[Gluster-devel] release-3.7 regression tests stability

[Sachin Pandit]

comments inline.

----- Original Message -----
> From: "Sachin Pandit" <spandit at redhat.com>
> To: "Atin Mukherjee" <amukherj at redhat.com>
> Cc: "Gluster Devel" <gluster-devel at gluster.org>
> Sent: Monday, June 8, 2015 11:40:05 AM
> Subject: Re: [Gluster-devel] release-3.7 regression tests stability
> 
> 
> 
> ----- Original Message -----
> > From: "Atin Mukherjee" <amukherj at redhat.com>
> > To: "Vijay Bellur" <vbellur at redhat.com>, "Gluster Devel"
> > <gluster-devel at gluster.org>
> > Cc: "Vijaikumar Mallikarjuna" <vmallika at redhat.com>, "Sachin Pandit"
> > <spandit at redhat.com>
> > Sent: Sunday, June 7, 2015 10:25:24 AM
> > Subject: Re: [Gluster-devel] release-3.7 regression tests stability
> > 
> > 
> > 
> > On 06/05/2015 11:38 PM, Vijay Bellur wrote:
> > > On 06/05/2015 09:39 AM, Atin Mukherjee wrote:
> > >> I still see lot of patches in release-3.7 are failing regression, be it
> > >> Linux or Netbsd. Does this mean all the spurious fixes which we did in
> > >> mainline are yet to be in 3.7? If so, what are we waiting for?
> > > 
> > > Most of the fixes are in release-3.7. Apart from glupy.t, are we aware
> > > of others that fail in release-3.7 consistently but pass the mainline
> > > regression runs?
> > I see inode-quota.t failed in these following two runs:
> > 
> > http://build.gluster.org/job/rackspace-regression-2GB-triggered/10189/consoleFull
> > http://build.gluster.org/job/rackspace-regression-2GB-triggered/10210/consoleFull
> > 
> > Ccing Vijai & Sachin
> 

I did some analysis on this and found out below mentioned
log during the case of failure.

[2015-06-08 03:45:21.317352] I [MSGID: 115071] [server-rpc-fops.c:1591:server_crea
te_cbk] 0-patchy-server: 203: CREATE /test_dir/test1.txt (72fa9c7d-d4cd-474e-baae-
28dde9b15990/test1.txt) ==> (Disk quota exceeded) [Disk quota exceeded]

The test case which fails is "TEST_IN_LOOP touch $M0/test_dir/test$i.txt",
while creating a file "test1" (i.e i=1). This should not fail as "test1" is
the first file being created in the directory where quota limit is set.
The only possible explanation is that before creating the file "test1",
we delete all the files present in that directory, and that information
has not been updated by marker. Because of which the new file creation
fails thinking the quota limit is still exceeded.

The test case basically validates both the types of quota (size and inode).
The reason for this spurious failure might be because of the delay in
updating the information my marker. I will verify the same by inducing
some delay before the execution of above mentioned testcase.

Thanks,
Sachin.


> Thank you Atin for reporting this failure.
> We will have a look at the failure and update
> this thread as and when we make any progress.
> 
> Thanks,
> Sachin Pandit.
> 
> > 
> > > 
> > > I suspect that we also might be hitting some new failures due to the
> > > previously failing tests now passing.
> > > 
> > > -Vijay
> > > 
> > > 
> > 
> > --
> > ~Atin
> > 
> _______________________________________________
> Gluster-devel mailing list
> Gluster-devel at gluster.org
> http://www.gluster.org/mailman/listinfo/gluster-devel
>