hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jing Zhao (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDFS-8580) Erasure coding: Persist cellSize in BlockInfoStriped and StripedBlocksFeature
Date Mon, 15 Jun 2015 17:59:01 GMT

    [ https://issues.apache.org/jira/browse/HDFS-8580?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14586398#comment-14586398

Jing Zhao commented on HDFS-8580:

The 002 patch looks good to me. One comment is that while saving an fsimage, the {{fsn.getErasureCodingZoneForPath(n.getName())}}
call is very expensive since it has to regenerate a full path and resolve it. Thus how about
just retrieving the cell size from the INode's local striped block info?
+        final int cellSize = fsn.getErasureCodingZoneForPath(n.getName())
+            .getCellSize();
+        builder.setCellSize(cellSize);

> Erasure coding: Persist cellSize in BlockInfoStriped and StripedBlocksFeature
> -----------------------------------------------------------------------------
>                 Key: HDFS-8580
>                 URL: https://issues.apache.org/jira/browse/HDFS-8580
>             Project: Hadoop HDFS
>          Issue Type: Sub-task
>            Reporter: Walter Su
>            Assignee: Walter Su
>         Attachments: HDFS-8580-HDFS-7285.01.patch, HDFS-8580-HDFS-7285.02.patch, HDFS-8580.00.patch
> Zhe Zhang, Kai Zheng and I had a offline discussion. Here is what we thought:  Add a
cellSize field in BlockInfoStriped as a workaround, and deal with memory usage in follow-on.(HDFS-8059)
> discussion in HDFS-8494:
> from Walter Su:
> {quote}
> I think BlockInfoStriped needs to keep cellSize.
> {quote}
> from [~vinayrpet]:
> {quote}
> I too was thinking the same when the FSImageLoader problem has came up. This will increase
the memory usage by ~4bytes for each block though.
> {quote}
> from [~jingzhao]
> {quote}
> -Also, we should consider adding a chunk size field to StripedBlockProto and removing
the cell size field from HdfsFileStatus. In this way we can access the chunk size information
in the storage layer.-
> {quote}
> ==
> update:
> from [~jingzhao]
> {quote}
> For fsimage part, since HDFS-8585 just removes StripedBlockProto, I guess what we can
do here is to either 1) add the cellSize information into StripedBlocksFeature in fsimage.proto,
or 2) bring StripedBlockProto back and put block info and cell size there.
> {quote}

This message was sent by Atlassian JIRA

View raw message