www-infrastructure-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Santiago Gala <santiago.g...@gmail.com>
Subject Re: Best Practices so far?
Date Fri, 02 May 2008 10:11:35 GMT
El vie, 02-05-2008 a las 10:33 +0100, Justin Mason escribió:
> Jukka Zitting writes:
> > Hi,
> > 
> > On Fri, May 2, 2008 at 11:22 AM, Davanum Srinivas <davanum@gmail.com> wrote:
> > > So, what's the learning so far? Example how does one set up a git-svn for a
> > > specific Apache project?
> > 
> > I've cloned a few smaller projects like this:
> > 
> >  git svn clone -s http://svn.apache.org/repos/asf/incubator/<project>
> > 
> > So far I've avoided doing that with TLPs because of the load on the
> > svn server, but now that svn.eu.apache.org is in place I'm running the
> > following:
> > 
> >  git svn clone -s http://svn.eu.apache.org/repos/asf/jackrabbit
> > 
> > Note that these clones take quite a while (hours) to create, but after
> > that it's quite easy and fast to keep them up to date with latest
> > changes.
> This is useful info -- thanks!
> I noticed on the infra list there was mention of a parameter which
> restricts how many commits to span in the SVN history; it might be useful
> to add that to this "best practices" thread...

--log-window-size=100 (default) It was /me mentioning it on IRC, IIRC

I had to use higher values to "hack" over mod_dontdothat (my guess) and
get a couple of TLPs cloned. Values of 1000 or 10000 revisions were not
being harmful at the time, IIRC I talk about it with Justin a couple of
years ago, when I did my first experiments. The higher the value the
more memory the client takes, the more revisions asks the server pre
request (actual load depends on setup-tear down in the server, I'm not
sure). The higher the value the faster the cloning will complete.

Not sure about safe values, though. In our case, even asking for 10000
revisions in a subtree can make little load, if most of those revisions
fall outside of the tree, I guess.


> --j.
Santiago Gala

View raw message