[Bugs] [Bug 1751722] Gluster fuse mount crashed during truncate

[bugzilla at redhat.com]

https://bugzilla.redhat.com/show_bug.cgi?id=1751722

Krutika Dhananjay <kdhananj at redhat.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
              Flags|needinfo?(kdhananj at redhat.c |
                   |om)                         |



--- Comment #15 from Krutika Dhananjay <kdhananj at redhat.com> ---
(In reply to Nir Soffer from comment #14)
> Krutika, this makes sense, and explaining the randomness of the failure.
> 
> Can you confirm that using a unique file for every host will avoids this
> issue?
> 

Yes, using a unique file for every host would ensure that the cached values in
memory are correct
and are not modified by a different client.


> I'm thinking about this flow:
> 
> 1. Generate random filename e.g ".prob-cacb3ecf-9445-4638-aa00-af0ae99dc09e"
> 2. Open the prob file with O_WRONLY |O_DIRECT | O_CREAT | O_SYNC | O_EXCL |
> O_CLOEXEC
> 3. Delete the file
> 4. Use the open fd to detect block size
> 
> So the probe file can be accessed only from single host, and we never
> truncate it.
> 
> Do you see issue with gluster in particular when using shard translator?
> 
> Does gluster support O_TMPFILE? this will simplify steps 1-3 to:
> 
>     fd = open("/mount/path", O_TMPFILE | O_WRONLY | O_EXCL | O_DIRECT);

I think it should work, though I haven't tested it.

-Krutika

-- 
You are receiving this mail because:
You are on the CC list for the bug.