hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Zhe Zhang (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDFS-8461) Erasure coding: fix priority level of UnderReplicatedBlocks for striped block
Date Sat, 19 Dec 2015 00:45:46 GMT

    [ https://issues.apache.org/jira/browse/HDFS-8461?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15065081#comment-15065081
] 

Zhe Zhang commented on HDFS-8461:
---------------------------------

Thanks Rui for finding the issue. I think we should update the comment. The {{* 3}} logic
is used to be consistent with the contiguous case. Roughly speaking, {{QUEUE_VERY_UNDER_REPLICATED}}
means the current redundancy is 1/3 of desired redundancy.

> Erasure coding: fix priority level of UnderReplicatedBlocks for striped block
> -----------------------------------------------------------------------------
>
>                 Key: HDFS-8461
>                 URL: https://issues.apache.org/jira/browse/HDFS-8461
>             Project: Hadoop HDFS
>          Issue Type: Sub-task
>            Reporter: Walter Su
>            Assignee: Walter Su
>             Fix For: HDFS-7285
>
>         Attachments: HDFS-8461-HDFS-7285.001.patch, HDFS-8461-HDFS-7285.002.patch
>
>
> Issues 1: correctly mark corrupted blocks.
> Issues 2: distinguish highest risk priority and normal risk priority.
> {code:title=UnderReplicatedBlocks.java}
>   private int getPriority(int curReplicas,
>   ...
>     } else if (curReplicas == 1) {
>       //only on replica -risk of loss
>       // highest priority
>       return QUEUE_HIGHEST_PRIORITY;
>   ...
> {code}
> For stripe blocks, we should return QUEUE_HIGHEST_PRIORITY when curReplicas == 6( Suppose
6+3 schema).
> That's important. Because
> {code:title=BlockManager.java}
> DatanodeDescriptor[] chooseSourceDatanodes(BlockInfo block,
>   ...
>      if(priority != UnderReplicatedBlocks.QUEUE_HIGHEST_PRIORITY 
>           && !node.isDecommissionInProgress() 
>           && node.getNumberOfBlocksToBeReplicated() >= maxReplicationStreams)
>       {
>         continue; // already reached replication limit
>       }
>   ...
> {code}
> It may return not enough source DNs ( maybe 5), and failed to recover.
> A busy node should not be skiped if a block has highest risk/priority. The issue is the
striped block doesn't have priority.



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

Mime
View raw message