Re: [gentoo-dev] New feature proposition (make.conf)

[Chris Gianelloni]

On Thu, 2003-08-21 at 00:44, Stewart Honsberger wrote:
> Zack Gilburd wrote:
>
> > One of the problems that plagues Gentoo is that the distro's growth is leading
> > to an insane amount of distfiles downloads (and even excessive downloads)
> > along with excessive rsync'ing.  Automatic removal of distfiles would
> > substancially increase the amount of distfiles downloads due to the fact that
> > packages often have revisions made to them and revisions mean that the
> > software has to be recompiled -- thus the distfile would have to be
> > redownloaded of already removed.
> >
> > Until the general downloading (and rsyncing) etiqutte improves, I don't see
> > how this could be The Right Thing(TM) to do.
>
> I've been thinking along the lines of a cronjob being pulled in with,
> say, Portage. Said cron job would, on a daily/weekly basis remove old
> distfiles based upon age, and perhaps even have a setting to consider
> size. (Eg; always/never remove files greater/less than a certain size).
>
> That's where it gets tricky. OpenOffice, Mozilla et al. are two great
> examples of packages whose source tarballs are *LARGE*. On one hand,
> those would, in one fell swoop, free up the most HDD space = most
> benefeit. On the other hand, they'd also cost more bandwidth to
> re-download = most detrimental.
>
> Operating on a strictly age-based system based around file access time
> could potentially work, except that Gentoo's install defaults and/or
> suggests strongly the notion of 'noatime' in fstab entries.
>
> The script for the cronjob is easy, especially if it's only date-based.
> The politics involved in implementing it, however, could be hairy. I'll
> throw my script out there for review if it interests anybody and let
> someone who's more proficient in Bash scripting add in the filesize details.

I use a script which deletes any files in distfiles which do not belong
to an installed package.  This keeps old packages out of my distfiles,
but also allows me to keep enough distfiles locally to keep from
hammering the mirrors constantly.  I think possibly a combination of the
two should be in order?  Maybe remove all non-installed distfiles on a
weekly basis and also remove all distfiles that have not been accessed
in something like 90 days?  This would keep things around long enough to
probably make it through any revision changes but also manage to clear
up room for the people that are complaining about space.

-- 
Chris Gianelloni
Developer, Gentoo Linux