The Packaging committee has been discussing users and groups for a long time, coming up with increasingly elaborate methods of solving the problems with dynamic allocation of uids.
As an alternative we want to explore biting the bullet and just allocating another uid range for static assignment. We thought of two possible ways to do this that made some sense:
1) Allocate the uids at the high end of the 32 bit range: 2**32-(range) to avoid the uids being used on as many systems as possible.
2) Work with another distro to share the range of static uids.
What do people think of this? Will it cause pain for too many sites? Is it an acceptable cost to avoid having to debate dynamic vs static uids for every package review in the future?
-Toshio
On Tue, Jul 24, 2007 at 12:34:34PM -0700, Toshio Kuratomi wrote:
The Packaging committee has been discussing users and groups for a long time, coming up with increasingly elaborate methods of solving the problems with dynamic allocation of uids.
As an alternative we want to explore biting the bullet and just allocating another uid range for static assignment. We thought of two possible ways to do this that made some sense:
- Allocate the uids at the high end of the 32 bit range: 2**32-(range)
to avoid the uids being used on as many systems as possible.
- Work with another distro to share the range of static uids.
What do people think of this? Will it cause pain for too many sites? Is it an acceptable cost to avoid having to debate dynamic vs static uids for every package review in the future?
Ville's proposal was fine, what are we really trying to fix that will force us to break with FHS and several large sites?
If static uids are ***really*** required for a networked application, then you don't only have to coordinate with "another distro", but with all of them as well as all Unices, so it's a lost cause to start with.
And if one argues that one doesn't need the same uid/gid on all systems to start a Unix-global registry, then why doe Fedora need it?
Don't fix something that isn't broken.
On Thursday 26 July 2007, Axel Thimm wrote:
Ville's proposal was fine
For the record, it was me who brought this (finding out if we could go almost-all-static mappings) up in the meeting. I'm still fine with "my" draft, but I thought it'd be a good idea to explicitly discuss and evaluate the static alternative for real too before the dynamic approach is voted on.
On Sat, Jul 28, 2007 at 12:05:29PM +0300, Ville Skyttä wrote:
On Thursday 26 July 2007, Axel Thimm wrote:
Ville's proposal was fine
For the record, it was me who brought this (finding out if we could go almost-all-static mappings) up in the meeting. I'm still fine with "my" draft, but I thought it'd be a good idea to explicitly discuss and evaluate the static alternative for real too before the dynamic approach is voted on.
OK, no problem. Next meeting we'll start discussing about switching to deb format for packages. ;)
Le Mar 24 juillet 2007 21:34, Toshio Kuratomi a écrit :
- Work with another distro to share the range of static uids.
What do people think of this? Will it cause pain for too many sites? Is it an acceptable cost to avoid having to debate dynamic vs static uids for every package review in the future?
+1 for finding interested contacts in other distros and just submit a request to the LSB to have the static range expanded
You have to realise static UID scarcity is a big reason UIDs vary from distro to distro - given UID shortage each distro priorises the packages it cares most about and as a result the attribution choices vary.
With a non-scarce range new UIDs and (and later historical UIDs) could just nicely converge.
On Thu, 2007-07-26 at 10:07 +0200, Nicolas Mailhot wrote:
Le Mar 24 juillet 2007 21:34, Toshio Kuratomi a écrit :
- Work with another distro to share the range of static uids.
What do people think of this? Will it cause pain for too many sites? Is it an acceptable cost to avoid having to debate dynamic vs static uids for every package review in the future?
+1 for finding interested contacts in other distros and just submit a request to the LSB to have the static range expanded
You have to realise static UID scarcity is a big reason UIDs vary from distro to distro - given UID shortage each distro priorises the packages it cares most about and as a result the attribution choices vary.
With a non-scarce range new UIDs and (and later historical UIDs) could just nicely converge.
With 4 billion UIDs claiming UIDs are a scarce resource is a bit excessive. It's just that nobody has the courage to break with the past and bump up the reserved UID space or even just move to a more creative approaches.
Simo.
On Thu, 26 Jul 2007, Nicolas Mailhot wrote:
+1 for finding interested contacts in other distros and just submit a request to the LSB to have the static range expanded
Mats Wichmann of the LSB, and I discussed Linux Assigned Names And Numbers Authority matters at OLS a few weeks ago at some length, and the matter was discussed in the LSB weekly conference call on Wednesday with him and me leading the discussion [the calls are open to persons willing to participate and work -- ask off list if Fedora or Red Hat wish details for calling in -- each has been absent from th process for reasons unknown to me in recent years. -- Alan Cox occastionally chimes in on the mailing list, always with insightful comment, however] http://www.lanana.org/ and there is certainly an interest at the LSB in helping reach inter-vendor coordination in this space.
Please feel free to CC me on any proposal for extension as it is worked up.
-- Russ Herrold herrold@owlriver.com
On Fri, 2007-07-27 at 10:43 -0400, R P Herrold wrote:
On Thu, 26 Jul 2007, Nicolas Mailhot wrote:
+1 for finding interested contacts in other distros and just submit a request to the LSB to have the static range expanded
Mats Wichmann of the LSB, and I discussed Linux Assigned Names And Numbers Authority matters at OLS a few weeks ago at some length, and the matter was discussed in the LSB weekly conference call on Wednesday with him and me leading the discussion [the calls are open to persons willing to participate and work -- ask off list if Fedora or Red Hat wish details for calling in -- each has been absent from th process for reasons unknown to me in recent years. -- Alan Cox occastionally chimes in on the mailing list, always with insightful comment, however] http://www.lanana.org/ and there is certainly an interest at the LSB in helping reach inter-vendor coordination in this space.
Please feel free to CC me on any proposal for extension as it is worked up.
Russ, it seems to me like there is some logic in having a centralized "assignment" for static UID/GIDs, across distributions. Is this something that LANANA would be interested in maintaining?
~spot
On Fri, 27 Jul 2007, Tom "spot" Callaway wrote:
Mats Wichmann of the LSB, and I discussed Linux Assigned Names And Numbers Authority matters at OLS a few weeks ago at some length, and the matter was discussed in the LSB weekly conference call on Wednesday with him and me leading the ...
http://www.lanana.org/ and there is certainly an interest at the LSB in helping reach inter-vendor coordination in this space.
Russ, it seems to me like there is some logic in having a centralized "assignment" for static UID/GIDs, across distributions. Is this something that LANANA would be interested in maintaining?
While I need to run it past the LSB side, it is a natural fit for LANANA --- we'll discuss it Wednesday, I imagine.
-- Russ
packaging@lists.fedoraproject.org
-
Axel Thimm -
Nicolas Mailhot -
R P Herrold -
Simo Sorce -
Tom Callaway -
Toshio Kuratomi -
Ville Skyttä