[indiana-discuss] beadm activate issue while upgrade to snv_94

[Dave Miner]

Mike Gerdts wrote:
> On Tue, Jul 29, 2008 at 10:06 PM, Praveen Kumar <praveen at sun.com> wrote:
>>> As I just posted to caiman-discuss and pkg-discuss, it seems as though
>>> the only way to have a good deal of certainty that you will be able to
>>> create, update, and activate a new boot environment, you need to be
>>> sure that you have 2x the amount of free space as the amount of space
>>> used by your active boot environment.
>>>
>>> http://mail.opensolaris.org/pipermail/caiman-discuss/2008-July/004739.html
>>>
>>> I see this as a bug or at least a really important RFE related to how
>>> "zfs promote" works.
>> praveen at athena:~$ beadm list
>>
>> BE                 Active Active on Mountpoint Space
>> Name                      reboot               Used
>> ----               ------ --------- ---------- -----
>> opensolaris-snv_93 yes    yes       /          8.67G
>> opensolaris-snv_94 no     no        -          1.03G
>>
>> Does it mean that I should have atleast 17G free before I can activate my
>> new be? How am I gonna afford that!
> 
> 
> The 2x is the worst case and assumes that everything will be replaced
> in the new BE.  However, your new BE only used about 1 GB.  Before
> "zfs promote" (and similar operations by beadm or pkg) will work you
> seem to need 8.67 GB free at the time that the promote is done.  As
> such, rather than 17 GB the total needed to be able to create,
> populate, then activate the BE in this case seems to be 9.70 GB
> (before the BE is created).  When you get done with it, you will still
> have the same amount of free.
> 
> The essence of the bug that I believe exists is that an unreasonable
> amount of free space is required to do a promote.  I don't believe
> that there is any point in time or failure mode that actually requires
> this space - I think it is an erroneous check in the zfs promote code
> path.  Then again, I don't know that code so others are certainly
> better placed to provide an explanation as to why it is required.
> 

I think part of the problem here is that we may be excessively 
aggressive in promoting BE's.  With the current behavior you end up with 
all snapshots attached to the most recent BE, which exacerbates the 
issue, I think.  Needs some investigation to determine if we can be 
smarter about this.

Dave