►
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
Hello,
everyone
today
is
thursday
june
9th,
and
this
is
the
tesseract
provider
azure
office
hours,
cluster
api
provider.
Azure
is
a
sub
project
of
six
class.
Your
lifecycle,
as
always
please
make
sure
that
you're
during
today's,
since
you
have
good
conduct
and
if
you'd
like
to
speak,
please
raise
your
hand
and
we'll
make
sure
you
get
a
chance
to
talk.
A
The
agenda
is
below
I'll
share
the
link
in
the
chat
and
then,
once
you
add
anything
before
it
in
order
to
have
edit
access,
you
need
to
join
the
successor,
lifecycle,
mailing
list
and,
let's
see,
let's
get
started
first
off.
Do
we
have
anyone
here,
who's
new
to
the
meeting
and
just
wants
to
say
hi
or
introduce
themselves
so
go
for
it.
B
Hey,
I
can't
recall
if
I
actually
have
introduced
myself
before,
but
I'm
josh
farrow
with
vmware
I'm
working
on
stock
ashosh
on
cap
c,
so
great
to
meet
you.
A
Hi
josh
yeah
we've
interacted
on
slack.
A
All
right,
let's
get
started
on
the
open
discussion,
then,
let's
see
out
of
three
csi
migration
dash
go
for
it.
C
All
right,
so
I
was
working
on
this
pr
and
then
you
know
left
out
in
the
middle,
but
I
got
a
feedback
that
we
want
to
kind
of
install.
C
You
know
out
of
three
csi
drivers
for
all
the
flares
so
that
you
know
we
test,
maybe
windows,
node
or
other
linux
nodes,
so
that
makes
sense.
But
I
have
a
question
like
this
mean
that
for
the
base
flavor,
that
we
have,
we
just
put
all
the
flags
that
is
required
to
use
the
out
of
three
csa
migration.
C
D
C
E
C
E
I
haven't
taken
a
close
look
at
the
current
csi
implementation.
Cecile.
Are
there
existing
like
csi
template
injections
that
would
need
to
be
rolled
back
and
replaced
with
this
or
we're
just
basically
not
testing
this
at
all
right
now,
this
is
a
gap.
C
All
right,
I
did
it
man
I
did
it
manually,
like
you
know
whatever
I
wanted
to
do
so,
maybe
I'll
give
it
a
try
on
those
lines
and
see
if
it
works
so,
but
I
do
have
a
question
like
let's
say
we
wanted
to
do
the
way
I
was
doing
so.
Do
you
mind
if
I
start
my
screen
and
ask
a
couple
of
questions
on
that.
C
So
so
I
can
see
that
inside
test,
where
is
this
config,
so
this
also
typed
a
channel?
A
C
Yeah,
so
so
this
there
is
config
and
the
ci
like
jack
did
one
pr
for
external
cloud
provider.
You
know
so
I
wanted
to
understand.
Are
these
yamls
auto
generated
or
do
we
write
that
and
put
it
in
here
like
there
is
one
cluster
template,
brow
external
cloud,
and
then
we
have
one
more
like
cluster
template
external
cloud
provider.
So
how
are
these
two
different.
E
I
would
be
delighted
to
do
a
quick
ad
hoc
session
because
it's
sort
of
complicated,
but
I
could
I
could,
if
we
had
time
in
this,
is
this
an
hour
slot
in
this
meeting.
E
E
A
Cool
yeah
so
yeah,
let's
circle
back
to
this
one
but
yeah
thanks
for
looking
into
that,
you
had
another
topic
on
here
about
workflow
identity.
Do
you
want
to
talk
about
that
now.
C
Yeah,
so
I
was
just
exploring
it
the
last
day
and
you
know
just
wanted
to
you
know
understand:
is
there
any
so
this
workload,
identity
thing
that
is
there
in
this
github
repo?
I
see
like
there
is
no
sdk
for
that,
but
this
aad
identity
is
a
bit
different
like
it
works
on
that
you
know,
custom
resources
so
like
what's
the
way
to
like
I
know,
I
could
see
the
code
and
scope
where
we
are
using
in
the
scope
package,
where
we
are
using
aad,
pod
identity
and
all
those
things.
C
A
So
actually
danny
was
here
danny
and
I
actually
went
through
a
workflow
identity
tutorial
recently,
because
we
were
trying
to
figure
out
how
it
works.
So
we
could
integrate
it
with
the
easy
cappy
cli
extension
that
he's
working
on.
So
I
would
say,
that's
probably
the
best
way
to
get
started
to
understand
a
little
bit
what's
required.
They
have
really
good
documentation
if
you
go
in
there
with
the
identity,
page
there's
a
website,
and
then
I
think
I
recommend
going
through
the
quick
start
and
going
through
the
subsequence
manual.
A
So
you
can
understand
what
exactly
needs
to
be
set
up
and
then
we'll
want
cavsy
to
use
it's
going
to
be
very
different
from
how
it's
working
right
now
with
ad
card
identity.
A
Right
now,
what
we're
doing
is
we're
creating
this
azure
cluster
identity,
crd
and
then
from
that
we're
creating
the
azure
identity,
one
that
that
age
enemy
expects
it's
no
longer
using
its
own
crd,
as
my
understanding
and
work
with
it
and
it'll
be
announced
using
a
service
account
and
then
we're
gonna,
basically
add
a
label
to
the
cabzi
controller,
the
calves
manager
and
then
that's
going
to
tell
my
workload
identity
to
that.
It
can
use
that
identity.
A
So
I
think
we'll
need
to
make
some
changes
to
the
cavs
deployment
itself
to
add
that
label
and
then
we'll
probably
need
the
user
to
do
some
sort
of
like
setup
beforehand
to
set
up
that
identity,
because
it's
a
federated
identity.
I
don't
think
we're
going
to
be
able
to
have
all
that
built
into
kepsci,
but
that's
where
maybe
we
can
talk
about
how
to
like
make
that
easier
like
having
the
extension.
Do
that
for
you
or
something
like
that?
A
Sure
so
what
I
think
we
should
look
into
doing
maybe
is
trying
to
see
if
there's
any
way
of
doing
this
work
without
removing
a
deep
part
identity
right
away.
So
we
can
have
both
side
by
side
and
have
it
maybe
behind
the
feature
flag
at
first.
A
Just
because,
like
you
know,
it's
a
pretty
big
change,
so
we
want.
We
don't
want
this
to
be
breaking
behavior
for
users
and,
as
we
like
test
it
out,
it
would
be
nice
to
have
the
integration
with
that
making
it
required
right
away.
C
So
one
one
more
question
falling
to
that.
So
do
we
change
the
ux
for
this
like
right
now,
using
ad
pod
identity
is
like
you
know,
you
have
you
reference
that
in
azure
identity,
how
is
the?
How
is
the
user
gonna
say
this?
Like
I
mean,
do
we
want
to
say
that
if
you
want
to
use
workload,
let's
say
we
want
to
support
both
for
a
while,
like
a
d
identity
as
well
as
workload
identity,
and
how
do
we
tell
capsi
that
we
have
to
use
workload,
identity.
G
A
It
might
be
good
to
start
with
a
like
some,
maybe
research,
tlc
and
then
get
a
proposal
started
for
this.
This
pretty
significant
rework
of
hoffman's
work.
We
definitely
wanted
the
multi-tenancy
aspect
which
lets
you
have
like
a
cluster
in
certain
tenant
and
then
another
cluster
in
another
tenant
living
on
the
same
management
cluster.
E
A
So
what
we're
talking
about
right
now
and
correctly?
But
I
think
what
we're
talking
about
is
right
now
we're
using
a
d-pod
identity
to
give
an
identity
to
capsule
in
order
to
create
workflow
clusters
like
making
azure
channels
that
lets
it
create
and
we
can
sell
at
workforce
right.
A
I
id
has
been
deprecated
and
is
now
being
replaced
by
azure
workflow
identity,
which
is
another
project
which
also
does
something
similar,
but
pretty
differently.
It's
a
v2,
it's
a
lot
more
integrated
with
kubernetes,
and
so
we
want
to
switch
to
that
model
instead.
But
because
it's
like
the
base
identity,
that's
used
to
create
things.
It's
a
pretty
disruptive
change.
E
Right
yeah,
that
makes
sense.
So
it's
a
from
a
user
perspective,
it's
kind
of
a
black
box
yeah.
It
would
seem
that
the
most
sensible
high-level
approach
would
be
to
pick
a
release
cycle
to
sort
of
implement
this
under
the
hood
changing
so
say
from
v,
one
beta
one
to
one
beta
two
or
something
and
the
one
beta
two.
E
C
A
You,
and
if
you
you
know,
if
you
go
through
it
and
you
get
a
good
understanding
of
the
differences
and
all
that
I'm
sure
this
audience
would
you
know
really
be
interested
in
seeing
a
little
like
tldr.
You
know
these
are.
This
is
what
we're
going
towards,
and
this
is
what's
changing
and
that
would
be
very
valuable
if
you
want
to
do
that.
F
A
All
right,
let's
move
on
to
the
next
one,
then
let's
see
mama
touch
community
galleries.
H
All
right,
okay,
sorry,
hi
everyone,
so
I've
been
working
on
platform
support
for
captain
for
a
while
now
and
the
next
item
on
finishing
this
work
or
almost
the
last
one,
is
the
community
gallery
support?
So
then
we
can
use
the
community
images
which
platform
will
provide
to
for
the
ci
and
yeah
I
just
wanted
to
so
like,
and
we
in
this
vr
we
had
some
some
discussions
around
how
we
do
the
mapping
from
the
azure
compute
api
and
the
image
references
to
the
types
which
kevzie
uses
and
yeah.
H
I
just
wanted
to
rise.
Some
yeah
some
focus
on
this
vr
and
if
I
can
like
what
are
the
steps
next
steps
for
this
or
like
what's
missing
and
yeah,
if
maybe
some
reviewers
could
have
a
look
again
because
I
think
like
like,
I
think
we
reached
a
conclusion
on
all
the
conversations
there
and
I
think
it's
like
in
the
mergeable
state.
A
Yeah
thanks
I'll
take
another
look.
I
know
james
is
also
looking
at
it
and
if
anyone
else
has
to
come,
please
with
you.
This
is
a
pretty
big
change,
but
hopefully
one
we
can
leverage
in
the
future
for
all
our
images.
So
we
should
get
the
us
rights.
A
D
Yeah
thanks
for
sticking
with
that
one,
I
think
the
last
update
looks
good
to
me
I'll
review
it
one
more
time
verify,
but
I
think
yeah
good
nice
work
on
that.
Thank
you.
A
F
Yes,
hello,
I'm
sorry
are
you
able
to
hear
me
yeah.
F
Hey
hello,
everyone:
it
is
just
a
quick
announcement
that
we
have
released
the
latest
version
0.5
for
ac
copy
extension
so,
and
then
like.
This
is
like
the
latest
changes
that
we
have
is
support
for
custom
templates
to
bring
out
your
workload
clusters
and
other
other
live
features.
So
please
make
sure
to
try
it
out
and
give
us
any
feedback
that
you
might
have.
A
Awesome,
thank
you
so
much
and
yeah.
If
you
want
to
try
it
out
instructions,
I
guess
are
in
the
weekly
on
this
link
and
this
is
really
to
integrate
cluster
api
with
azure
still
life
with
the
folks
that
are
already
used
to
that
interface.
But
if
you're,
you
know
using
cluster
ctl
and
that's
working
fine
for
you
for
free
to
keep
doing
that.
I
Yeah
pardon
me,
I
just
want
to
say,
maybe
there's
nobody
on
this
call,
but
if
you
know
someone
who
kind
of
wants
to
get
into
cap
c,
this
is
a
real
easy
way
to
do
it.
You
know,
there's
several
there's
several
low-hanging
fruit
pr's.
I
forgot
the
label
that
we
needs.
We
have
several
needs,
help
vr's,
that
if
you
know
any
python,
they'd
be
really
easy
to
get
into.
So,
if
you
know
anybody
who's
interested
in
you
know
getting
into
cap
c.
I
Good
first
issue:
we
have
several
good
first
issue
issues
that
would
be
really
simple
to
knock
off.
A
Awesome
thanks
for
calling
that
out.
Maybe
we
should
link
to
it
in
the
casino
as
well.
At
some
point,
it's
trying.
I
I
mean
I
feel
like
when
we
get
to
more
of
a
release
candidate
when
we're
and
we're
fairly
close,
that
we
might
want
to
consider
vectoring
people
over
to
the
extension
from
the
capsid
docs
in
general.
Once
we
think
it's
solid.
I
A
A
So
we
are
at
1.4
all
right,
so
these
are
the
things
that
we
are
currently
tracking
as
the
not
the
things
we're
committed
to,
but
the
things
that
we
would
like
to
see
get
into
the
1.4
release,
which
we
are
planning
on
releasing
july
6.
So
in
a
bit
less
than
a
month.
This
is
like
the
estimated
date
based
on
last
release,
plus
two
months
that
we're
trying
to
adopt
as
well
students,
let's
see
so
flat
car
support
nature's
working
on
that.
A
So
we've
talked
about
community
galleries
being
the
last
step
for
that
of
tree
csi
drivers
and
cloud
providers.
So
I
think
shitasha
is
working
on
the
csi
driver,
part
of
it
jack.
Do
we
still
have
plans
to
move
the
other
templates
out
of
your
cloud
provider
like
what
should
be
our
timeline
for
that
or
not
not
right
away?
Yeah.
E
Good
question:
I
actually
suspect
it's
fairly
trivial
at
this
point
for
folks
who
don't
have
the
background
we
originally
cecil
and
I
originally
took
a
stab
at
this
sort
of
back
february
march
timeline
and
because
attitude
cloud
provider
was
something
that
was
being
delivered
inside
the
template
themselves
as
a
cluster
resource
set.
E
It
was
really
hard
to
invert
all
that
yaml
from
the
current
there's
one
out
of
three
template
and
then
every
other
test
is
entry.
So
the
high
level
goal
is
to
invert
that
and
have
the
default
template
out
of
tree
and
to
continue
to
sort
of
fall
back
with
a
single
cluster
template
or
a
single
cluster
test
scenario
that
validates
entry.
E
Now
that
we're
using
helm
to
install
out
of
tree
cloud
writer,
I
think
it
would
be
fairly
trivial
to
simply
remove
all
the
cluster
resource
set
definitions
from
everywhere
and
then
yeah
and
then
have
a
single
template
that
has
the
correct
configuration
and
then
have
helm
insult
out
a
tree
club
everywhere.
E
A
Okay,
so
we'll
keep
that
in
there
for
now.
I
think
so.
E
A
A
Okay,
I'll
do
that
afterwards,
services
async,
so
this
one
is
this-
is
the
epic
for
the
mega
issue
for
all
the
services.
So
this
one's
almost
done,
I
think
we
have
yeah.
We
made
some
really
good
progress
here.
We
have
tags
that
are
currently
in
progress
and
public
ips
and
a
few
others
remaining
around
skill
sets
afterwards.
A
Skill
sets
will
be
the
last
one,
so
I
think
yeah
async
publications
and
tags
are
here
and
they're
being
worked
on
and
we
have
pr's
open
early.
So
those
are
good.
I
don't
know
if
anyone
here
is
working
on
that
one
looks
like
it's
assigned
to
someone
node
image
upgrades.
Does
anyone
know
anything
about
that?
Looks
like
it's
rotten?
A
G
A
Thanks,
okay
and
then
we
have
the
last
one
which
is
documenting
our
release.
Statements
based
on
this
new
approach.
I
think
bridget
might
help
with
that
one
so,
but
I
think
anyone
else
wants
to
get
involved
in
help.
Please
feel
free
to
comment
on
the
issue.
A
Okay,
so
that's
all
we
have
right
now.
I
know
there
are
two
things
also
that
are
not
in
here.
That
probably
should
be
in
here
I'm
thinking
of
doing
workload,
identity,
community
galleries,
all
that
stuff.
So
maybe
we
should
just
add
those
whenever
whoever
is
working
on
the
issue-
and
you
just
notice
that
something's
not
in
here
please
edit
any
questions
for
the
monsoon
or
the
release
or
anything
in.
A
Between
right,
if
not,
I
think
I
will
stop
sharing
my
screen
and
then
I
will
hand
it
off
to
reject
if
you
want
to
go
there
test
templates.
Anyone
who
will
stay
for
this
feel
free
to
stay.
E
G
A
Shift
sorry
before
you
start
sharing
actually
just
a
little
announcement
psa
that
we
just
found
someone
added
at
the
bottom
that
I
almost
forgot.
I
just
wanted
to
say
thank
you
cheyenne
for
all
the
hard
work.
Cheyenne
is
stepping
down
as
maintainer
he's
taking
a
new
opportunity
outside
of
his
current
company,
so
he
won't
be
actively
maintaining
taxi
anymore,
and
then
I
also
want
to
say
welcome
jack
for
as
a
capsule
maintainer
jack.
Just
recently
got
added
from
your
video
to
maintainers.
A
We're
super
excited
to
have
him
joining
the
maintainer
group
and
yeah.
Thanks
for
all
the
hard
work,
jeff.
E
So
if
I
alt
tab,
safari,
vs
code
and
terminal
we've
got
it
all
cool
all
right,
so
I'll
start
with
vs
code,
so
yeah
ashtosh.
You
asked
about
how
all
of
the
there's
a
lot
of
templates
and
a
lot
of
like
different
types,
the
templates
under
subdirectories.
How
do
they
all
relate
to
one
another,
like
I
think,
specifically,
you
were
observing
that
we've
got
a
similar.
E
So
if
we,
if
we
think
about
external
cloud
provider,
for
example,
here's
one
under
test
ci
and
then
here
is
one
under
simply
under
templates.
And
how
does
this
all
work?
E
So
this
I
think
the
the
easiest
way
to
sort
of
introduce
what's
going
on
here
is
to
think
of
everything
under
templates
as
being
organized
according
to
like
a
sort
of
I
don't
know,
front
end
view
layer,
so
go
ahead
and
just
start.
E
Okay,
I
think
I
see
what's
going
on,
I
must
took
a
file
name
that
started
with
clustering
provider
azure
with
the
revo,
okay,
okay.
How
is
this
for
capsi
file
structure,
yep,
okay,
great
cool,
thank
you
cecile.
So
the
way
that
I
think
about
it-
and
this
is
maybe
my
bias-
because
I
I
have
done
some
front-end
development
in
the
past-
is
that
these
are
the.
There
are
three
sort
of
rendered
templates.
E
One
is
under
templates
and
it's
these
directories,
these
files
in
this
flat
structure
directly
into
the
templates
directory.
So
all
of
these
yaml
files
that
I'm
selecting
here
and
all
of
these
under
test
ci
and
I
believe
the
same
or
there
might
there
might
be
more
that
I'm
not
even
aware
of,
but
the
these
are
rendered
templates
that
derive
from
a
set
of
partials
for
lack
of
better
term
that
that
we
organize
and
decompose
and
then
put
back
together
at
the
end
using
a
script
called
gen
flavors.
E
E
I'm
not
sure
if
there's
like
a
canonical
project
that
it
was
built
for
whoops
in
the
wrong
directory,
but
it
allows
you
to
decompose
all
these
files
in
ways
like
this,
where
we
have
so
we've
got
like
sort
of
high
level
categories,
I'm
going
to
move
my
chat
window
over,
so
I
can
notice
it
so
much
added
just
in
case
someone's
asking
questions.
E
Customize
is
part
of
sig
cli,
okay
cool,
so
we've
got
these
high
little
level
categories
like
add-ons
and-
and
these
are
all
essentially
like
sections
of
what
we
want
to
inject
into
an
overall
template.
So
if
it's
not
clear
by
default,
we
are
using
in
our
tests
a
single
yaml
file,
which
is
basically
is
all
inclusive
of
everything
in
the
cluster.
E
And
so,
if
you
look
at
one
of
these
here's,
the
ad
template,
you
can
see
a
cluster
azure
cluster
cubanium
control
plane
so
on
no!
No.
This
is
not
the
only
way
that
you
can
create
a
capacity
cluster.
You
could
you
could
organize
this
with
one
yaml
file
per
resource
and
you
could
set
up
your
test
implementation
to
be
able
to
just
do
the
equivalent
cube
ctl
apply
through
an
enumerated
set
of
those,
but,
as
you
can
imagine,
that's
a
little
bit
more
complicated.
E
So
the
way
that
we
have
organized
this
is
to
move
that,
like
sort
of
per
resource
cluster
definition
burden
into
our
the
way
that
we
manage
these
flavors.
So
we've
got
like
I'm
gonna
demonstrate
something
I've
been
looking
at
recently,
so
under
add-ons
calico.
E
I
don't
want
to
go
too
deep
into
customized,
but
you
can
sort
of
see
here
that
we
have.
Is
this
a
single
resource?
No
there's
a
few
resources
here,
but
these
are.
These
are
calico
specific
resources,
and
so,
if
I
this
is
going
to
be
applied
to
basically
every
template
right
now.
Oh
I'm
in
the
wrong
branch.
Look
what
I
mean!
E
Okay!
So
this
is
going
to
basically
be
applied
to
every
every
template.
So
if
I
go
to
the
default
template,
I'm
not
seeing
it
there
where's
the
calculator.
Oh
okay,
it's
a
little!
I've
got
my
back
myself
into
it
more,
so
there
is
a
little
bit
more
that
we
do
with
cluster
resource
set.
So
I'm
going
to
demonstrate
something
else,
because
I
don't
want
to
confuse
folks,
let's
see
here,
I'm
going
to
skip
over
add-ons
all
right,
so
I'm
gonna
look
at
dual
stack.
So
in
dual
stack
we
have
some.
E
We
have
this
ability
to
basically
do
strategic
merges
of
various
resource
types.
So
here
we
are.
This
is
by
convention,
if
you're
not
familiar
with
this
you'll
just
have
to
trust
me,
but
this
is
typically
the
name
that
we
give
to
our
first
machine
deployment.
We
give
it
an
index
zero.
So
what
we're
doing
here
in
dual
stack
is
we're
essentially
defining
something
that
has
a
set
of
differences
and
then
customize
will
then
with
declarative
config.
E
E
Basically
overlaid
on
top
of
the
existing
say
resource
definition,
so
we've
got
a
cluster
definition
here,
an
azure
cluster
definition,
and
this
allows
us
to
override
very
specific
sections.
So
in
the
dual
stack
situation,
this
is
really
the
the
the
stuff
that
we
care
about,
and
so
we
want
to
essentially
take
our
our
default
cluster
spec,
which
maybe
I
can
find
easily.
E
Yeah
here
we
go
so
let's
say
this
is
this:
is
a
kind
of
generic
foundation
for
how
we
define
a
cluster
and
then
in
the
dual
stack.
E
Where
did
I
find
that
here
we
go
we're
essentially
going
to
overlay
this
on
top
of
this,
and
what
it'll
do
is
it'll,
replace
this
base,
cidr
block
definition
with
the
dual
stack
specific
one,
and
then,
if
we
look
at
the
outcome
of
all
that
kind
cluster,
we
will
see
that
the
rendered
dual
stack
template
is.
E
Has
the
cr
block
that's
as
much
detail
as
I
want
to
go
into
for
an
overview,
but
but
the
key
sort
of
high-level
concept
is
that
we
organize
all
of
these
sort
of
sub-components
of
each
template
like
this
into
the
sort
of
functional
like
into
these
functional
organizational
directories,
and
then
we
use
customize
to
put
them
back
together,
and
then
we
do
all
of
all
the
actual
rendering.
With
this
script.
E
C
E
That's
right
so
that,
from
from
a
maintenance
standpoint,
the
first
thing
that
you
have
to
internalize
is
that
when
you
want
to
move,
if
you
want
to
move
this
forward,
if
you
want
a
slash,
24z
16,
you
can't
just
do
it
here.
So
if
I
I'll
do
that
and
then
this
is
actually
going
to
take
a
long
time,
so
you
can
see
up
here
that
my
diff
suggests
that
I've
changed
the
yaml
from
16
to
slash
24.
I'll,
let
that
run
in
the
background.
So
what
you
actually
want
to
do.
E
Is
you
there's
a
little
bit
of
detective
work
that
goes
into
it,
but
hopefully
these
have
been
organized
in
a
sane
way,
where
the
patterns
sort
of
become
clear
fairly
quickly.
So
you
can
see
under
here
we
have
this
flavors
directory,
and
so
this
is
by
convention.
I
think
we
inherited
this
from
from
cappy
by
a
cluster
ctl.
E
You
can
sort
of
reverse
engineer
these
rendered
templates,
so
you
see
all
of
them
start
with
a
substring
prefix
cluster
dash
template.
I
think
that
fulfills
a
cluster
ctl
sort
of
standard.
E
So
that
way,
you
can
do
a
cluster
ctl
gesture
and
just
refer
to
the
flavor
and
it'll
look
for
a
file
called
cluster
dash
template
dash
with
the
flavor
name
so
under
flavors
for
aed.
There's
an
underplayed
there's
an
aed
directory.
So
we
can
assume
that
that's
where
we
would
actually
maintain
this.
So
the
the
kind
of
basic
initialization
file
is
customization.yaml,
and
so
this
is.
This
will
tell
us
where
all
of
the
the
partials
are
derived
and
how
they're
composed
into
a
rendered
outcome.
E
So
you
can
see
that
this
is
relying
on
that
that
base
set
of
of
definitions
and
then
we're
adding
to
it
a
machine
deployment
in
the
in
the
current
working
directory.
You
can
see
that
here
and
we
are
then
telling
it
to
take
this
yammel
partial
and
this
yama
partial
and
strategically
merge
it
on
top
of
essentially
the
outcome.
E
So
the
sort
of
hierarchy
is
like
this
is
your
sort
of
foundational
set
of
things
that
need,
and
there
are
some
assumptions
that
you
need
to
organize
in
some
way
where
there
aren't
collisions
that
are
going
to
produce
unpredictable
results
and
then,
if
we
go
patches
cuban
control
plane,
this
is
a
bunch
of
stuff,
that's
going
to
be
overlaid
on
top
of
and
it's
by
kind
on
top
of
the
the
any
existing
kind
cubanium
control
plane
with
this
name
that
it's
going
to
find
yeah
anyway.
E
Let's
go
look
back
here,
so
this
is
finished.
So
now,
if
I
go,
do
it
get
diff
so
now
we
have
this
this
sort
of
classic
scenario
where
you
know
this
is
a
common
thing
that
you'll
see
across
different
projects
where
you
think
you're
updating
the
thing,
but
actually
ci,
then
overwrites
it
because
the
thing
you
updated
is
generated
code.
That's
not
meant
to
be
modified
directly.
So
if
we
wanted
to
do
this
correctly,
let's
do
a
little
detective
work
here.
So
we
are
updating
the
crdr
block.
E
So
if
we
go
to
I'm
going
to
go
to
base
first
of
all
and
so
dot
slash
base
cluster
template
cool,
so
I'm
going
to
change
this
to
a
24,
and
so
at
this
point,
if
I
do
a
diff,
I
see
that
the
yellow
under
that
flavor
base
slash
cluster
temple.yaml
has
been
updated
and
I
will
now
do
a
make
generate
flavors.
E
So
this
should
hopefully
give
you
a
taste
of
the
complexity
here,
because
this
is
a
super,
simple
change
and
there's
all
kinds
of
things,
and
even
even
this
aad
definition
is
relatively
simple.
If
you
do
a
little
investigation,
you'll
see
some
that
are
super
complex,
so
still
you're,
saying
something:
no.
Okay,
sorry.
D
E
Oh
good
call:
well
I'm
actually
I'm
glad
that
that
that
I
accidentally
stumbled
into
that,
because
that's
a
really
important
thing
to
demonstrate.
So
this
is
one
of
the.
So
this
is
a
challenge
for
for
work
streams
like
I'm
working
on
where
say
I
wanna,
I
wanna
change
something
that
exists
everywhere:
fifteen
fifty
percent
of
the
templates,
so
that
requires
a
kind
of
like
a
more
abstract
refactoring
of
the
way
all
this
has
been
organized.
E
E
So
does
it
do
do
folks
understand
why
this
had.
I
was
just
like
kind
of
steamrolling
through
this
and
why
this
had
way
larger
side
effects
than
I
was
intending.
E
E
Okay,
so
what
I
would
want
to
do
again,
where
are
we
in
this
we're
on
the
cluster?
Okay
cool?
So
what
do
we
have
here?
Machine
deployment?
Where
would
you
put
this
in
patch?
It
would
you
put
it?
E
E
E
There
we
go
okay,
so
I'm
going
to
include
the
minimum
amount
of
information
needed
to
make
this
change,
which
means
I'm
going
to
whoops,
not
that
I'm
going
to
get
rid
of
all
this,
because
I
don't
want
to
change
any
of
this,
and
I
don't
want
to
change
this.
I
is
this:
the
minimum
amount
of
information
to
do
as
successful
strategic
workers.
Do
you
think
cecil.
E
A
E
So
I've
been
leaning
into
the
this
is
super
complicated
part
of
the
equation.
It's
also
really
powerful.
So
the
reason
we
there
are
good
reasons
why
the
you
know
sig
cli,
created
customized,
and
there
are
good
reasons
why
we
organize
our
template.
This
way,
if
you
do
this
same
and
and
in
a
well-earned
way,
then
this
is
a
great
great
way
to
organize
all
these
different
templates
and
make
maintenance
easier
going
forward.
But
it's
just
something
we
need
to
be
aware
of.
E
It
also
is
very
easy
to
make
this
too
complicated
all
right.
There
we
go
so
we've
got
this
net
new
file
in
our
git
diff
and
we
have
overridden
the
base
template
under
template
cluster
80.
yaml
with
the
new
value,
so
that
is
this
is
the
sort
of
the
muscle
that
we
exert
over
and
over
again
as
we
modify
these
templates.
C
Yeah
so
so
my
question
was
like:
if
you
go
to
the
template
static
today
yeah,
so
it
was
mode
coming
from
the
ci
perspective.
Let's
say
I
want
to
write
an
e2e
yeah
like
you
did,
for
you
know
external
cloud
provider,
and
if
you
go
to
the
test
directory
right
inside
the
test
in
the
ci,
we
again
have
you
know
the
ci
version
and
the
you
know
pro
external
cloud
provider,
ci
version,
the
directory.
E
Yep,
okay,
great
question,
so
this
is
a
little
opaque,
so
the
what
I've
been
what
I've
been
modifying
are
what
I-
and
hopefully
this
is
documented
in
a
sane
way.
What
I've
been
modifying
are
the
set
of
templates
that
we
provide
as
a
reference
for
users
to
introspect
and
hopefully
reuse
in
useful
ways
on
their
own
clusters.
So
these
templates
are
not
actually
input
for
any
of
our
tests,
so
this
change
that
I've
made
here
isn't
actually
going
to
affect
any
of
the
tests.
E
So
if
the
goal
one
would
do
this,
if
there
were
a
good
reason
to
update
the
reference
templates,
but
by
convention
we
there's
not
quite
a
one-to-one
mapping
because,
as
you
can
see
under
test,
so
so
really
quickly
under
test
ci
are
the
the
set
of
templates
that
we
actually
test
using
our
end
and
test
runner
is.
Is
that
a
correct
statement
to
seal,
or
is
that,
like
only
partially
correct?
I.
A
Know
that's
correct,
though
the
flavors
are
the
one
we
publish
as
release
artifacts
the
proud
ci
ones
are
the
ones
that
we
use
in
proud
to
test
them.
Usually
we
want
the
ci
templates
to
be
based
on
the
templates.
We
publish
with
just
a
few
additions
around
tags
that
allow
us
to
do
automated
cleanup.
A
E
Yeah,
I
would
I
would
echo
that
that
the
the
aspiration
would
be
for
for
the
difference
between
all
these
templates
in
test,
ci
and
and
under
just
templates
is
only
the
set
of
changes
that
are
needed
to
to
sort
of
inform
end-to-end
test
functionality,
but
that
functionally
they're
equivalent.
So
that
way
that
all
of
these
templates
that
we're
sharing,
as
as
references
for
our
user
community,
are
actually
tested,
but
we're
not
quite
the
aspirations
that
quite
realize
like,
for
example,
this
aed
template
that
we
have
a
reference
for.
We
don't
actually
test.
E
So
we
could
have
a
discussion
about
what
we
would
want
to
do
about
that
like
should
we
clean
up
that
reference
template
because
we
actually
don't
know
that
it
works,
or
should
we
add
an
additional
test
scenario
here,
so
that
we
can
make
sure
that
we're
backing
all
of
our
reference
templates
with
known
working
test
signal?
Does
that
make
sense.
C
Yeah
so
again
like
there
are
two
different
external
cloud
provider:
yammers
like
one
is
provider,
hyphen
ci
and
this
one
is
top
channel,
so
why
this
two
things
exist
there
like.
E
Yeah
great
question:
let's:
let's
quickly,
I
think
the
easiest
way
to
sort
of
talk
about
that
would
be
to
diss
them.
So
let
me
let
me
dip
the
two
hope.
Let
me
hopefully
there
I
think
they
probably
derive
yeah.
They
I
think
they
derive
from
common
foundation.
So
the
diff
should
sort
of
tell
the
story:
okay,
cool.
So
on
the
left
we
have
the
external
cloud
provider
and
on
the
right
we
have
external
cloud
providers,
ci
version.
E
I
see
okay
cool
so
the,
but
the
key,
the
key
difference
is
going
to
be
in
this
big
blob.
That's
on
the
right
here
so
dash
ci
version
is
our
that's
the
way
that
we've
named
a
few
of
our
templates
to
indicate
that
we
are
going
to
build.
I
think
we're
going
to
build
we're
using
these
templates
to
scaffold
clusters
that
are
using
arbitrarily
build
bits
built
bits
of
kubernetes.
Is
that
right?
E
So
all
of
this
foo
inside
this,
this
green
section
on
the
right
column
is
foo
that
we
have
developed
to
ensure
that
cube.
Adm
has
knowledge
of
how
to
sort
of
like
rationalize
the
kubernetes
version
aspect
of
these
arbitrarily
built
bits.
So
there's
a
bunch
of
foo
in
there.
I
said
this.
This
is
actually
the
command
that
runs
that
foo.
So
this
is
like
a
essentially
a
payload
that
we're
delivering
by
a
cloud
init,
and
this
is
where
we're
telling
cube
adm
to
execute
the
that
as
a
shell
script.
E
So
to
answer
your
question,
the
difference
between
these
two
is
that
the
the
sort
of
vanilla
external
cloud
provider
template
is
a
template
that
we
maintain
that
creates
a
cluster
that
can
accommodate
out
a
tree
cloud
provider,
so
I
say,
can't
accommodate
because
it
doesn't
actually
install
the
outreach
cloud
writer.
We
we
now
use
helm
to
do
that.
So
the
reason
we
have
a
separate
template
for
that
is
because
we
don't
want
to
well
there's.
There
are
a
few
configuration
changes
that
essentially
tell
like.
E
If
I
don't
know
here,
we
go
like
this
right
here,
so
we
need
a
separate
template
to
declare
these
configurations
to
the
control,
plane,
components
and
then
the
ci
version.
Suffixed
template
is
one
that
has
these
external
cloud
provider
declarations
here,
plus
it
has
all
of
the
known
working
the
way
that
we
have
integrated
all
of
our
tests
to
scaffold
clusters
that
can
run
arbitrary
bits
of
kubernetes.
E
So
from
a
from
a
dev
point
of
view,
what
this
one
on
the
right
fulfills
is.
It
helps
the
folks
maintaining
the
cloud
provider
azure
bits
as
they
move
forward
to
to
test
their
code
against
head
actually
to
test
their
code
against
kubernetes
at
head.
So
as
kubernetes
moves
forward,
we
can
wire
up
end-to-end
solutions
that
build
arbitrary
bits
of
kubernetes
and
then
use
well-known
environment
variables
to
refer
to
those
urls,
and
then
this
foo
will
tie
it
all
together.
Does
that
make
sense.
C
E
Okay,
let
me
close
this,
I
mean,
I
think
the
answer
to
your
question
is
definitely
yes,
because
that's
just
how
we
put
together
all
these
templates.
So
if
we
go
go
ahead,
okay,
cool!
So
if
we
go
under
test
ci,
you
can
see.
We've
got
two
of
these,
so
we
can
actually
dip
these
again
somewhere.
So
here's
my
vanilla
one
for
compare
selected,
so
the
one
on
the
right
wins
the
battle
so
much
simpler.
Oh
no,
maybe
not
reading
the
diff
right
yeah!
This
one
is
super
simple.
E
So
let's
jump
right!
This
one
is
so
simple.
Maybe
this
was
developed
like
after,
but
yeah
these
are
completely
different,
so
actually
to
go
into
detail
about
them
would
take
some
time
so
it
looks
like
maybe
this
is
offloading
some
of
the
complexity
to.
E
Here
so
this
is
probably
okay,
so
this
is
yeah,
here's
the
complexity,
so
the
reason
I
followed,
that
is,
is
that
this
is
a
way
to
sort
of
easily
fork
and
just
slightly
mutate,
an
existing
template.
So
it's
referring
to
this.
So
if
I
go
one
directory
down
and
then
under
prow
ci
version,
if
you
give
it
a
directory,
then
it's
essentially
going
to
render
the
outcome
of
this.
E
So
this
has
all
this
crazy
complexity
and
then,
in
addition
to
that,
we
have
an
external
cloud
provider,
yaml
spec,
that
we
are
doing
a
strategic
merge
on
top
of
what's
that,
and
so,
if
we
go
three
directories
down
to
flavors.
E
External
cloud
provider
patches,
so
this
is
the
interesting
sort
of
reusable
stuff
that
configures
the
control
plane.
So
this
is
where
we,
this
is
the
kind
of
old
way
of
doing
things
that
uses
the
cluster
resource
set.
We've
still
got
that
there,
so
that's
not
super
meaningful
that
the
really
meaningful
stuff
is
under
here.
So
this
cube
adm
control,
plane
configuration
needs
to
be
different
for
out
of
tree
clock
provider.
So
you
know
specifically,
we
need
this.
E
C
Yeah
so
let's
say,
for
example,
if
somebody
has
to
write
a
new
e2a
test
that
want
to
use
a
different
flavor.
That
does
not
exist,
so
you
go
to
the
ci
directory
and
you
know
put
your
own.
You
know
custom
mammals
that
would
actually
use
the
base
bra
or
the
ci
version
and
would
generate
these
templates
and
then
finally,
it
can
be
used
in
really
to
test.
E
That's
right
I
mean
I
could
just
for
fun.
We
can
actually
do
that
really
quickly.
So
I
could
this
since
this
one
is
sort
of
easy,
I'm
going
to
just
fork
this
one
and
I'll
call
this.
E
Super
cool
cloud
provider,
ci
version
and
all
I
need
to
do
to
do
this-
is
I'm
going
to
create
a
new
patch
just
under
this
directory
patches,
and
I'm
going
to
call
it
super
cool,
actually
I'll,
just
call
it
a
cube.
Adm.
E
E
That
and
if.
E
I'm
not
sure
that's
all
I
need
to
do.
I
think
so.
I
can't
remember
if
there's
any
extra
accounting
that
I
need
to
do
for
this
new
net
new
template,
but
let's
just
see
what
happens.