Re: cvs [commit aborted]: out of memory;

[Muraleedhara Reddy]

Hi Bernd, Fmclean and Larry,

Thanks a lot for your help.

Finally I figured out that the culprit is file of size 1.9 GB on CVS server
that is unable to fti into kernel virtual memory. So we have successfully
started checking in file with a different name that started a fresh revision
of file on repository.

Thanking you once again,
Muraleedhara.

-----------------------------------------------------------------------------------------------------------------------------------
On Wed, Jul 16, 2008 at 10:23 PM, Bernd Jendrissek <
bernd.jendrissek@gmail.com> wrote:

> On Wed, Jul 16, 2008 at 4:16 PM, Muraleedhara Reddy <msirigir@gmail.com>
> wrote:
> > We are trying to check-in a binary file of size 228MB. We have its
> revision
> > 1.24 on original repository and 1.1 on test repository. Test repository
> has
> > no data except the file we are trying and it is not identical to original
> > one.
> >
> > I can see that the repository file size on original repository is 1.9 GB,
> > and on test repository, it is 228MB.
>
> I bet you're running a 32-bit kernel, or at least a 32-bit CVS, with a
> 2GB / 2GB userland/kernel virtual memory split.  Your repository file
> is so big that if you were to map it into the process's address space,
> there'd be no address space left for the process and its libraries
> themselves!
>
> Maybe you could run the CVS server on a 64-bit system, if preserving
> all of this file's history in one piece is very important.
>
> > But we have the /tmp space only around 1 GB. So do we need to remove any
> > intermediate/obsolete revisions so that we can make the
> > whole repository file fit into CVS virtual memory?
>
> If history is important, you could "cvs delete" the file, then rename
> the repository file (FOO,v) to something that indicates that it should
> not be used again, but also not deleted.  Perhaps FOO-oldhistory,v
> Then "cvs add" the last non-deleted revision of your file and start
> again, until you hit 1.24 or 1.23 again, and repeat.
>
> I recommend against checking such big files into a *source* control
> system.  It's unlikely to be source if it's that big.
>