Hi Alexander,
I wanted to follow up with you on this bug. I spoke with Ethan
yesterday, and I think for now we have to put your fixes for this bug on
hold.
The overall concerns I have are:
1. Changing the behavior of install to actually write to the disk before
the install has actually started. I know in the case of AI the user has
really started the install by booting the client, but it seems as if
what you are proposing is to label the disk part way in to reading the
manifest, and then proceed with the rest of the manifest processing to
start the installation.
2. Adding this flag to the schema doesn't seem like the right thing to
do in general. My philosophy about adding things to our schema is that
we only do this if it is something that we must have user input on. I am
not sure that this bug indicates that this is the answer.
3) I agree that we need the disk geometry information but, I think there
are several issues we need to look at with regard to this bug:
A) First, liborchestrator shouldn't fail if we don't have
all the attributes for a disk. liborchestrator will be going
away eventually so making the change in this code is likely not
the right thing to do.
B) We need to figure out if we can get geometry information in
libdiskmgt even if there isn't a label, for sparc. Right now we
call GGEOM to get information about a disk on sparc, but if
there is no label, it doesn't succeed. I am investigating this.
C)If it isn't possible to get geometry information about a disk
on sparc without the disk being labeled, we need to understand
what the user experience should be to make a disk usable by the
installers.
Thank you for working on this! It is very much appreciated. However, at
this time let's keep this work on hold until we understand the best way
to proceed.
Regards,
sarah
****
On 04/ 6/10 09:18 AM, Sarah Jelinek wrote:
On 04/ 6/10 09:06 AM, William Schumann wrote:
This new element deserves some scrutiny, since it affects the design and
configuration of the Automated Installer. There needs to be some
explanation of how this works: under what circumstances the disk is
actually labeled, what is the behavior if the manifest is reused, what
if an EFI label exists? Suggest that the design changes be floated in
caiman-discuss by proposing changes (perhaps using an algorithm) and
soliciting input in bugzilla, and afterwards, fix it accordingly.
auto_install.c:
As coded, the Target Instantiation functions should not be called
directly, but through ti_create_target(), and only during the TI phase.
This is presently in perform_slim_install:do_ti()
ai_manifest.rng:
The new element belongs in the ai_target_device section, since it
pertains to a particular disk target. Provide some brief comments in the
schema, since we are moving to improving documentation within the schema.
Actually, I am not sure I am comfortable with adding something to the
existing schema. In particular, since it doesn't appear to me this bug
is a stopper in any way.
We are working on schema redesign now. Adding additional tags to the
existing schema is going to be throw away code. I also think that as
part of the new schema and semantic validation the automatic creation of
a label is more about the behavior of AI, as opposed to a specific user
specification.
I can certainly consider this as an element in the new schema, but if
the user has explicitly indicated they want to use a device(which I
believe this bug addresses), and it doesn't have a label, I think the
behavior should be that we label the device so we can use it.
thanks,
sarah
****
To be consistent with other elements in the section, it should be
prefaced with "target_device_". Suggest rather
"target_device_label_disk" so that "label" is an active verb. Should
this be named "...label_if_absent" or something to more closely reflect
what actually happens or the intended usage of this element?
On 04/ 6/10 03:12 PM, Alexander Eremin wrote:
Please review the fixes for:
6260 sparc AI need a way to mechanism to explicitly ask for creating
label on Sparc disk
Bug:
http://defect.opensolaris.org/bz/show_bug.cgi?id=6260
Webrev:
http://cr.opensolaris.org/~alhazred/6260/
Tested on sparc and x86 (not affected) successfully.
Part of sample install_log in attachment.
What is the behavior if the disk is already labeled? If the disk has
existing slices and custom slices are specified in the manifest?
William
_______________________________________________
caiman-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/caiman-discuss
_______________________________________________
caiman-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/caiman-discuss
_______________________________________________
caiman-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/caiman-discuss
