Drew,
Thanks for putting this together so quickly:
I've carried out the following zone related tests :
1. Install a zone specifying no ai_manifest or sc_profile thus it uses
default zone_default.xml manifest and runs sysconfig on first boot.
Zone installed, and on first boot prompted for root password and
initial user account details
Zone booted cleanly, and logged in as new initial user and verified
dataset rpool/export/home's mountpoint is inherited from parent
rpool/export and set to /export/home
2. Install a zone specifying a sc_profile which automatically sets
root
password and creates a new initial user.
Zone installed, and boot cleanly. Logged as initial user and verified
dataset rpool/export/home's mountpoint is inherited from parent
rpool/export and set to /export/home
3. Install a zone specifying a custom AI manifest which sets
rpool/export's mountpoint to be "/foo"
Zone installed and on initial boot prompted only for root password
as not prompted for initial user details.
Zone booted cleanly, and logged in as root and verified dataset
rpool/export was mounted at /foo and also verified
dataset rpool/export/home's mountpoint is inherited from parent
rpool/export and was set to /foo/home.
4. Install a zone specifying a custom AI manifest, which sets
rpool/export's mountpoint to be "/foo" and also specify a sc_profile
to set user password and create initial use.
Zone booted cleanly, logged in as initial user. and verified dataset
rpool/export was mounted at /foo and also verified
dataset rpool/export/home's mountpoint is inherited from parent
rpool/export and was set to /foo/home.
All tests passed successfully, just one quirk in test 3, where only
root password was prompted for on initial boot and not initial user
details. My guess is the sysconfig must check the existence of
/export/home and if not there will not prompt for initial user
details.
cheers
Matt
On 05/31/12 20:54, Drew Fisher wrote:
All:
I went ahead and implemented the fix from within
Filesystem.from_xml()
and the corresponding fix for DC. Mary is still sanity testing the
images, but since I'm out of the office Friday and Monday and Matt is
out Monday, Ethan and I felt that these changes needed to be reviewed
ASAP to make b18's closure on Monday. These changes supersede Matt's
changes to instantiation_zone.py. That file will not be changed with
these changes.
Here's a pointer to the webrev:
https://cr.opensolaris.org/action/browse/caiman/drewfish/filesystem_from_xml/webrev/
The code in Filesystem.from_xml() was added way back in b156 for
DC to
handle a target specification in which a zpool/fs dataset didn't
mount
at /zpool/fs on the filesystem. What DC has been doing since then is
passing strings around like dataset.mountpoint, dataset.action,
dataset.name, etc, rather than simply passing the DOC object. The
changes I made removes these strings and instead simply passes the
DOC
object in and out of the various DC functions.
The fix is to simply let ZFS deal with the mountpoint on its own. If
the
user supplies it as a attribute on the Filesystem element, we'll
explicitly set it. If not, ZFS will handle it for us. If we do it
this
way, we can remove the need for an entire target-instantiation
checkpoint from within DC. We can simply use Filesystem.create()
on the
4 datasets we need.
Matt has test data from various zone installs that I'll ask he
provide
as a reply to this mail. Also, I fixed a bug Jack just filed about an
invalid zpool name on the system.
Unittests are clean -
DC:
[mox:test] > sudo nosetests -s test_*.py
Ran 68 tests in 54.928s
OK
Target:
[mox:test] > sudo nosetests -sx test_*.py
Ran 268 tests in 24.744s
OK
I've also tested a DC manifest to make sure ZFS is doing the right
thing:
<target name="desired">
<logical>
<zpool name="tank" action="use_existing">
<filesystem name="mp/dcmp/ai" action="preserve"/>
</zpool>
</logical>
</target>
In which the mountpoint for tank/mp is at:
[mox:test] > zfs get mountpoint tank/mp
NAME PROPERTY VALUE SOURCE
tank/mp mountpoint /export/home/overhere local
The DC datasets are:
[mox:test] > zfs list -r tank/mp
NAME USED AVAIL REFER MOUNTPOINT
tank/mp 320K 648G 36.0K /export/home/overhere
tank/mp/dcmp 284K 648G 36.0K /export/home/overhere/dcmp
tank/mp/dcmp/ai 248K 648G 38.6K /export/home/overhere/dcmp/ai
tank/mp/dcmp/ai/build_data 78.6K 648G 37.3K
/export/home/overhere/dcmp/ai/build_data
tank/mp/dcmp/ai/logs 95.9K 648G 95.9K
/export/home/overhere/dcmp/ai/logs
tank/mp/dcmp/ai/media 34.6K 648G 34.6K
/export/home/overhere/dcmp/ai/media
As you can see, zfs create / DC is doing the correct thing.
Thanks!
-Drew (on behalf of Matt and Ethan)
_______________________________________________
caiman-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/caiman-discuss
