[Gluster-devel] Bug regarding quota.t

[Krishnan Parthasarathi]

I am not sure whether its incorrect for quota to expect healing of extended attributes or its a dht bug, http://review.gluster.org/#/c/6392/ is the commit that adds the test case that you observe failing.

HTH,
KP

----- Original Message -----
> On Sun, Aug 24, 2014 at 10:32 PM, Harshavardhana
> <harsha at harshavardhana.net> wrote:
> > While trying to fix some porting issues i came across a rather a bug
> > in quota.t testcase implementation or rather an inherent bug of
> > distribute self-heal.
> >
> > Quota.t in its 63/64th tests does an add-brick and rebalance to expand
> > the testing volume. But this is done with a premise of self-heal of
> > directories after the rebalance would help in propagating
> > "trusted.quota.limit-set" extended attributes.
> >
> > Now after the rebalance is completed 66/67th test use the following
> > CLi style for 'getfattr'.
> >
> > ~~~
> > TEST getfattr -d -m "trusted.glusterfs.quota.limit-set" -e hex \
> >               --absolute-names $B0/brick{3,4}/$TESTDIR/dir{1..10};
> >
> > TEST getfattr -d -m "trusted.glusterfs.quota.limit-set" -e hex \
> >               --absolute-names $B0/brick{3,4};
> > ~~~
> > One needs to be aware that 'getfattr' with valid or invalid "xattr"
> > regexes while using "-m" option does not exit with a non zero exit
> > code.
> >
> > Observe the following example
> >
> > # touch test
> > # getfattr -d -m "trusted.foo" test
> > # echo $?
> > 0
> >
> > The problem here is that on "$B0/brick{3,4}" the attributes do not
> > exist, while we are expecting it to be healed back after the
> > rebalance? is that really necessary? since the 66th test validates
> > these attributes on "dir{1..10}" ?
> >
> > The issue i reproduced was after editing the above code to be more
> > predantic by using "-n"
> > ~~~
> > TEST getfattr -n trusted.glusterfs.quota.limit-set -e hex \
> >                --absolute-names $B0/brick{3,4}/$TESTDIR/dir{1..10};
> > TEST getfattr -n trusted.glusterfs.quota.limit-set -e hex \
> >                --absolute-names $B0/brick{3,4};
> > ~~~
> >
> > Now this leads to 67th test to be a failure always, due to the
> > aforementioned fact that the xattrs do not exist or neither do they
> > get healed to its parent directory. Is this a Distribute bug? or is
> > this a wrong expectation from quota to get the quota xattrs to be
> > healed on parent brick directory?
> 
> Does anyone care to clarify this?
> 
> --
> Religious confuse piety with mere ritual, the virtuous confuse
> regulation with outcomes
> _______________________________________________
> Gluster-devel mailing list
> Gluster-devel at gluster.org
> http://supercolony.gluster.org/mailman/listinfo/gluster-devel
>