On Tue, 2006-10-10 at 11:46 +0200, Jan Engelhardt wrote:

> >
> >unfortunately, it was never that stable and it performance sucked
> >(basically only useful for append only things like log files).
> 
> There is a problem (or advantage) with sparse copyup. Consider the 
> following action table (dirs=/B=rw:/A=ro):
> 
> Current unionfs CVS
> 
> Action                              | BRANCH A | BRANCH B | UNION VIEW
> ------------------------------------+----------+----------+------------
> mount union                         |   0000   |     -    | 0000
> make changes to some range          |   0000   |   0001   | 0001
> umount union                        |   0000   |   0001   | 0001
> make changes to the underlying file |   1000   |   0001   | 0001
> 
> 0 = some bytes filled with 0
> 1 = some bytes filled with 1
> - = sparse bytes
> 
> With sparse copyup:
> 
> Action                              | BRANCH A | BRANCH B | UNION VIEW
> ------------------------------------+----------+----------+------------
> mount union                         |   0000   |     -    | 0000
> make changes to some range          |   0000   |   ---1   | 0001
> umount union                        |   0000   |   ---1   | 0001
> make changes to the underlying file |   1000   |   ---1   | 1001
> 
> 0001 vs 1001 - whoops! Though modifying underlying files is unsupported while
> mounted, you see that this issue persists _across_ a not-mounted union.

yes, in my model, the backing store only was ever accessed via the
union.  It wasn't perfect, and I ended up deciding there were better
ways to achieve what I wanted.

_______________________________________________
unionfs mailing list
[email protected]
http://www.fsl.cs.sunysb.edu/mailman/listinfo/unionfs

Reply via email to