hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Yanbo Liang (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDFS-3219) Disambiguate "visible length" in the code and docs
Date Tue, 23 Oct 2012 07:57:13 GMT

    [ https://issues.apache.org/jira/browse/HDFS-3219?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13482182#comment-13482182

Yanbo Liang commented on HDFS-3219:

There is a scenario:
1) Dfs client writes to a file on HDFS, then the client exit with exception.
2) All the replicas of the last block on the DataNodes are in RBW state and may be with different
We assume the last block has three replicas that the visible length is R1 < R2 < R3.
3) When we open the file for read, DFSInputStream construction with invoking openInfo().
At this moment we connect to the DataNode with replica R2 and set the parameter lastBlockBeingWrittenLength
with R2.
4) When we read from the stream with fail over to other DataNode such as the DataNode with
replica R1,
we can not get the last data of R2-R1.

I wonder whether this scenario is make sense?
> Disambiguate "visible length" in the code and docs
> --------------------------------------------------
>                 Key: HDFS-3219
>                 URL: https://issues.apache.org/jira/browse/HDFS-3219
>             Project: Hadoop HDFS
>          Issue Type: Improvement
>            Reporter: Eli Collins
>            Priority: Minor
> HDFS-2288 there are two definition of visible length, or rather we're using the same
name for two things:
> 1. The HDFS-265 design doc which defines it as property of the replica:
> {quote}
> visible length is the "number of bytes that have been acknowledged by the downstream
DataNodes". It is replica (not block) specific, meaning it can be different for different
replicas at a given time. In the document it is called BA (bytes acknowledged), compared to
BR (bytes received).
> {quote}
> 2. The definition in HDFS-814 and DFSClient#getVisibleLength which defines it as a property
of a file:
> {quote}
> The visible length is the length that *all* datanodes in the pipeline contain at least
such amount of data. Therefore, these data are visible to the readers.
> {quote}
> According to this definition the visible length of a file is the floor of all visible
lengths of all the replicas of the last block. It's a static property set on open, eg is not
updated when a writer calls hflush. Also DFSInputStream#readBlockLength returns the 1st visible
length of a replica it finds, so it seems possible (though unlikely) in a failure scenario
it could return a length that was longer than what all replicas had.
> This has caused confusion in a number of other jiras. We should update the design doc,
java doc, perhaps rename DFSClient#getVisibleLength etc to disambiguate this.

This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

View raw message