Hi Jack,
On 10/ 5/10 04:52 PM, Jack Schwartz wrote:
Hi Jan and Sanjay.
On 10/05/10 00:04, Jan Damborsky wrote:
On 10/ 5/10 02:47 AM, sanjay nadkarni wrote:
On 09/15/10 05:05 PM, Jack Schwartz wrote:
Hi Jan.
Looks like a DDU bug. See below...
On 09/15/10 06:05 AM, Jan Damborsky wrote:
Hi Jack,
thank you very much for looking into this !
I have done more investigation in order to narrow
down the issue - please see my response below.
Jan
On 09/15/10 03:13 AM, Jack Schwartz wrote:
Hi Jan.
On 09/14/10 02:01 AM, Jan Damborsky wrote:
Hi Jack and Zhong-Yuan,
as part of evaluating potential fix for bug 16714, I have built
AI Sparc image
based on 147 skipping compress() step in boot_archive_archive.py
finalizer [1].
I basically commented out following section:
...
if IS_SPARC:
# print "Doing compression..."
# try:
# compress(BA_BUILD, BA_LOFI_MNT_PT)
# except Exception:
# release_archive()
# raise
...
However, in this case DDU fails with following messages in
/tmp/install_log:
$ more /tmp/install_log
<AI Sep 10 15:29:24> /tmp/ai_combined_manifest.xml manifest
setup and validated
<AI Sep 10 15:29:26> ai_du_get_searched_pkg_list: Doing a device
scan for devices which are missing drivers...
<AI Sep 10 15:29:34> ai_du_get_searched_pkg_list: Querying
manifest for explicit repo for getting missing driver packages...
<AI Sep 10 15:29:34> ai_du_get_searched_pkg_list: Found repo in
manifest: publisher:opensolaris.org, origin:http://ipkg.sfbay/dev
<AI Sep 10 15:29:50> ai_du_get_searched_pkg_list: Manifest
allows adding of third-party drivers
<AI Sep 10 15:30:03> ai_du_get_searched_pkg_list: DDU returned
package info for "video" type device "3DLabs Wildcat Realizm 500".
<AI Sep 10 15:30:03> ai_du_install_packages: Installing packages
to /
Looks like the problem starts here:
<AI Sep 10 15:30:03> UNK package at origin:
but I suspect there are two problems.
1) DDU really couldn't find the package (even though the
installer says it did ("DDU returned package info for...").
It seems this might be it - I have built two identical AI images
with boot archive
uncompressed. The only difference is in publisher configured for
created image:
AI image where DDU works reports:
# pkg publisher
PUBLISHER TYPE STATUS URI
opensolaris.org (preferred) origin online
http://ipkg.sfbay/dev/
And the one with DDU failing has:
# pkg publisher
PUBLISHER TYPE STATUS URI
opensolaris.org (preferred) origin online
http://ipkg.sfbay/dev/
opensolaris.org (preferred) mirror online
http://ipkg.czech:8000/
So in the latter case, mirror was added to the 'post install'
publisher.
It seems that the mirror is not problem for 'pkg search' itself -
it seems to do the right job:
# pkg search SUNW,XVR-2500
INDEX ACTION
VALUE
PACKAGE
alias driver ['pci3d3d,170', 'SUNW,XVR-2200', 'pci3d3d,176',
'SUNW,XVR-2400', 'pci3d3d,171', 'pci3d3d,1052', 'pci3d3d,1055',
'pci3d3d,178', 'pci3d3d,177', 'pciex3d3d,177', 'pciex3d3d,178',
'SUNW,P25-LITE', 'SUNW,P25-HEAVY', 'SUNW,XVR-2500',
'pci3d3d,1050', 'pciex3d3d,1050']
pkg:/driver/graphics/[email protected]
So the issue seems to be on DDU side
and it is not related to fix for bug 16714 (now tracked in bugster
as 6984618), since DDU works with uncompressed Sparc boot archive.
Based on that, from DDU point of view, are you fine with
omitting compression step as a fix for 6984618 or do you happen
to envision any DDU problems if we go with uncompressed
Sparc boot archive ?
Yes. I verified the same:
When I run comp_lookup.sh I see the bugus error message I
mentioned yesterday when publishers are set as you have them above,
with ipkg.czech:8000 as a mirror:
Can not find /tmpsparc/find_driver or not executable.
(Note to DDU folks: I copied the scripts to /tmp so I could do set
-x in them and trace them.)
When I remove the mirror and keep only ipkg.sfbay/dev and rerun
comp_lookup.sh, it terminates with a more normal output:
SUNW,XVR-100:IPS
Note that I called comp_lookup.sh the way the DDU does as part of
its processing, specifying SUNW,XVR-100 as an argument (for this
particular system/situation). Not sure of the relevance that
comp_lookup.sh just returned the same argument passed into it, with
":IPS" appended, when the mirror was removed.
That said, it's a bug that comp_lookup.sh bombed out with an error
under an acceptable pkg publisher configuration, but terminated
normally under a different acceptable configuration.
FYI, I could get to ipkg.czech:8000 in the browser, although it
said I couldn't look at packages there as it was running as a mirror.
I have filed the following bug:
6985156 DDU can't search mirrored IPS repositories
This is no change. Perhaps an RFE. Since the mirrors *only* contain
the data and not the metadata I therefore think that one cannot do a
search on the mirrored repos.
In this particular case, DDU failed due to the fact that besides
IPS repo with metadata, publisher also contained IPS content mirror:
# pkg publisher
PUBLISHER TYPE STATUS URI
opensolaris.org (preferred) origin online
http://ipkg.sfbay/dev/
opensolaris.org (preferred) mirror online
http://ipkg.czech:8000/
That can happen, since it seems DDU uses existing image's IPS metadata
to manage DDU publishers and DistroConst post-install IPS publishers
are not cleaned up first.
If <search_all> does not include a publisher, the one already on the
system is used.
Thank you for pointing this out. I was not aware of this default case.
The DDU failed if that configuration had a mirror. However, we have
tested a fix from the DDU team and this bug has been resolved.
Thinking about this, it seems there are two aspects of this issue:
* DDU IPS publisher should exactly match configuration specified in
AI manifest
in section add_drivers/source/publisher
The DDU should now do this. If it doesn't, that is a bug.
I see. I didn't try the newest bits. So it seems that the observations I
did some time ago
are likely outdated for now.
Note though that the add_drivers/source/publisher entry, at least at
this time, takes only a single publisher name and repository.
* DDU should work correctly when IPS content mirror is part of DDU
publisher.
It will likely be not an issue if former part is addressed, since I
assume
that amount of data transferred during DDU process is likely small, so
configuring DDU publisher with IPS mirror is not a big win.
See above.
Thank you !
Jan
_______________________________________________
caiman-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/caiman-discuss
