Karen Tung wrote:
> Glenn Lagasse wrote:
>> Hey Dave,
>>
>> * Dave Miner (dminer at opensolaris.org) wrote:
>>  
>>> Glenn Lagasse wrote:
>>>    
>>>> Could I get a reviewer (or more) to have a look at the changes I'm
>>>> proposing for 5265 (Distro constructor should auto-generate image
>>>> hashes) please?
>>>>
>>>> http://defect.opensolaris.org/bz/show_bug.cgi?id=5265
>>>> http://cr.opensolaris.org/~glagasse/slim_source/
>>>>
>>>> In case anyone is curious, I'm generating MD5, SHA-1 and SHA-256
>>>> cryptographic hashes.  SHA-256 may be a little overkill but it's *more
>>>> secure* than the others and I figured I'd try and appease the security
>>>> conscious crowd while I was at it (SHA-256 checksums are also starting
>>>> to show up in the wild from other distributors of electronic media such
>>>> as ours so I figured I'd get a jump up).
>>>>
>>>>       
>>> I'd suggested this be an option controllable in the manifest, which  
>>> perhaps is overkill, though on many systems it's going to be a bit  
>>> time-consuming to compute three different hashes on gigabyte-ish 
>>> files.  That could also be accomplished somewhat more obliquely if 
>>> this were  made a separate step in the finalizer, at least; any 
>>> reason you didn't  choose that design?
>>>     
>>
>> I did actually consider those designs (manifest option, finalizer
>> script).  In the end I didn't really see any compelling reason to
>> implement them.  They're more flexible certainly but I didn't think this
>> sort of feature needed that.  I considered image checksums part of the
>> same process of preparing the image done in create_{iso|usb}.  I suppose
>> it's possible that someone may not want to generate checksums for their
>> images.  In my testing (on a quad core 2 duo with 8g of ram) generating
>> checksums took on the order of seconds to generate the hashes for the
>> liveCD and usb images (less than 20 for a complete set for both) so I
>> didn't consider time as a significant issue even if you consider
>> building on 'slower' hardware.
>>
>> But, it's not difficult to implement either of the other designs.  I
>> think I'd lean more towards a combination of both, specify which hashes
>> you want in the manifest which a new finalizer script then parses.  You
>> then have flexibility in deciding which hashes you want generated as
>> well as checkpoint/resume ability when generating them.
>>
>> How does that sound?
>>
>>   
> IMO, generating of the hashes should be part of the same finalizer script
> that generates the associated images, that way, there's no ambiguity about
> which image the hashes are being generated for.  If we were to generate
> the hashes in a separate finalizer script, one will have to check 
> whether the
> given image exist or not, and the name also need to be pass in.
> It can work, but I think it is not as clean of a solution.
> 

I don't see where there's potential ambiguity, or why names need to be 
passed in.  My understanding of the architecture was that we maintained 
the manifest as a document in the ManifestServ that could be queried by 
each task for the data it needs.  This information is seemingly 
available there.

Dave

Reply via email to