►
Description
SIG Cluster Lifecycle - Cluster API Office Hours - 20230315
B
All
right
now,
I
can
read
off
the
the
notes.
Thank
you
everyone.
This
is
the
cluster
API
meeting
a
subgroup
of
Sig
cluster
lifecycle
meet.
This
meeting
is
recorded,
remember
to
raise
your
hands,
since
this
is
the
element
of
the
cluster
life
cycle
cluster
life
cycle.
Please
follow
the
cncf
community
guidelines
essentially
sums
up
to
be
nice
to
each
other,
raise
your
hands
when
you
have
questions
and
am
I
missing
anything
I'm
skipping.
This
really
quick,
I,
think
that
covers
everything
with
those
all
right.
B
A
B
D
I
can
go.
My
name
is
Ogden
Mills
and
I'm
on
the
data
collection
team
for
Adobe
analytics,
and
we
we
do
some
of
our
own
cluster
Management
in
China,
since
this
isn't
over
there.
So
I'm
trying
to
keep
track
of
what
ethos
core
is
doing
and
how
stuff
is
going.
So
we
can
know
how
we
can
leverage
it
for
our
China
clusters.
B
A
Thank
you
Mike.
So
first
one
is
about
I'm,
proposing
a
small
amendment
to
our
back
support
policy.
So
the
reason
why
I'm
proposing
this
is
that,
while
now
the
project
is
settling
down
to
a
four
months,
Cadence,
which
is
great
for
the
project
overall,
but
it
can
somehow
sometimes
slow
down
the
work
on
experimental
feature.
So
following
something
similar
that
has
been
that
it
exists
in
the
capacity
policies
I'm
proposing
basically
to
a
low
to
back
part,
experiment,
enhancement,
enhancement
or
addition
to
experimental
feature,
but
with
some
guardlay
guard
rails.
A
So
guardrail
number
one
is
that
this
should
not
become
an
habit
because
the
recommended
way
to
deploy
enhancement
or
addition
to
experimental
features
to
grab
the
group
them
together
and
into
the
next
minor
release.
The
second
Gallery
is
that
we
allow
this
kind
of
backward
only
to
the
latest
stable
branch
in
in
this
moment.
It
is
one
to
three
and
the
second
and
the
last
guardrail,
and
the
most
important
is
that
that
top
priority
for
the
reviewer
will
be
to
keep
the
release
Branch
stable,
so
reviewer.
A
We
we
will
have
to
make
sure
that
the
the
back
part
is
well
scoped.
It
is
not
breaking
and
also
it
does
not
touch
part
of
the
code
base
which
are
used
by
stable
feature
and
eventually
additional
garbage
depending
on
the
type
of
PR,
and
so
it
is
possible.
It
is
a
back
door
in
case
we
forgot
to
to
add
something
or
people
ask
something,
and
we
do
not
want
to
wait
for
months,
but
it
should
not
become
diabet.
It
is
the
main
call
out
for
it.
A
B
B
E
Yeah
I
think
I'm
also
plus
one
to
the
idea
I
think
it's
needed
for
you
know
being
able
to
make
progress
on
these
features.
Although
I
will
say
I
think
the
devil
is
in
the
details
and
we
should
you
know,
be
very
careful
of
how
we
word
this
because
I
think,
if
it's
too
subjective
it
can
lead
to
different
interpretations.
E
So,
for
example,
like
I
see
right
now,
it
says
small
enhancements
like
we
should
Define
what
is
small
like
what
is
considered
small
and
then
what
does
it
mean
not
make
it
a
habit?
So
is
this
like
maybe
like
we
can
have
rules
by,
for
example,
it
needs
to
be
approved
by
two
maintainers
in
order
to
Grant
an
exception,
so
that
can
be
more
like
by
default.
It's
not
authorized,
but
you
can
get
an
exception
for
it.
E
Otherwise,
I
could
see
this
sliding
and
then
I
think
the
hardest
part
will
be
to
make
sure
we
don't
touch
other
parts
of
the
code
base
because
very
often
PRS
that
are
targeting
mostly
experimental
files,
end
up
fixing
like
a
util
function
or
adding
a
new
util
function
or
changing
something
in
a
non-experimental
file.
So
there
will
be
something
to
watch
out
for,
but
I'll
comment
on
the
pr.
Thank
you
for
bridge
silver,
making
this.
B
E
Thanks
this
is
very
update,
except
now
the
release
is
actually
out
so
tab.
Z
1.8.0
has
been
released.
The
two
it
contains
a
lot
of
great
goodies
and
features,
but
two
main
things
I
want
to
call
out
is
manager.
Oh
sorry
manage
clusters,
so
AKs
is
now
graduated
out
of
experimental.
This
means
it's
no
longer
behind
a
feature
flag,
so
huge
thanks
to
everyone
who
helped
make
that
happen.
Second
thing
is:
we
are
now
defaulting
to
out
of
tree
cloud
provider,
first
non-managed
or
self-managed
clusters
and
not
AKs
clusters.
E
F
Yep,
it's
for
the
alternative
communication
patterns
speech
group,
so
we
met
earlier.
Just
two
of
us
I
mean
Florian
from
from
AWS.
We
decided
that
we're
probably
going
to
move
from
a
two
weeks
Cadence
to
a
monthly
Cadence,
just
because
progress
is
slow
in
all
honesty,
so
I
know
from
both
our
parts.
There
isn't
the
focus.
B
F
So
this
is
to
see,
if
it's
possible
well,
it
is
possible.
People
have
done
this
to
instead
of
having
the
Assumption,
where
the
management
cluster
has
direct
connectivity
to
all
child
clusters
that
something
like
a
a
tunnel
reverse
tunnel
is
used.
The
child,
cluster
or
infrastructure
connects
back
to
the
copy
management
cluster.
F
A
Yeah,
this
is
a
a
question
for
Richard,
so
recently,
I
was
looking
to
kcp
the
kcp
from
Reddit,
so
the
kcp.dev
project
and
I
saw
that
they
have
an
interesting
solution
for
solving
a
similar
problem,
which
is
the
Sinker,
which
is
basically
an
agent
that
runs
on
the
physical
cluster.
In
this
case
for
copy
it
will
be
in
the
workload
cluster
and
this
agent
is
responsible
to
think
data.
In
this
case
our
sources
back
to
a
central
control
plane
in
the
case
of
kcps
kcp
itself.
In
the
case
copy
could
be
the
management
cluster.
F
That's
really
interesting.
Another
discussion
that
came
up
today,
which
was
you
know,
Cappy
manager
of
managers,
essentially
which
the
sync
would
be
really
really
interesting
for
that,
but
yeah
not
as
part
of
this,
but
maybe
we
should
and
I
can
I
can
add
that
to
the
discussion
document
as
well,
because
we've
got
a
separate
dock
with
some
of
the
Alternatives
that
we
have
considered
so
I
will
definitely
add
it
to
that.
A
A
I,
don't
know
enough
about
its
implementation
detail,
but
I
think
that
the
pattern
that
they
are
implementing
makes
sense
so
yeah,
maybe
some
time
down
the
line
we
I
will
reach
out
to
you
and
I
don't
know,
do
a
deep
dive
or
figure
it
out.
If
there
is
something
that
we
can
leverage
our
own
or
some
idea
that
we
can
look
at.
B
E
Yeah
I
I,
just
I
put
this
in
chat,
but
just
in
case
people
didn't
see
it
for
the
alternative
communication.
Patterns
group
I've
seen
multiple
people
ask
what
it's
about
and
I
think
every
other
time.
I
forget
what
it's
about
so
my
view
worth
like
thinking
of
like
a
more
like
targeted
name,
I,
think
I,
really
like
Daniel's
suggestion
of
the
management
workflow
cluster
connectivity
patterns.