hadoop-hdfs-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "J.Andreina (JIRA)" <j...@apache.org>
Subject [jira] [Created] (HDFS-8731) Erasure coding: For a small file missing and under replicated ec-block calculation is incorrect
Date Wed, 08 Jul 2015 04:47:04 GMT
J.Andreina created HDFS-8731:
--------------------------------

             Summary: Erasure coding: For a small file missing and under replicated ec-block
calculation is incorrect 
                 Key: HDFS-8731
                 URL: https://issues.apache.org/jira/browse/HDFS-8731
             Project: Hadoop HDFS
          Issue Type: Sub-task
            Reporter: J.Andreina
            Assignee: J.Andreina


Write a small file (whose replica detail would be : one data replica and 3 parity replicas
.)

Now :
	totalReplicasPerBlock = 4 
	targetFileReplication = 9 (calculated based on default RS-6-3 schema)

Hence the small file ec-blocks is calculated as always missing and underreplicated. which
is wrong... 

{code}
targetFileReplication = (short) (ecSchema.getNumDataUnits() + ecSchema.getNumParityUnits());
{code}
{code}
      // count missing replicas / under replicated blocks
      if (totalReplicasPerBlock < targetFileReplication && totalReplicasPerBlock
> 0) {
        res.missingReplicas += (targetFileReplication - totalReplicasPerBlock);
	res.numUnderReplicatedBlocks += 1;
{code}



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

Mime
View raw message