[Gluster-devel] 3.5.1-beta2 Problems with suid and sgid bits on directories

[Anders Blomdell]

On 06/19/2014 03:39 PM, Anders Blomdell wrote:
> On 2014-06-19 13:48, Susant Palai wrote:
>> Adding Susant
> Unfortunately things don't go so well here, with --brick-log-level=DEBUG, I get very
> weird results (probably because the first brick is slower to respond while it's
> printing debug info), I suspect I trigger some timing related bug.
>
> I attach my testscript and a log of 20 runs (with 02777 flags).
>
> The real worrisome thing here is:
>
>    backing: 0 0:0 /data/disk2/gluster/test/dir1
>
> which means that the backing store has an unreadable dir, which gets propagated to clients...
I have an embryo of an theory of what happens:

1. directories are created on the first brick.
2. fuse starts to read directories from the first brick.
3. getdents64 or fstatat64 to first brick takes too long, and
    is redirected to second brick.
4. self-heal is initiated on second brick.

On monday, I will see if I can come up with some clever firewall tricks 
to trigger this behaviour in a reliable way.

/Anders

-- 
Anders Blomdell                  Email: anders.blomdell at control.lth.se
Department of Automatic Control
Lund University                  Phone:    +46 46 222 4625
P.O. Box 118
SE-221 00 Lund, Sweden