Dave Miner wrote:
> Evan Layton wrote:
>> Dave Miner wrote:
>>> Sarah Jelinek wrote:
>>> ...
>>>>
>>>> I was actually thinking we would run the ICT stuff before we did the 
>>>> pkg image-update. Much like we do a pkg image-update on a live 
>>>> system today. So, I assume ips would take care of the configuration 
>>>> changes that are required during an image update.
>>>>
>>>> The process would be:
>>>> 1. Install the bits that match the current booted client.
>>>> 2. The normal ICT and post install stuff is run by AI.
>>>> 3. Then run pkg image-update if we have detected that the user asked 
>>>> for later bits than the booted client version.
>>>>
>>>> Do you think this would work?
>>>>
>>>
>>> I'm sure it would work, since it's exactly the path you'd follow in 
>>> updating from version R(unning) to version I(nstalled).  But I 
>>> question whether it's necessary...
>>>
>>>> As for upgrading the zpool version if we detected there was a 
>>>> version higher that could be supported.. I think this is required 
>>>> since the user is expecting to get the latest version supported with 
>>>> their request. We can manage this in a fairly straightforward way I 
>>>> think.
>>>>
>>>
>>> I think updating zpool versions is a fairly minor issue, not one I'd 
>>> focus on.  The only problem it causes is newer features that aren't 
>>> available until the upgrade, but that isn't an entirely unexpected 
>>> path, since if you've been doing updates, you'll have the same 
>>> problem, and hence there'll be documentation and so on covering the 
>>> issue.
>>
>> Good point that this is documented. However This does represent an 
>> issue where the result of the install is somewhat unexpected. In this 
>> case the zpool version of the system installed this way could differ 
>> from say a live CD install where the booted image and installer match 
>> the bits being installed. We had listed that this unexpected result 
>> was a problem and that it was a requirement that the installed systems 
>> should match. Does this seem like a requirement that we should remove 
>> and not worry about?
>>
> 
> It would be better if we can deal with it, but given that the 
> workarounds are a) install with the matching booted bits or b) run zpool 
> upgrade, it seems not to be something over which I'd torture the design. 
>  Postponing it to a first-boot task would seem to work well enough, 
> wouldn't it?

Yes doing this as a first-boot task would take care of it and that seems to be 
the way to deal with it. I just wanted to make sure this seemed to be a valid 
requirement.

-evan

Reply via email to