> On 12/21/2011 9:15 AM, Arthur Prokosch wrote:
> > Jonathan, or others who experienced this bug -- have you seen corrupt
> > excel files with modification times from before when a volume was
> > served by 1.6.0?
>

Hmm, I hadn't actually thought of that scenario, or looked for evidence to
support it. The first few cases of corruption were for files modified after
a volume was moved to 1.6 servers.  Since then, all volumes have been moved
back to 1.4 servers and I have not looked as closely at the evidence (just
performing restores based on the modification timestamp).


> In all of the cases I have seen the corruption was triggered by opening
> and closing an excel spreadsheet which results in 8 bytes being written
> to the spreadsheet header.  The header becomes corrupted with either the
> first 1K or 4K of the file being replaced by NULs instead of the actual
> header file.  The last modification timestamp is preserved by Excel.


Ah, that explains it! Thanks for discovering this behavior, Jeffery. Since
I did not know that Excel had the behavior, for each file I tested
restoring many different versions of before finding that I could reliably
restore the one from the latest modification timestamp.

-- 
Jonathan.Nilsson at uci dot edu
Social Sciences Computing Services
SSPB 1265 | 949.824.1536

Reply via email to