[Bugs] [Bug 1175551] Intermittent open() failure after creating a directory

[bugzilla at redhat.com]

https://bugzilla.redhat.com/show_bug.cgi?id=1175551

Niels de Vos <ndevos at redhat.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |Triaged
                 CC|                            |ndevos at redhat.com,
                   |                            |rgowdapp at redhat.com,
                   |                            |srangana at redhat.com
          Component|core                        |distribute
         Depends On|                            |1121920, 1094724



--- Comment #1 from Niels de Vos <ndevos at redhat.com> ---
> When a job starts it spawns numerous tasks in parallel. Each of these tasks,
> at startup, effectively does:
> 
> if [ ! -d /shared/scratchdir ]; then
>     mkdir -p /shared/scratchdir
> fi

I suspect that this concurrent creation of the directory is the issue. Other
bugs have been linked with this one where this problem is observed too.

It sounds possible that one process was able to create the directory on some of
the bricks, and an other process created the same directory on other bricks.
This leads to a directory that has different GFIDs on different bricks.

A workaround would be to create the /shared/scratchdir directory before the job
starts the additional tasks.

Adding the maintainers of DHT on CC who should know more about this issue and
the current state of the solution (http://review.gluster.org/4846).


Referenced Bugs:

https://bugzilla.redhat.com/show_bug.cgi?id=1094724
[Bug 1094724] mkdir/rmdir loop causes gfid-mismatch on a 6 brick distribute
volume
https://bugzilla.redhat.com/show_bug.cgi?id=1121920
[Bug 1121920] AFR : fuse,nfs mount hangs when directories with same names
are created and deleted continuously
-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are the assignee for the bug.