they're completed, before the other DNs in the pipeline have a
chance to report the new block. This results in a lot of extra replication work on the cluster,
as we replicate the block and then end up with multiple excess replicas which are very quickly
deleted.

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira

Mime
View raw message