hadoop-common-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Raghu Angadi (JIRA)" <j...@apache.org>
Subject [jira] Reopened: (HADOOP-1070) Number of racks and datanode double temporarily when upgrading from 0.10.1 to 0.11.2
Date Tue, 13 Mar 2007 01:26:09 GMT

     [ https://issues.apache.org/jira/browse/HADOOP-1070?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Raghu Angadi reopened HADOOP-1070:
----------------------------------



> Because namenode was restarted before check point in the prev run, both hostName and
ip are added to n/w topology. 
> When a datanode is removed in the image or edits log, we dont delete its entry from n/w
topo, I am not sure why. Now only ip 
> entries get updated with the new topo info. This explains extra nodes. Now new image
is written to disk (equivalent to a
> checkpoint in the first start up). 

Looks like this could lead to a situation where a datanode is marked dead but Namenode tries
to schedule blocks to it since it exists in NetworkTopology clusterMap.

One suggested fix ( by Konstantin ) is not to add any nodes to clusterMap while loading fsimage.
So the clusterMap is updated only when nodes register and when they are marked dead. Will
submit a patch for this. 

> Number of racks and datanode double temporarily when upgrading from 0.10.1 to 0.11.2
> ------------------------------------------------------------------------------------
>
>                 Key: HADOOP-1070
>                 URL: https://issues.apache.org/jira/browse/HADOOP-1070
>             Project: Hadoop
>          Issue Type: Bug
>          Components: dfs
>    Affects Versions: 0.11.2
>            Reporter: Nigel Daley
>         Assigned To: Raghu Angadi
>             Fix For: 0.13.0
>
>
> When upgrading from Hadoop 0.10.1 to 0.11.2, I see the number of racks and datanode double
after the 2nd startup of the Namenode.  After the 3rd startup, they correct themselves:
> Namenode Log After 1st Startup:
> 2007-03-06 18:27:27,045 INFO org.apache.hadoop.dfs.StateChange: STATE* Network topology
has 1 racks and 4 datanodes
> Namenode Log After 2nd Startup:
> 2007-03-06 18:27:43,201 INFO org.apache.hadoop.dfs.StateChange: STATE* Network topology
has 2 racks and 8 datanodes
> Namenode Log After 3rd Startup:
> 2007-03-06 18:28:09,730 INFO org.apache.hadoop.dfs.StateChange: STATE* Network topology
has 1 racks and 4 datanodes

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Mime
View raw message