hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Tsz Wo Nicholas Sze (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (HDFS-9710) Change DN to send block receipt IBRs in batches
Date Mon, 15 Feb 2016 16:10:18 GMT

     [ https://issues.apache.org/jira/browse/HDFS-9710?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Tsz Wo Nicholas Sze updated HDFS-9710:
--------------------------------------
    Attachment: h9710_20160216.patch

Many existing tests assume that datanodes send block receipt immediately.  For example, TestDecommission#testRecommission
creates a file with #replication == 5 and closes it.  Then it calls getFileBlockLocations
and expects the #locations must be  5.  The assumption is incorrect since the test will fail
if one of the 5 datanodes is slow.  I will fix these tests later.

Let me test the patch with the new feature disabled.

h9710_20160216.patch: sets DFS_BLOCKREPORT_INCREMENTAL_INTERVAL_MSEC_DEFAULT to 0.

> Change DN to send block receipt IBRs in batches
> -----------------------------------------------
>
>                 Key: HDFS-9710
>                 URL: https://issues.apache.org/jira/browse/HDFS-9710
>             Project: Hadoop HDFS
>          Issue Type: Improvement
>          Components: datanode
>            Reporter: Tsz Wo Nicholas Sze
>            Assignee: Tsz Wo Nicholas Sze
>         Attachments: h9710_20160201.patch, h9710_20160205.patch, h9710_20160216.patch
>
>
> When a DN has received a block, it immediately sends a block receipt IBR RPC to NN for
report the block.  Even if a DN has received multiple blocks about the same time, it still
sends multiple RPCs.  It does not scale well since NN has to process a huge number of RPCs
when many DNs receiving many blocks at the same time.



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

Mime
View raw message