[Savannah-hackers-public] Re: GNU Rush 1.6

[Sergey Poznyakoff]

Sylvain Beucler <address@hidden> ha escrit:

> Savannah gets less than 1 SSH connection per minute for
> CVS/SVN/Git/Hg, so it doesn't sound like something that needs
> optimization.

I agree with that. Gicen current connection rates the performance gain
is not noticeable, although stress tests do show it.

> Meanwhile writing software in C is usually more tedious.

It depends. In my case it is the other way around :^)

> This reminds me: I saw in the last SourceForge announcement that
> they're providing unrestricted shell access limited to 4h with limited
> view on the system, which they claim to be based on "virtualization".
> Any idea on what this is based on?

Interesting. I can't say right now how that can be achieved (putting
aside chrooted implementation). Let me think.

> > 4. It allows to execute programs in chdir-ed environment. This feature,
> > among others, is used when executing `cvs server'. This allows to limit
> > access to a single root directory. It is also used when handling
> > uploads of files for distribution from download.gnu.org.ua (see
> > below);
> 
> (chdir'd or chroot'd?)

Oops, chrooted, of course.

> I tend to distrust chroot'd environment: they are likely to introduce
> breakages.

Their use simply requires a very careful attention to details (such as
needed libraries, devices, etc.).

> At Savannah the overzealous usage of chroot and sanitizing
> in early 2004 made CVS register 'uidXXXX' instead of the actual user
> name, ruining the history.

Ah, yes, I remember that.

> > Login      Rule         Start            Stop      Time   Command
> > sergiusz   git          Wed Feb 11 20:28 Wed 20:28 00:00  
> > /usr/bin/git-upload-
> > andy_shev  svn          Wed Feb 11 19:05 Wed 19:05 00:00  /usr/bin/svnserve 
> > -r
> > gray       sftp-upload  Wed Feb 11 18:00 Wed 18:01 00:01  /bin/sftp-server
> 
> It sounds interesting to produce stats out of these logs :)

It is possible. Any idea what to include in these statistics?

> post-socket is a nice concept.
> Do you know how interrupted connections are handled?

They are handled by the wydawca utility. When started, it collects
information about triplets (tarball.directory.asc, tarball,
tarball.sig). Any complete triplets (i.e. ones that contain all three
components and ones that contain a lone self-sufficient directory file
of v.1.1) are processed immediately. Incomplete triplets are held on
quarantine for a certain time, in the hope that submitter will supply
missing parts some time in the future. If he does not, such dangling
incomplete triplets are removed after the timeout expires.

Regards,
Sergey