On 06/13/2017 03:47 PM, Peter Robinson wrote:
For actual artifacts such as cloud/disk/installer images I agree but
at least pushing out individual packages so people can do "dnf
upgrade" picks up issues such as dependency issues that also kill the
compose and allows people to still test explicit parts and have the
composite parts of a "rolling release" and get things fixed.... I feel
that's better than dragging everything to a blinding halt like we have
for the last 13... or is it 14 days?
Perhaps so. Note that I personally have been updating from the koji
repo, but I know that has some limitations (no multiarch, etc).
I can't help but feel this is like British politics is ATM where
people are claiming everything is "strong and stable" while the wheels
have fallen off and are rolling down the road. I don't think pushing
out the Everything repo stops the "kill off Alpha" process from
happening, in fact I believe it means it's more likely to happen
because if the last two weeks shows anything all that happens is that
if we wait for a "everything is perfect ship it" we never will and
because nothing is shipped nothing is tested so once we get to the
"computer thinks it's good" we can get to the actual testing and then
we get to "what the hell changed in the last two weeks broke X, Y and
Z, are they related or completely independent?" process.
I think the all or nothing actually makes it less likely for us to
ever ship anything! I don't think shipping the traditional
"Everything" repo breaks the "kill Alpha" proposal, I think we need
to
be pragmatic and realise that people actually consuming content helps
that.
Peter "yes I live in the as strong and stable as a house of cards
country so I can joke/comment on it" R
Well, I am not even an owner of that Change, so I will leave it to Adam
and Dennis to chime in on the no alpha part.
I do think now that we do have a compose that worked we are in a better
place and can now do what Colin was suggesting (If something fails now,
just untag the offending thing and file a bug on it and recompose).
There's definitely tons of room for improvement all around here including:
* Composes are really slow (likely related to storage slowness), if they
were faster or could fail faster we could untag/fix/iterate more. Right
now we are lucky to get 2 chances a day.
* There's plans pretty far along to 'pre test' base stuff with test
composes, that should help a lot. ie, new anaconda or lorax or whatever
arrives, a test compose is done and openqa tests it. If it fails we know
we can't use that in the main compose.
kevin