William Schumann wrote:
> Jan,
> My responses below are from testing performed today against Solaris 98 
> and my modified auto-install and liborchestrator.
>
> jan damborsky wrote:
>> Hi William,
>>
>> the overall document looks pretty good,
>> I have only couple of questions/comments -
>> please see below.
>>
>> Thank you,
>> Jan
>>
>> Slice and partition management actions
>> --------------------------------------
>> * Partition
>>   Will be starting offset required ? What happens
>>   is the partition overlaps the exiting one ?
> There is currently little checking at this layer.  If overlapping 
> occurs, fdisk fails, the error message is logged, an error code 
> returned, and AI fails.

ok.

>>   I might recommend to support similar approach which
>>   is to be taken for slices - specify only partition
>>   size or allow AI to pick up the largest contiguous
>>   chunk of available space.
> This would indeed be simpler, but in the little time remaining before 
> the release, we might avoid this change and require the user to specify 
> the starting sector and size for both creation and deletion.  The user 
> can also use other partitioning tools for this preview release.

Agreed - given the time constraints, this should be postponed for later.

>>
>> Defining desired behavior
>> -------------------------
>> * <target_device_use_solaris_partition> = false
>>
>>   - partition information specified
>>     What would be the format of partition information ?
>>     Is user requested to specify complete partition
>>     information - what is optional, what is mandatory ?
> The AI engine manifest should include this information for partitions:
> action = create or delete
> partition number (currently ignored)
> starting sector
> size in sectors
> partition type (currently ignored - creates Solaris only)
>
> For both creation and deletion, the starting sector and size only are used.
>
> To delete a partition, there must be an exact match on starting sector 
> and size, or an error will be returned and the installation will terminate.
>
> Here is a sample of the AI engine manifest for a partition:
> <ai_device_partitioning>
>    <partition_action>create</partition_action>
>    <partition_number>1</partition_number>
>    <partition_start_sector>200</partition_start_sector>
>    <partition_size>20480</partition_size>
>    <partition_type>99</partition_type>
> </ai_device_partitioning>
>
> Testing reveals that any partition information missing will cause a 
> manifest verification failure (as well as a core dump in the version I'm 
> working on).  This includes information that is ignored by the partition 
> editing code I've written.
>
> Since for the November release, the user edits the AI engine manifest by 
> hand, we should remove the fields that are required but ignored from the 
> sample template, and to remove the requirement from the manifest parser.

 I see - thanks a lot for clarifying this.

>>   - no partition information
>>     I might recommend not to take 'whole disk' as
>>     default if no partition information is specified.
>>     This seems dangerous - I think that if user wants
>>     to wipe out all the disk, it should be explicitly
>>     stated in the manifest.
> Opinions from anyone else?
>> Current behavior
>> ----------------
>> Only to confirm - it seems this section describes current
>> implementation in AI - might it be correct ?
> Yes, we're talking about AI and code that is under development but not 
> yet in the repo.

ok.

>> To be done for November
>> -----------------------
>> - What will be implemented as far as creating Solaris2
>>  partition is concerned ?
> The plan is only to create Solaris2 partitions with create and delete 
> actions, as described above.  The current code supports this.

ok.

>> - With respect to minimum and recommended sizes, they
>>  are slightly different comparing with the ones
>>  in GUI installer.
>>
>>  The reason is that we don't have /.image_info file for
>>  AI, which might tell us, how much space is needed for
>>  installation.
>>  Current behavior is that if this information is not
>>  available defaults are picked up by the orchestrator:
>>  4GiB for minimum, ~12GiB for recommended.
> Sundar indicated that this will not be added for November.  There is a 
> field in the AI engine manifest of minimum device size (as opposed to 
> partition size) that is used in criteria checking.

Speaking about /.image_info file, I think it is not going to be very
meaningful for AI, since the bits to be laid out are not hardcoded as
for LiveCD, they are provided in AI manifest. I think we will need
to implement other mechanism, for instance deduce the required disk
space from package size information.

>> - Switching to read_extvtoc() and write_extvtoc() will
>>  allow to install on <1TB; 2TB) disks. Only 2TB will
>>  be usable on >2TB disks.
> OK.  There are existing OpenSolaris bugs with terabyte disks that can 
> hopefully be fixed by the release.

ok.

Thanks !

Jan

> William
>>
>>
>> William Schumann wrote:
>>> Slice (VTOC) and partition (fdisk) management for the Automated 
>>> Installer have not yet been specified, yet safe and sensible behavior 
>>> is expected.
>>>
>>> I have prepared a document (attached) with the following objectives:
>>> - document the expected features and behavior
>>> - document the current behavior
>>> - document what is yet to be coded
>>> - serve as a starting point so we may achieve consensus on what the 
>>> proper behavior should be
>>>
>>> Please review the the attached document and comment.  Check my 
>>> assumptions, offer corrections and suggestions.
>>> William
>>>
>>> ------------------------------------------------------------------------
>>>
>>> _______________________________________________
>>> caiman-discuss mailing list
>>> caiman-discuss at opensolaris.org
>>> http://mail.opensolaris.org/mailman/listinfo/caiman-discuss
> _______________________________________________
> caiman-discuss mailing list
> caiman-discuss at opensolaris.org
> http://mail.opensolaris.org/mailman/listinfo/caiman-discuss


Reply via email to