hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jakob Homan (JIRA)" <j...@apache.org>
Subject [jira] Commented: (HDFS-1751) Intrinsic limits for HDFS files, directories
Date Thu, 17 Mar 2011 20:01:29 GMT

    [ https://issues.apache.org/jira/browse/HDFS-1751?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13008101#comment-13008101
] 

Jakob Homan commented on HDFS-1751:
-----------------------------------

A quota by any other name would still limit the number of objects that can be created in the
namespace just as sweetly.  

The following code from the patch is very telling:
{code}   private <T extends INode> T addChild(INode[] pathComponents, int pos,
       T child, long childDiskspace, boolean inheritPermission,
       boolean checkQuota) throws QuotaExceededException {
+	// The filesystem limits are not really quotas, so this check may appear
+	// odd.  It's because a rename operation deletes the src, tries to add
+	// to the dest, if that fails, re-adds the src from whence it came.
+	// The rename code disables the quota when it's restoring to the
+	// original location becase a quota violation would cause the the item
+	// to go "poof".  The fs limits must be disabled for the same reason.
{code}

Essentially it's saying we're not doing quota checking, except that we're throwing a FSLimitException
(which extends and is therefore a QuotaExceededException) and we have to do the exact same
workaround that the quota check has to do in this situation - but we're still not a quota
check.  

bq. after a couple of incidents where user applications exhausted some resource by behaving
in a way that was deeply wrong and (probably) unintended.
Could these incidents have been prevented by judicious use of the existing namespace quota?

My concern remains that this quota is implemented separately and in parallel from the main
quota checking mechanism, adding more state, more paths and more opportunity for bugs.  Could
we accomplish the same thing by defining a default NSQuota for new directories and allowing
this to be specified via configuration file or command line (as with the other quotas)?  

I'm -1 on this patch as it stands.

> Intrinsic limits for HDFS files, directories
> --------------------------------------------
>
>                 Key: HDFS-1751
>                 URL: https://issues.apache.org/jira/browse/HDFS-1751
>             Project: Hadoop HDFS
>          Issue Type: New Feature
>          Components: data-node
>    Affects Versions: 0.22.0
>            Reporter: Daryn Sharp
>            Assignee: Daryn Sharp
>             Fix For: 0.23.0
>
>         Attachments: HDFS-1751-2.patch, HDFS-1751-3.patch, HDFS-1751-4.patch, HDFS-1751.patch
>
>
> Enforce a configurable limit on:
>   the length of a path component
>   the number of names in a directory
> The intention is to prevent a too-long name or a too-full directory. This is not about
RPC buffers, the length of command lines, etc. There may be good reasons for those kinds of
limits, but that is not the intended scope of this feature. Consequently, a reasonable implementation
might be to extend the existing quota checker so that it faults the creation of a name that
violates the limits. This strategy of faulting new creation evades the problem of existing
names or directories that violate the limits.

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira

Mime
View raw message