►
From YouTube: CAPZ Office Hours 05-01-2020
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
B
A
All
ready,
let's
get
started,
so
we
have
a
Google
Doc
that
basically
house
all
our
notes,
etc.
If
you
want
to
go
ahead
and
add
your
name
to
it,
go
for
it.
I'll
share
it
in
this
chat
right
now
and
if
you
have
any
discussion,
items
feel
free
to
add
it
anything
goes
in
this
forum
and
who
wants
to
be
a
note-taker
all
right.
Thank
you.
David.
A
A
C
We
just
merged
the
updates
to
the
dependency.
It
includes
your
3.4
as
well,
because
0.34
has
a
header
bug
for
cluster
cuddle
in
it.
So
we
held
that
back
and
waited
for
3.5.
It's
a
pretty
big
pet
release.
There's
a
lot
of
changes
in
there.
So
you
know
if
you
find
anything
in
the
like
next
week
or
so,
and
you
see
anything
we're
there
any.
D
C
Like
please
report
them
I'm
also
doing
for
it,
but
yeah
lots
of
changes
in
there.
A
Cool
and
then
another
fun
one
I'm
David
Justice
is
now
our
viewer
for
cap,
Z,
Congrats
yay
for
viewers,
and
then
we
have
a
net
and
then
I
think
we're
going
into
discussion
items
right
after
this,
starting
with
cecile
man
seeing
etsy
d-theta.
This
proposal.
C
Yeah
I'll,
sorry
also,
if
anyone
has
discussion
items,
please
add
them:
it's
not
too
late.
Okay,
so
I
just
wanted
to
share
this
proposal
that
I've
been
working
on.
So
basically
it's
a
way
to
be
able
to
mount
that's
D
on
a
data
disk
right
now
we
run
at
Sandy
on
the
US
disk
and
it's
causing
some
performance
issues
and
here's
the
cabinet
issue
that
yeah
I,
like
the
Cavs,
the
issue
that
started
this
proposal.
The
reason
I
made
it
a
proposal
is
because
I'm
gonna
need
changes
in
the
qadian
bootstrap
provider.
C
In
order
for
this
to
work,
because
we
need
to
change
cloud
in
it
to
be
able
to
mount
the
disk
and
what
I'm
proposing
right
now
is
a
general
way
to
add
disk
partitioning
and
fast
systems
to
cloud
in
it
using
qadian
config,
similar
to
how
we
do
user
than
NCP.
That
way,
it's
not
just
for
SSD,
it's
reusable
and
we
can
specify
the
way
we
learn
about.
That's
the
as
part
of
the
cabs
II
cluster
templates.
So
please
take
a
look.
Let
me
know
what
you
think.
C
E
C
E
A
F
I
think
this
is
gonna
become
a
thing
so
initially
the
conversations
around
multi-tenancy
wear,
hey
we're
just
going
to
have
multiple
instances
of
a
manager
and
a
individual
manager
will
have
one
set
of
credentials
and
to
do
multi-tenancy.
We
would
spin
up
new
instances
of
managers
each
one
having
its
own
tenant
and
a
tenant
was
associated
to
you,
a
set
of
credentials
and
those
are
loaded
out
via
the
environmental
configuration
for
the
manager
and
the
hour.
Edger
SDK
for
dough
picks
those
up
from
the
environmental
variables
there
now.
F
F
The
folks
over
in
Kappa
had
brought
up
that
they
were
looking
at
using
having
multiple
tenants
within
a
single
instance
and
I
actually
think
that
this
to
me,
this
makes
the
most
sense
if
we
can
build
up
a
good,
a
pay,
robust
security
model
around
this
to
me,
it
makes
sense
to
you
host
that
code.
Together
and
but
still
have
the
opportunity
to
possibly
have
multiple
instances,
because
I
mean
I'm
sure,
there's
gonna
be
situations
where
you
have
Coke
and
Pepsi
not
wanting
to
share
the
same
memory
space
or
something
like
that.
F
You're
good
you're
going
to
have
these
this
case.
So
the
work
that
was
done
to
break
out
the
webhooks
from
the
manager
from
the
reconciler
reconciler
was
is
still
good.
Work
still
necessary
work,
but
we
need
to
also
have
a
model
where
we
can
have
multiple
credentials,
possibly
credentials
from
secrets
or
perhaps
a
secret
store,
perhaps
and
then
a
instance
full
or
in
Azure
terms
a
managed
identity,
running
and
possibly
user
assigned
identities,
so
that
the
controller
can
have
multiple
identities
that
it's
running
under.
F
A
A
So
my
question
is:
are
we
trying
to
what
scenarios
with
multi-tenancy
are
we
train
enable?
Are
we
trying
to
enable
like
I'm,
building
out
platform,
so
other
folks
can
now
use
like
copy
securely
or
copies
securely,
or
is
it
for
like
folks
internally
developing
like
what
what's
the
model
like?
What's
the
use
case,
I
guess.
F
So
the
use
case-
I
think
is,
you
have
you
know
possibly
multiple
tenants
and
they
control
like
multiple
tenants
within
your
cluster.
So
maybe
a
single
tenant
is
confined
to
a
namespace
and
they're
going
to
have
all
their
infrastructure
live
in
that
namespace
and
maybe
that's
like
company
one
and
then
Company
2
is
going
to
be
separated
off
into
the
subroutine
space.
Now
you
don't
want
to
have
a
single
identity
that
has
access
to
you
both
of
those
companies
right
right
as
your
infrastructure
out
for
both
of
those
companies
like
one
master,
SuperDuper
identity.
A
Yeah
please
so,
for
example,
a
development
team
working
under
the
same
cluster
say
they're,
building
like
Starbucks
is
going
in
and
building
out
like
their
applications
on
top
of
a
cluster,
usually
there's
multiple
teams
working
in
the
same
clusters,
because
one
cluster
per
team.
It's
a
lot
of
money.
It's
a
lot
of
resources!
So
that's
why?
Finding
specifically
what
we're
doing
here,
I
guess
and
thanks
for
the
explanation,
David
it's
trying
to
figure
out.
A
Yeah
yeah,
that's
kind
of
the
high
level
of
it
there's
another
level
to
this
as
which
might
come
up
in
the
future,
and
what,
if,
like,
customers,
wanted
to
develop
hard
multi-tenancy
within
cab,
Z
itself,
like
through
hardware
through
the
computer,
because
they
were
offering
some
sort
of
platform
to
offer
like
compute
to
their
customers
and
then
they
would
have
to
isolate.
But
that's
a
different
discussion,
so
yeah.
F
Yeah
I
think
it
is
I'm
really
curious
to
hear
all
the
use
cases
from
Kappa
2
to
be
completely
Frank.
I
I
haven't
thought
through
all
two
use
cases
yet
and
I
I'm
really
curious
to
sit
in
in
here.
What
they're
thinking
about
I
know
this
also
will
come
down
to
possibly
on
behalf
of
identities
and
so
like
b-but,
the
user.
So
if
we
can
authenticate
somebody.
F
What
is
it?
Oh
I
do
see
the
boudin
I
see
open,
ID
connect.
If
we
can
authenticate
somebody
with
that
and
then
use
that
identity
in
a
ad,
then
you
know
provision
infrastructure
for
them
and
understand
what
roles
and
rights
they
have
with
that
measure.
I
think
we
can
start
to
extend
out
that
rights
boundary
and
understand
who's,
actually
operating
and
use
that
identity.
On
behalf
of
the
controller,
I
think
we
need
to
start
thinking
about
deeper,
deeper
identity
integration
into
the
controller
in
general
and
I.
A
C
C
F
I'm
thinking
as
your
tenant
and
I'm
thinking
the
way
that
the
way
that
the
way
that
the
churches
work
right
now
or
at
least
with
that
PR,
is
that
we're
still
looking
at
the
same
time
so
model
we
still
only
have
one
set
of
creds
and
that
set
of
creds
has
to
have
access
to
you.
Any
of
the
subscriptions
that
it's
going
to
talk
to
you
right,
yeah.
E
D
Focaccia
one
use
case
is,
if
you
have
a
resource
like
GPB
GPU,
which
might
have
lots
of
threads.
Are
we
GPUs,
then?
If
it
is
in
a
cluster,
somebody
might
want
to
use
it
in
the
data
plane.
So
that
is
one
use
case.
I
would
say,
because
it's
a
expensive
resource
and
needs
to
be
shared
by
multiple
tenants
in
a
given
cluster.
That
could
be
a
possible.
A
A
C
So
this
I
just
wanted
to
bring
up
the
question
it
might
be
easily
answered.
If
it's
not,
it
might
be
an
issue
or
proposal,
but
I
just
was
wondering
so
right
now.
What
we
do
is
when
we
reconcile
like
a
German
or
a
Jew
cluster,
and
that
needs
to
create
Azure
infrastructure,
for
example
like
creating
a
resource
group
or
creating
a
virtual
machine.
C
We
wait
for
the
future
completion,
so
we
send
the
request
SDK
and
then
we
say
wait
until
this
is
done,
and
so
the
we
don't
progress
further
until
that
resource
is
succeeded
and
it's
created
which
isn't
bad,
but
it
just
means
that
practically
if
you
are
checking
like
constantly
for
status
of
your
Azure
cluster
Azure
machine,
it
will
never
go
into
progress
or
progressing
state.
So,
for
example,
we
have
VM
state
for
machines,
and
the
VM
state
will
always
be
either
not
started
or
succeeded
or
failed.
C
It
will
never
show
you
that
it's
creating
or
that
it's
deleting
because
that
operation
it
never
like
loops
back
and
that's
not
I
was
just
wondering.
Is
that
something
that
you
know
we?
We
think
it's
okay,
like
that
or
the
other
thing
I
was
thinking
is
I
know.
David
wants
to
say
something
about
that,
because
I
I
was
actually
thinking
about
that.
H
Was
just
gonna
I'm,
just
gonna
say
it
because
I
was
kind
of
doing
that
anyway,
so
because
the
Ricans
are
always
going
to
keep
and
then
picking
up
the
different
status
if
there
are
different
statuses
or
should
be
doing
that.
So,
if
we,
if
we're
not
reading,
if
we're
not
waiting
for
the
future
to
come
back,
the
reconciler
only
check
the
status
and
then
have
the
different
pieces
has
its
route.
The
same
life
cycle
so
I
might
not
have
to
make
this
right.
E
F
Yes,
so
this
this
kind
of
stinks,
because
that
future
I
don't
think
the
future
is
actually
serializable
either,
so
that
future
that
we
get
back
in
fact,
is
a
pointer
to
a
other
like
some
sort
of
the
URI
and
a
way
to
look
into
the
body
of
that
URI
to
get
back
what
you
want
so
in
Azure
I'll
give
you
the
short
short
version
in
Azure.
When
you
create
an
instance
of
something
resource,
you
have
possibly
three
different
types
of
responses
that
you
could
get
back.
F
I'm
super
super
quick
ones
and
you
could
just
get
it
200
and
you
can
say
or
201
with
no
long-running
operation
and
it
comes
back,
has
succeeded
and
give
it
a
high-five
and
you'll
walk
away.
No,
no!
No!
But
no!
You
know
we
try.
You
know
find
out
when,
when
it's
done
other
times
to
get
it
to
to
and
that's
just
an
accepted
and
what
that
ends
up
giving
you
is
a
response
back.
Some
of
them
actually
have
a
pointer
to
an
operation
and
that
operation.
F
You
can
then
use
that
URI
to
get
operation
and
keep
pulling
it.
You
end
up
getting
the
status
of
that
operation
that
that
can
take
a
bit
and
then
the
other
one
is
actually
the
statuses
in
the
body
so
like
when
you
build
a
VM,
you
get
provisioning
stay
back
and
you
have
your
gonna
201
back
to
the
service,
but
the
provisional
state
and
the
body
is
the
thing
that
changes.
So
you
have
to
keep
asking
it
until
you
get
a
terminal
state.
F
F
It
is
so
like
Java
and
on
that,
but
that
state
machine
is
actually
serializable
and
you
can
get
it
back
out
for
with
reconciler
it'd,
be
really
really
cool
to
have
the
async
reconciliation
right,
because
if
we
could
do
something
like
that,
we
just
kick
it
off
and
we
say
later
down
the
line.
We're
gonna
keep
coming
back
and
checking
out
the
resource
state.
Now
we
we
have
a
little
project.
That's
going
on
that
I
think
we.
F
You
should
probably
progress
a
little
further
and
then
introduce
it
here
where
we
can
do
kind
of
that
that
basic
reconciliation-
and
that
seems
maybe
that
would
be
the
time
that
we
really
address
this
as
you're
saying
so.
Co
I
think
you're,
you're,
spot-on
and
recognizing
that
issue,
and
it's
it's
tough
for
a
consumer,
because
you
don't
have
a
good
way
to
do
it
right
now,
so
maybe
putt
for
just
a
little.
C
Also
I
just
want
to
add:
it
decides
they're
like
practical
effect
of
not
being
able
to
pull
status.
There's
also
the
effect
of
the
fact
that
we're
waiting
for
resources
and
cereal
as
I
see
realizing
the
deployment
of
different
resources
that
might
have
no
dependency
on
each
other
is
making
us.
You
know
waste
time
so,
for
example,
creating
the
I
don't
know.
What's
a
good
example,
creating
the
virtual
network
might
not
have
anything
to
do
with
creating
actually
virtual
network
and
is
connected
to
everything
but
anyways.
C
You
have
some
resources
that
are
not
connected
to
each
other
that
you
need
to
deploy
that
you
could
deploy
in
parallel,
but
by
blocking
the
deployment
of
the
second
one
on
the
finish
or
on
the
completion
of
the
first
one
we're.
Essentially,
we
could
reduce
the
time
it
takes
to
create
clusters.
That's
what
I'm
saying
by
doing
something
like
that
and
yeah
I'm
fine
with
punting.
For
now.
If
we
think
that
the
breaking
out
the
SDK
usage
will
be
a
natural
solution
that
will
come
soon,
just
some
thought.
I
had
that
I
thought
I'd,
chair.
D
D
What
they
do
is
they
call
ionic
ionic
is
the
bare
metal
operator
if
they
have
a
response
back
saying
whether
it's
complete
or
not
completed,
because
you
cannot
wait
indefinitely
for
IO
completions,
so
either
you
have
answer
within
30
seconds
or
whatever
30
milliseconds
whatever
you
time,
and
then
it
will
not
pick
up
that
device
or
that
node
or
that
specific
host,
whatever
you
call.
So
in
that
case
there
is
a
separate
CRD
for
remediation,
which
is
independent
of
so
this
happens
in
the
operator
framework.
D
So
I
am
not
sure
what
we
use
here,
because
I
have
not
looked
at
the
architecture,
fully
to
understand
how
we
are
doing
it.
So
what's
the
effect
of
what
we
call
the
timeouts,
should
it
not
proceed
after
timeout
and
if
it
is
failing,
somebody
is
some
other
see
already
we'll
handle
that
that's
the
question:
does
it
make
sense?
I,
don't
know
that
is
the
solution.
D
C
So
yeah,
so
essentially
our
works
in
cabs.
You
right
now
is
the
reconciler
will
say:
okay,
I
need
to
create
a
virtual
machine
and
then
that
create
virtual
machine
function
will
call
out
to
the
azure
s
and
K
forego,
which
will
send
the
request
to
create
the
machine
and
that
request
will
send
back
what
we
call
it
future
and
the
future
is
basically
something
that
tells
you
like.
C
David
was
describing,
you
know
when
is
done
and
what
we
do
is
we
wait
for
that
future
to
be
completed,
so
we
don't
keep
going
until
the
future
either
tells
us
like
this
machine
timed
out
or
it
succeeded,
or
it's
failed,
and
so
we
do
use
Rican
cells.
They
were
not
using
the
reconciler
part
of
like
constantly
check
back
on
the
VM
to
see.
If
it's
created
we're
just
going
through
it
once
and
either
it
fails,
or
it
succeeds.
C
D
D
D
A
C
A
C
A
C
A
C
A
H
A
Done
honestly,
that's
a
bit,
but
let's,
let's
talk
to
Steven
behind
the
scenes,
I
mean
just
double-check
and
then
we'll
go
ahead
and
change
the
name
to
just
done.
If
the
list
gets
too
big
to
manage,
I
mean
ideally
we're
tracking
all
of
our
done
things
in
milestones
anyway,
so
like
past
mile
since
we'll
be
we'll,
be
able
to
see
that
issues
and
tears
linked
to
it.
I'm
such
to
be
fine
but
yeah.
Anybody,
elderly.
A
Okay,
cool
we'll
talk
to
Steven,
really
quick
and
then
change
that
thanks
for
bringing
up
Cecile
alrighty.
So
let's
go
quickly
through
the
project
word.
Everything
should
be
pretty
up-to-date
but
there's
probably
new
cards
that
we
need
to
add
so
there's
three
here
we
can
kind
of
fear
where
they
go:
they're,
probably
just
going
to
end
up
in
the
backlog.
A
These
guys,
don't
think
I
don't
have
anything
attached
to
them.
Alright,
let's
go
into
the
back,
look
really
quick
and
make
sure
there's
nothing
that
should
be
in
to
do
I
guess
if
you
guys
see
something
that
should
be
moved.
Let
me
know,
but
if
there's
no
person
attached
to
it,
we
probably
can
keep
it
there.
Oh.
A
J
A
A
A
A
A
So
many
things
are
getting
picked
up:
okay,
no
Matt's
working
on
it!
I
guess
he
hasn't.
Has
you
put
up
a
PR
yet
nope?
Okay,
we'll
keep
it
in
to
do
that
until
we
get
up
here
on
that
all
righty,
and
then
this
guy
has
a
PR
League
to
it.
We
made
it's.
A
C
I
think
I'm
assigned
to
the
ones
I'm
reviewing,
but
I
don't
have
that
many
that
are
moving
right
now,
yeah
well,
yeah,
I,
think
all
of
these
are
actually
being
worked
on
right
now,
except
maybe
the
control
plane,
NSG
thing
which
has
been
standing
there
for
a
bit
and
I
commented
on
it
saying
maybe
we
should
close
it,
but
I'm,
not
a
response.
It's
okay.
H
A
H
B
A
Okay,
I
had
some
okay,
not
all
of
it.
I
heard
a
kiss
engine
and
I
heard
a
Don's.
A
B
B
A
Well,
happy
to
have
you
and
I'm
gonna
set
up
some
time
with
you
to
discuss
further
kind
of
what
you
guys
are
doing
with
a
kiss
engine
and
then
what
we
could
do.
The
cabs,
II
and
I
can
help
you
get
up
to
speed
to
just
from
just
because
our
arc
is
so
internal
and
then
we
can
figure
out
what
use
cases
we
want
to
bring
up
into
Katzie,
etc.
A
C
So,
there's
a
good
reason
for
that.
The
good
first
issues
are
actually
not
in
the
mouth
soon,
because
one
of
the
criteria
is
that
they're
not
critical
they're,
not
on
the
critical
path,
but
there
are
good
versus
issues
that
you
can
pick
up.
They're,
just
not
part
of
what
we
commits.
That's
part
of
the
work
so
yeah
feel
free
to
pick
those
up.
You
can
filter
by
label
yep.
A
F
Was
awesome,
work
they
ingested
awesome
work.
Thank
you
and
Richard
I'm,
not
sure
ground
right
now,
good
progress.
Almost
there
I
really
enthusiastic
about
all
of
the
awesome
work
that
is
going
on
across
the
project.
There's
a
wide
variety
of
people
contributing
and
that's
that's
inspiring.
Thank
you,
everyone
for
the
Freder
commitment
and
if
there's
anything,
we
can
do
to
help
support
Riya
seal.
Any
of
us
we're
always
were
always
there,
but
you
you
all
are
first.
A
Okay
cool
well.
On
that
note
thanks,
everyone
have
a
good,
Friday
and
yeah
we'll
see
thanks
Rio,
bye.