Please provide process for small, targeted fixes in tools/regression

[Peel, Casey]

> Knowing now what I knew then, it's probably one of the modifications
> that was made to geom_mirror to be more robust by automatically

Unlikely since this was failing well before we brought in our gmirror
changes.

# mdconfig -a -t malloc -s 10m -u 0
# mdconfig -a -t malloc -s 10m -u 1

# gmirror label mirror /dev/md0GEOM_MIRROR: Device mirror/mirror launched
(1/1).
# gmirror insert mirror /dev/md1
GEOM_MIRROR: Device mirror: rebuilding provider md1.
GEOM_MIRROR: Device mirror: rebuilding provider md1 finished.

# gmirror remove mirror /dev/md0GEOM_MIRROR: Device mirror: provider md0
destroyed.
# gmirror remove mirror /dev/md1
gmirror: mirror: Can't remove the last ACTIVE component /dev/md1.


-- Casey


-- 
Performance Test Ninja
casey.peel at isilon.com / 206.777.7945






On 5/1/14 3:53 PM, "Garrett Cooper" <yaneurabeya at gmail.com> wrote:

>On Thu, May 1, 2014 at 3:44 PM, Peel, Casey <casey.peel at isilon.com> wrote:
>>> but out of curiosity why are you
>>> removing the geom remove testing in the geom_mirror testcases? Does
>>> the test not work out of the box on vanilla FreeBSD?
>>
>> To be honest, I haven't tested on a stock FreeBSD system, but on our
>> internal almost-but-not-quite FreeBSD-CURRENT system removing the last
>> partition of a gmirror fails. To remove the last partition you have to
>> destroy the gmirror. It should be easy to confirm this is still true on
>> FreeBSD-CURRENT. That said, I have the gmirror tests entirely moved over
>> to ATF so I'm not sure its worth messing with.
>
>Knowing now what I knew then, it's probably one of the modifications
>that was made to geom_mirror to be more robust by automatically
>rebuilding the mirrors with Isilon secret sauce. I wonder if turning
>off the secret sauce (kernel bits, middleware) would allow the test to
>pass without issue.
>
>Thanks!
>-Garrett
>