Hi,
On 5/25/21 4:07 PM, Yedidyah Bar David wrote:
Hi all,
As part of the discussion around [1], we talked about changing the way
we create and use the ansible inventory.
I am now in the middle of doing something about this. It turned our
far more complex and possibly less elegant than
expected/hoped/intended.
I'd like to get a high-level review for my WIP [2]. It's not tested
and not ready for review. But if people think this is going in the
correct direction, I'll continue. Otherwise, I'll give up - and then
we have to decide how to continue otherwise. Parts of this can most
likely be done more nicely, other parts should probably be completely
replaced/removed/redone.
While the 'Inventory' class is very nice,
propagating that into other parts
of codebase indeed turned out to be quite a pickle... I was thinking about
it and did an attempt of a different approach - simply converting the
inventory
provided by the backend to be a directory [3]. It also meets the design
goals -
the backend is unaware of the HE VM and we can extend the inventory
dynamically
by simply dropping additional 'he.yml' file there when we're ready.
Preliminary
tests shown that there's no breakage so we should be able to avoid
the "global inventory fix" and stay with what we have right now.
WDYT?
We should probably also make some hard
decisions, which I am frankly not sure we can make without having more
concrete ideas about what other backends we want and how they would
look like. So for now I ignored all this and simply did a POC.
I definitely want to
keep OST code backend-(or in fact lago-)independent
cause
lago is not maintained and its raison d'ĂȘtre is inexistent... I doubt
however we'll
find the time to actually move/implement a different backend to run OST.
I'd definitely like to see oVirt tested by oVirt some day (as that
should be doable),
but I guess that's up to the community.
Regards, Marcin
[3]
https://gerrit.ovirt.org/#/c/ovirt-system-tests/+/114948/