1 |
Hi Gentoo Devs- |
2 |
|
3 |
Thank you to Martin MOKREJŠ and the others who contributed to the recent |
4 |
thread on new openafs ebuilds. I've been using my own ebuilds (they're ugly |
5 |
and I doubt that anyone in the Gentoo dev-team would be interested in them, |
6 |
but if someone's interested I'd be glad to share) with OpenAFS for about a |
7 |
year now and have been using OAFS with linux for maybe two years. OAFS has a |
8 |
great deal to offer and I'm glad to see somebody working on improving the |
9 |
support in Gentoo for OAFS. |
10 |
|
11 |
But I'm writing today to get some pointers on the subject of keeping various |
12 |
portage directories |
13 |
(/usr/portage, /usr/portage/packages, /usr/portage/distfiles, etc.) on afs |
14 |
volumes. |
15 |
|
16 |
I've been experimenting with this notion for a month or so and seem to be |
17 |
having no notable problems to speak of thus far, but I wonder if someone more |
18 |
expert in Gentoo than me and also familiar with OAFS could offer any |
19 |
comments. |
20 |
|
21 |
In my experiments, I have an afs volume called portage in the afs tree and I |
22 |
make a symlink in all the networked machines' local filesystems |
23 |
at /usr/portage that points to this volume. I have another afs volume called |
24 |
distfiles and there is a 'distfiles' symlink in the portage volume that |
25 |
points to it. I have also set PKGDIR to /usr/packages and |
26 |
symlinked /usr/packages to another afs volume. OAFS docs describe what seems |
27 |
to me to be a reasonably good system for storing system binaries on afs |
28 |
volumes with a general tree structure like this: |
29 |
|
30 |
/afs |
31 |
/afs/cell.name |
32 |
/afs/cell.name/i386_linux24/usr/afsws/{bin,lib} |
33 |
/afs/cell.name/i386_linux26/usr/afsws/{bin,lib} |
34 |
/afs/cell.name/ppc_linux26/usr/afsws/{bin,lib} |
35 |
|
36 |
and so on, and then making symlinks in the local fs to these volumes. I'm |
37 |
just trying to extend this notion to Gentoo. |
38 |
|
39 |
I made afs volumes under each of these architecture/linux-kernel trees called |
40 |
arch.kernel.packages. Then, going to each machine in the network, made |
41 |
symlinks from /usr/packages to the appropriate afs volume. |
42 |
|
43 |
This system seems to work for me pretty well, but I wonder if there are subtle |
44 |
issues that I'm overlooking that should be addressed. One issue that I have |
45 |
thought a little about is keeping readonly access to the afs volumes that all |
46 |
the machines need and obtaining write access to the appropriate afs volumes |
47 |
whenever running an emerge --sync or emerging a package or making a quickpkg |
48 |
out of an installed package. I have a scheme in place that works, but I'm |
49 |
sure there are many things I've overlooked with it. |
50 |
|
51 |
Does anyone have any thoughts to share on: |
52 |
|
53 |
a) general advisability of this (seems like a good thing to me---lots of |
54 |
savings on space across machines, oafs has a good authentication system in |
55 |
kerberos, seems better to me than running a local rsync server alone and also |
56 |
better in at least some ways than NFS, etc), |
57 |
b) what special considerations I should keep in mind with such a scheme, |
58 |
c) security, |
59 |
d) general reading material to help me think about a-c better. |
60 |
|
61 |
TIA. |
62 |
|
63 |
-- |
64 |
Kevin |
65 |
http://www.gnosys.us |
66 |
|
67 |
-- |
68 |
gentoo-dev@g.o mailing list |