hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Mingliang Liu (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDFS-10338) DistCp masks potential CRC check failures
Date Wed, 27 Apr 2016 22:52:12 GMT

    [ https://issues.apache.org/jira/browse/HDFS-10338?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15261127#comment-15261127
] 

Mingliang Liu commented on HDFS-10338:
--------------------------------------

I'm favor of propagating the IOException thrown by {{getFileChecksum()}}. The retriable command
will take care of it, and after all retry-attempts the copy mapper will handle failures accordingly.
Moreover, I believe this is orthogonal to the {{ignoreFailures}} option.

Another point to mention is that, the {{checksumsAreEqual()}} has a conflict/confusing javadoc.
It claims:
{code}
   * @return If either checksum couldn't be retrieved, the function returns
   * false. If checksums are retrieved, the function returns true if they match,
   * and false otherwise.
   * @throws IOException if there's an exception while retrieving checksums.
{code}
While it has a {{throws IOException}} signature, it does not really throw any exception.

> DistCp masks potential CRC check failures
> -----------------------------------------
>
>                 Key: HDFS-10338
>                 URL: https://issues.apache.org/jira/browse/HDFS-10338
>             Project: Hadoop HDFS
>          Issue Type: Bug
>          Components: distcp
>    Affects Versions: 2.7.1
>            Reporter: Elliot West
>
> There appear to be edge cases whereby CRC checks may be circumvented when requests for
checksums from the source or target file system fail. In this event CRCs could differ between
the source and target and yet the DistCp copy would succeed, even when the 'skip CRC check'
option is not being used.
> The code in question is contained in the method [{{org.apache.hadoop.tools.util.DistCpUtils#checksumsAreEqual(...)}}|https://github.com/apache/hadoop/blob/release-2.7.1/hadoop-tools/hadoop-distcp/src/main/java/org/apache/hadoop/tools/util/DistCpUtils.java#L457]
> Specifically this code block suggests that if there is a failure when trying to read
the source or target checksum then the method will return {{true}} (i.e.  the checksums are
equal), implying that the check succeeded. In actual fact we just failed to obtain the checksum
and could not perform the check.
> {code}
> try {
>   sourceChecksum = sourceChecksum != null ? sourceChecksum : 
>     sourceFS.getFileChecksum(source);
>   targetChecksum = targetFS.getFileChecksum(target);
> } catch (IOException e) {
>   LOG.error("Unable to retrieve checksum for " + source + " or "
>     + target, e);
> }
> return (sourceChecksum == null || targetChecksum == null ||
>   sourceChecksum.equals(targetChecksum));
> {code}
> I believe that at the very least the caught {{IOException}} should be re-thrown. If this
is not deemed desirable then I believe an option ({{--strictCrc}}?) should be added to enforce
a strict check where we require that both the source and target CRCs are retrieved, are not
null, and are then compared for equality. If for any reason either of the CRCs retrievals
fail then an exception is thrown.
> Clearly some {{FileSystems}} do not support CRCs and invocations to {{FileSystem.getFileChecksum(...)}}
return {{null}} in these instances. I would suggest that these should fail a strict CRC check
to prevent users developing a false sense of security in their copy pipeline.



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

Mime
View raw message