[cvsnt] Repository corruption

[Sebastian Wangnick]

Dear CVSNT folks,

please find at http://sebastian.wangnick.de/titeltext.zip one of the
seemingly corrupt repository files. The error message occurs *only* when
trying to access (or to compare with) version 1.1 of the file. This is also
why I did't detect the corruption earlier; in all those strange files I
detected, the more recent versions are perfectly accessible. Actually, I
detected the issue only today because (a very thin voice starting here) I
wanted to give Subversion a look and thus tried to convert the repository to
Subversion format to have a good size sample. This conversion process threw
the error messages at me, which I then confirmed by using TortoiseCVS to
access the respective versions.

I've also checked a backup done in 2004 and one of the corrupt files was
already corrupt back then. My corruption cases thus were not all introduced
in one go, but rather must have developed over time, with different versions
of CVSNT.

My statement on CVSNT reliability was purely reflecting my current alarm
state. After all, I did entrust our configuration management to CVSNT
originally, and so far was very happy with it. I didn't mean to step on
anybody's toes. However, I thought my utilisation of CVSNT to be rather
normal. I don't remember any power failure (but wouldn't want to exclude
one). Maybe I should check my hard disk for errors?

I also know from work that ClearCase has its issues, too. What concerns me a
bit is that such repository corruption goes unnoticed. If you would get a
warning about repository corruption right away, e.g., when checking in,
you'd have a chance to restore a proper repository file from backup. Or
maybe there is some sanity check tool?

Regards,
Sebastian Wangnick

-----Ursprüngliche Nachricht-----
Von: Flávio Etrusco
Gesendet: Montag, 28. August 2006 23:01
An: Sebastian Wangnick
Cc: cvsnt at cvsnt.org
Betreff: Re: [cvsnt] Repository corruption


On 8/28/06, Sebastian Wangnick <sebastian.home at wangnick.de> wrote:
> Dear CVSNT folks,
>
> for some small-scale Python and Webpage development environment for
> the company of my wife I'm using TortoiseCVS (1.8.11, CVSNT 2.0.58d)
> on a Windows XP system with a local CVSROOT.

I hope for "small-scale" you mean you alone, otherwise you definately
shouldn't be using :local:.


> Just now I recognised that for quite a number of files (10 out of 500) the
repository seems to be corrupted. When comparing the files against their
initial version I'm getting:
>         In C:\Temp\TCVca23.tmp: "C:\Programme\TortoiseCVS\cvs.exe" "-q"
"checkout" "-r" "1.1" "-d" "temp" "Foo/word.py"
>         CVSROOT=:local:C:\Dokumente und Einstellungen\All
Users\Dokumente\Cvsroot
>         cvs checkout: Dropping data: pos>vec->text.nlines
>         cvs [checkout aborted]: invalid change text in C:\Dokumente und
Einstellungen\All Users\Dokumente\Cvsroot/Foo/word.py,v
>         Fehler, CVS-Vorgang fehlgeschlagen
>
> I have been using older versions of TortoiseCVS in the past. However, one
of the corrupt files was checked in 2005.06.29, and at that time I had
installed 1.8.11/2.0.58d already.
>
> This issue occurs with Word documents, GIF files and in one case also with
a Python source file (which *could* have been modified from ASCII to UTF-8
format at a certain stage).

Do you mean it "occurs" (i.e. it's reproducible) or it "simply" occurred?
Why did it take so much time for you to detect the problem (2 months???)?

> Is this a known issue?
I don't think so. I don't remember anything like that reported on this list.


> Is it corrected in later versions?
I would sure hope so.
I'm certain there are "heaviers" cvsnt users than my employer out
there, and we've only had two episodes of repository file corruption
(only one file each)  in more than two years (starting with 2.0.51d or
.58d...), with about two hundred users, tens of projects, thousands
files of all kind and size (well, usually not beyond 10MB, except some
crazy guys that were storing 40MB-sized .pdb files in there :-/ ),
etc, and even those corruptions could be clearly associated with some
external cause since the server was running continually overloaded.


> Obviously this issue puts into question the reliability of CVSNT on
Windows.

Well, if you can rule out every and all other possible causes
(anti-virus, power failure, HD failure, OS crash or failure, etc)
then, yes, this would be a quite worrysome. I agree with the others in
that being it such an ancient version this isn't exactly a
showstopper. I disagree in that it's a stable release nonetheless and
any bug involving repository corruption is extremelly serious.
However, in my experience all of the other possibilities I've cited
are far more plausible than a cvsnt bug.

Cheers,
Flávio