Re: [Gzz] Paperbot status

[Janne Kujala]

On Tue, May 13, 2003 at 12:10:21PM +0300, Tuukka Hastrup wrote:
> The structure at the moment is following: Makondo runs bgfilegen.py from 
> libvob, generating texture files for request files and moving the textures 
> via ssh to Himalia. The request files are placed by a CGI script at 
> http://himalia.it.jyu.fi/paperbot/paper. I hope someone could review the 
> system for errors.

Overall, the cgi script (himalia:~paperbot/public_html/paper.cgi)
seems to be secure. However, a DoS attack could be severe
because the cgi (shell) script uses several subprocesses 
for parsing and processing the arguments (tr, perl, bc, sed, cut).
This also makes it hard to review all details.

The process for generating and transferring the papers 
is unnecessarily slow. With Tuukka, we identified several 
sources of delay:

  - an extra "sleep 1", which seems unnecessary
  - 2-3 seconds for the ssh connection for adding the request to makondo
  - 0-1 seconds for polling the request directory in makondo
  - 2-3 seconds for the ssh connection for copying the result to himalia
  - 0-1 seconds for polling the result directory in himalia

The paperbot could be made faster by 

  - synchronizing the different processes with signals 
  - either using non-authenticated connections or only using 
    a single ssh connection for the communications
  - getting rid of most of the extra processes 
    (e.g., by only using python) 
  - possibly making the actual paper generation faster
      - does the jpeg compression use imagemagick?
      - is there unnecessary disk io?

I think the delay should be reduced to few seconds.
Otherwise, I think the system is acceptable.

        Janne