On Fri, May 22, 2009 at 6:15 AM, Martin Bochnig <martin at martux.org> wrote:
> On Fri, May 22, 2009 at 6:07 AM, Shawn Walker <swalker at opensolaris.org>
> wrote:
>> Martin Bochnig wrote:
>>>
>>> On Fri, May 22, 2009 at 5:35 AM, Shawn Walker <swalker at opensolaris.org>
>>> wrote:
>>>>
>>>> Martin Bochnig wrote:
>>>>>
>>>>> #4.) ? pfexec pkg set-publisher -P -O http://pkg.opensolaris.org/dev dev
>>>>>
>>>>> #5.) ? pfexec pkg unset-publisher opensolaris.org
>>>>>
>>>>> #6.) ? pfexec pkg refresh
>>>>>
>>>>> #7.) ? pfexec pkg install entire
>>>>>
>>>>> #8.) ? pfexec pkg install redistributable
>>>>
>>>> The steps above are not correct. Here's why:
>>>>
>>>> A new publisher should not be added, instead the existing publisher's
>>>> repository should be changed. ?The pkg refresh isn't necessary as install
>>>> will automatically refresh if needed. The install entire isn't necessary
>>>> as
>>>> that is already installed.
>>>>
>>>> These are the correct steps:
>>>>
>>>> pfexec pkg set-publisher -O http://pkg.opensolaris.org/dev
>>>> opensolaris.org
>>>>
>>>
>>>
>>> Technically it doesn't matter how you call it (last field).
>>> It is just a name.
>>
>> Not quite. ?There are subtle behaviours in the system when you do it the way
>> you listed.
>>
>>> I _wanted_ to chose a different authority-/publisher- name, because as
>>> of now OS2009.06 is still only available as an unsupported *Preview*.
>>> And the authority which - as of now - officially carries the name
>>> "opensolaris.org" is still for 2008.11 .
>>> Therefore I randomly chose the name "dev" (because its mnemonic, as
>>> the branch itself is called "dev") which makes it necessary the
>>> explicitly run ...
>>
>> Nevertheless, you shouldn't do that. ?If you want the packages from the /dev
>> repository, you should be switching the opensolaris.org publisher to use the
>> /dev repository:
>>
>> pfexec pkg set-publisher -O http://pkg.opensolaris.org/dev opensolaris.org
>>
>> If you want to reset the publisher afterwards, you would do this:
>>
>> pfexec pkg set-publisher -O http://pkg.opensolaris.org/release
>> opensolaris.org
>>
>> But that's pointless until the official release is pushed to /release.
>>
>>> Well, it does not harm if you (re-) install "entire" before that.
>>> Probably it is redundant. But from your nice "official" instructions at
>>> least
>>>
>>> pfexec pkg refresh
>>>
>>> would be missing, which is run automatically as part of the
>>>
>>> pfexec pkg install entire
>>>
>>> which I suggested. What you get byy strictly following the official
>>> instructions would be an early end:
>>
>> The correct way to handle that is to do this after changing the publisher:
>>
>> pfexec pkg image-update
>>
>> Cheers,
>> --
>> Shawn Walker
>>
>
>
>
> Aha. Advice much appreciated and accepted.
> Thanks!
>
>
> --
> %martin
>
$ pfexec pkg install redistributable
DOWNLOAD PKGS FILES XFER (MB)
Completed 745/745 181408/181408 1519.70/1519.70
PHASE ACTIONS
Install Phase 74640/217564 Action install
failed for 'opt/onbld/etc/abi/ABI_i386.db' (pkg:/SUNWonbld):
MemoryError:
pkg: unable to create BE opensolaris_2009-05-22-07:40:32
pkg: An unexpected error happened during installation:
Traceback (most recent call last):
File "/usr/bin/pkg", line 2475, in ?
__ret = main_func()
File "/usr/bin/pkg", line 2424, in main_func
return install(mydir, pargs)
File "/usr/bin/pkg", line 755, in install
api_inst.execute_plan()
File "/usr/lib/python2.4/vendor-packages/pkg/client/api.py", line
520, in execute_plan
self.img.imageplan.execute()
File "/usr/lib/python2.4/vendor-packages/pkg/client/imageplan.py",
line 725, in execute
p.execute_install(src, dest)
File "/usr/lib/python2.4/vendor-packages/pkg/client/pkgplan.py",
line 254, in execute_install
dest.install(self, src)
File "/usr/lib/python2.4/vendor-packages/pkg/actions/file.py", line
165, in install
shasum = misc.gunzip_from_stream(stream, tfile)
File "/usr/lib/python2.4/vendor-packages/pkg/misc.py", line 291, in
gunzip_from_stream
ubuf = dcobj.decompress(buf)
MemoryError
pkg: This is an internal error. Please let the developers know about this
problem by filing a bug at http://defect.opensolaris.org and including the
above traceback and this message. The version of pkg(5) is '05eb4fbadc69'.
martin at opensolaris:~$ df -h /tmp
Filesystem Size Used Avail Use% Mounted on
swap 369M 24K 369M 1% /tmp
martin at opensolaris:~$ grep -n swap /etc/vfstab
10:swap - /tmp tmpfs - yes -
12:/dev/zvol/dsk/rpool/swap - - swap -
no -
martin at opensolaris:~$
martin at opensolaris:~$
martin at opensolaris:~$ prtconf|grep emory
Memory size: 1024 Megabytes
$ swap -l
swapfile dev swaplo blocks free
/dev/zvol/dsk/rpool/swap 182,2 8 1048568 499072
$ zfs list
[...]
rpool/swap 522M 1.95T 522M -
Why can the BE not be created, what happens? May this be related to
calling the authority "dev" (before Shawn corrected me, I then kept it
running for curiosity)?
It is not possible that it really ran out of memory, as more than
enough swap is available, so it must be a bug, which is the reason why
I write this follow-up message.
What I also noticed is that IPS forces the system into its knees (as
long as it runs, before it crashes).
With conary I never witnessed anything like this, not remotely. As
conary and IPS are implemented in the same languages (Python and for
its backends C), it must be possible to fix IPS.
As you see I removed all cross-postings from CC, because only the
engineers at caiman-discuss can help here.
--
%martin
