[
https://issues.apache.org/jira/browse/HDFS-14739?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16935534#comment-16935534
]
Jinglun edited comment on HDFS-14739 at 9/23/19 10:23 AM:
----------------------------------------------------------
Hi [~xuzq_zander], I ran into the same problem recently. The root cause is in
RouterClientProtocol.getListing() it passes a broken child to
getMountPointStatus(). The patchv03 RouterClientProtocol.getListing()#829~831
makes sense to me, you pass the full path of child to it.
If I understand right, v03 RouterClientProtocol.getListing()#748~764 is used
when there is no default namespace nor a / mount. For example:
/dir-0/dir-1 ns0->/dir
/dir-0/dir-2 ns1->/dir
If we list /dir-0, we will expect a result \{dir-1, dir-2}. So we must catch
the IOException, or the exception will be thrown to the client. I think we
should add a test case for this.
*Correct me if I'm wrong: if we list /dir-non-exist, we will expect an
Exception rather than {}.* Because {} means /dir-non-exist is an empty dir.
This case seems not handled in v03. Here we need a test case.
About the unit test, I've make
TestRouterMountTable.testMountTablePermissionsNoDest() stronger in HDFS-14867.
Hope it can help.
was (Author: lijinglun):
Hi [~xuzq_zander], I ran into the same problem recently. The root cause is in
RouterClientProtocol.getListing() it passes a broken child to
getMountPointStatus(). The patchv03 RouterClientProtocol.getListing()#829~831
makes sense to me, you pass the full path of child to it.
If I understand right, v03 RouterClientProtocol.getListing()#748~764 is used
when there is no default namespace nor a / mount. Shouldn't we simply let it
fail when there is no default and / mount ? I mean why we need to catch the
IOException, could we just let it throw ?
About the unit test, I've make
TestRouterMountTable.testMountTablePermissionsNoDest() stronger in HDFS-14867.
Hope it can help, just for reference.
> RBF: LS command for mount point shows wrong owner and permission information.
> -----------------------------------------------------------------------------
>
> Key: HDFS-14739
> URL: https://issues.apache.org/jira/browse/HDFS-14739
> Project: Hadoop HDFS
> Issue Type: Bug
> Reporter: xuzq
> Assignee: xuzq
> Priority: Major
> Attachments: HDFS-14739-trunk-001.patch, HDFS-14739-trunk-002.patch,
> HDFS-14739-trunk-003.patch, image-2019-08-16-17-15-50-614.png,
> image-2019-08-16-17-16-00-863.png, image-2019-08-16-17-16-34-325.png
>
>
> ||source||target namespace||destination||owner||group||permission||
> |/mnt|ns0|/mnt|mnt|mnt_group|755|
> |/mnt/test1|ns1|/mnt/test1|mnt_test1|mnt_test1_group|755|
> |/test1|ns1|/test1|test1|test1_group|755|
> When do getListing("/mnt"), the owner of */mnt/test1* should be *mnt_test1*
> instead of *test1* in result.
>
> And if the mount table as blew, we should support getListing("/mnt") instead
> of throw IOException when dfs.federation.router.default.nameservice.enable is
> false.
> ||source||target namespace||destination||owner||group||permission||
> |/mnt/test1|ns0|/mnt/test1|test1|test1|755|
> |/mnt/test2|ns1|/mnt/test2|test2|test2|755|
>
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
