[ https://issues.apache.org/jira/browse/HDFS-11268?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15823810#comment-15823810 ]
SammiChen commented on HDFS-11268: ---------------------------------- Hi [~jojochuang]], you're right, EC policy ID is stored using replication factor field. After I did more investigation, It turns out that the problem is the EC policy ID loading process from the FsImage file. I updated the JIRA description accordingly. > Correctly reconstruct erasure coding file from FSImage > ------------------------------------------------------ > > Key: HDFS-11268 > URL: https://issues.apache.org/jira/browse/HDFS-11268 > Project: Hadoop HDFS > Issue Type: Bug > Components: erasure-coding > Affects Versions: 3.0.0-alpha1 > Reporter: SammiChen > Assignee: SammiChen > Priority: Critical > Labels: hdfs-ec-3.0-must-do > Attachments: HDFS-11268-001.patch > > > Currently, FSImageFormatProtoBuf has the information about whether the file > is striped or not and saved file's erasure coding policy ID in replication > field. But later, when FSImage is loaded to create the name space, the > default system erasure coding policy is used to reconstruct file's block > structure. In case if the erasure coding policy of file is not the default > erasure coding policy, the content of the file cannot be accessed correctly. -- This message was sent by Atlassian JIRA (v6.3.4#6332) --------------------------------------------------------------------- To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org