Re: CVS, NFS Filesystem, and failover

[Paul Sander]

I have also personally experienced a situation with NetApp servers
where under heavy load a rename(a,b) call became an unlink(b) call,
which caused new committed data to be silently lost.  It's been a
number of years since this has happened, but keep an eye open for
it.

>--- Forwarded mail from address@hidden

>Andrew McGhee writes:
>> 
>> I've heard it mentioned on this newsgroup that mixing NFS with CVS is a bad
>> idea.
>>
>> Can anyone elaborate more one this?

>There have been lots of reports of repository corruption caused by
>interoperability problems between different NFS client and server
>implementations.  The most common problem is missing data: the file
>contains a block of NULs instead of the correct data.

>Since you're considering using a system that is specifically designed to
>be an NFS server, it's reasonable to conclude that it has been throughly
>tested with a wide variety of client implementations and thus is
>unlikely to have such problems, particularly if the NFS client machine
>is a common commercial platform.  I would be much more cautious with
>something like a Linux machine, since the server manufacturer has
>probably not tested it as extensively, if at all.  (Many of the reports
>of corruption have involved Linux, but it's impossible to tell whether
>that means that Linux is more trouble prone or just more popular.  If it
>is more trouble prone, it's impossible to tell whether that is because
>of bugs in the Linux NFS code or if it just exposes bugs in the other
>machine's server code.)

>So, it isn't necessarily a bad idea, but I would proceed cautiously. 
>You might want to grab the script that was posted here a while back to
>check the repository for damage and run it nightly for a while.

>--- End of forwarded message from address@hidden