openstack team mailing list archive

[David Dobbins <david.dobbins@xxxxxxxxxxxxx>]

Clay-

You are correct – the hadoop-swift filesystem is just an implement of the Hadoop FileSystem class that uses swift for storage.  As such, it's just translating filesystem calls into HTTP requests into swift and is not accessing the filesystem of the object servers directly.

Shashank-

I'm curious about what version of that Hadoop-Swift-integration project you're trying to run.  You shouldn't have been able to create containers with it in any of the more recent versions.  You might want to try the "converged" branch of https://github.com/hortonworks/Hadoop-and-Swift-integration.  This is the branch that's getting submitted back to Apache for inclusion in Hadoop.

Hope that helps,
-David

From: Vaidy Gopalakrishnan <gvaidy@xxxxxxxxx<mailto:gvaidy@xxxxxxxxx>>
Date: Friday, April 26, 2013 4:06 PM
To: Clay Gerrard <clay.gerrard@xxxxxxxxx<mailto:clay.gerrard@xxxxxxxxx>>, David Dobbins <david.dobbins@xxxxxxxxxxxxx<mailto:david.dobbins@xxxxxxxxxxxxx>>
Cc: Shashank Sahni <shredder12@xxxxxxxxx<mailto:shredder12@xxxxxxxxx>>, "openstack@xxxxxxxxxxxxxxxxxxx<mailto:openstack@xxxxxxxxxxxxxxxxxxx>" <openstack@xxxxxxxxxxxxxxxxxxx<mailto:openstack@xxxxxxxxxxxxxxxxxxx>>
Subject: Re: [Openstack] Swift container's rwx permissions


[Hi Clay]

Including David who can answer this much better.

Vaidy


On Fri, Apr 26, 2013 at 12:30 PM, Clay Gerrard <clay.gerrard@xxxxxxxxx<mailto:clay.gerrard@xxxxxxxxx>> wrote:
Wow, so I glanced quickly at the github project to try and get some context, but I think I actually start getting *more* confused when I see "swift" in the same class name as "file-system" ;)

I'd like you (or maybe vaidy, hi vaidy!) to correct me if I'm wrong, but this hadoop integration will *not* access the filesystem of the object servers directly?  Everything will happen on a pool of processing boxes that will talk to swift via HTTP - same as any other client?

In that case, the error message is just a leaky abstraction showing through.  HDFS probably has permission errors that it tries to helpfully map back to file system constructs which just don't apply when you're trying to "simulate" a file system on object storage.  You'll have to get down the the HTTP to understand what's causing the error.  Presumably a 401 from Swift, so access to swift logs would be helpful.

OTOH, if we're * actually* talking about filesystem permissions; then I'm totally confused.  But ACL's definitely won't help.  They're just a row sitting in a sqlite database - probably on a totally different server from where the one replica of this object is sitting on the filesystem. Nothing you can set in the api will change the filesystem permissions of the directory structure or files on the object servers.

Maybe do you have some more overview info on the general approach?  I don't really have any Hadoop experience, so maybe it'd be better if there's a hadoop expert out there that also has some experience with swift and can help get you on the right track...

-Clay



In


On Fri, Apr 26, 2013 at 1:11 AM, Shashank Sahni <shredder12@xxxxxxxxx<mailto:shredder12@xxxxxxxxx>> wrote:
Hi everyone,

I've been experimenting with using Swift(1.8) as Hadoop's DFS backend.

https://github.com/DmitryMezhensky/Hadoop-and-Swift-integration

After a few glitches, I'm able to create/access/delete objects/containers using hadoop's cli fs tool. But whenever I'm trying to run the job it fails with the following error.

ERROR security.UserGroupInformation: PriviledgedActionException as:dharmesh cause:java.io.IOException: The ownership/permissions on the staging directory swift://hadooptest.rackspace/tmp/app/mapred/staging/dharmesh/.staging is not as expected. It is owned by and permissions are rwxrwxrwx. The directory must be owned by the submitter dharmesh or by dharmesh and permissions must be rwx------

Note that, the local system username is "dharmesh" and the openstack account and associated tenant is "dharmesh" too.

I tried setting the permissions by creating "tmp" container using "swift post -r 'dharmesh:dharmesh'", but unfortunately ended up with the same result. Is there an other way to set rwx ACLs in swift?

--
Shashank Sahni

_______________________________________________
Mailing list: https://launchpad.net/~openstack
Post to     : openstack@xxxxxxxxxxxxxxxxxxx<mailto:openstack@xxxxxxxxxxxxxxxxxxx>
Unsubscribe : https://launchpad.net/~openstack
More help   : https://help.launchpad.net/ListHelp