►
From YouTube: 2020-11-12 CAPZ Office Hours
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
All
right,
hello,
everyone
today
is
november
12
2020,
and
this
is
the
cluster
api
azure
office
hours.
As
always,
please
follow
the
cncf
code
of
conduct
and
be
nice
to
everyone.
If
you
have
anything
that
you'd
like
to
discuss,
feel
free
to
add
it
to
the
open
discussion
section
of
the
agenda
and
if
you
can,
please
add
your
name
to
the
attendant
list
at
the
top
of
the
document.
If
you
don't
have
this
document,
I
will
paste
the
link
in
the
chat
and.
A
A
Okay,
so,
let's
start
with
welcoming
anyone
who's
new
to
this
meeting.
Is
there
anyone
who
hasn't
been
to
this
meeting
before
and
wanna
wants
to
quickly
introduce
themselves
just
say
hello
and
tell
us
like
what
brought
you
to
cabsie
office
hours.
A
I
don't
see
any
hands
or
anyone
on
muting,
so
I'm
going
to
assume
there
isn't,
if
I
already
said,
add
your
name
to
the
attendance
list
and
we'll
do
the
millstone
project
board
review
at
the
end,
I
think
yeah.
A
So,
let's
start
with
the
other
discussion
topics.
Maybe,
and
then
we
can
come
back
to
the
milestone
david.
You
want
to
talk
to
us
about
aci.
B
Absolutely
so
thought's
been
bouncing
around
my
head
with
the
way
that
we
build
up
infrastructure.
We
use
virtual
machines
and
we
use
virtual
machine
scale
sets
each
one
both
both
use
virtual
machines
behind
the
scenes,
so
they
spin
up
a
virtual
machine
within
azure.
B
B
Azure
also
offers
a
kind
of
abstraction
for
running
containers
called
azure
container
instances
and
to
deploy
out
a
container
there.
It's
very
very
fast,
it's
like
super
fast,
and
it
builds
like
by
the
second
and
it's.
I
think
it
could
offer
us
a
really
interesting
opportunity
to
spin
up
clusters
quickly
scale
them
quickly.
B
B
It
just
basically
becomes
capti
except
running
in
container
instances,
and
I
mean
there's
still
going
to
be
like
some
weirdness
around,
like
we
gotta
wire
up
networking
and
make
sure
everything
works
in
an
azure-ish
way
for
like
load,
balancers
and
v-nets
and
stuff
like
that,
but
I
think
it
actually.
I
think
it
could
work.
So
I
was
just
curious
what
people
thought
about
it.
If
this
is
something
that
would
be
interesting
or
just
get
the
idea
out
of
the
public,
anybody
have
any
thoughts.
B
That's
a
really
great
question,
okay
and
and
every
time
that
this
has
been
brought
up
just
like
in
like
one-off
discussions
that
that's
brought
up
too
so
vk
ends
up
spinning
up
in
aci
instance
right,
but
you
have
to
have
the
case
somewhere
or
like
we
don't
have
to
introduce
vk
like
it
could
just
be
at
the
controller
level
that
we're
doing
this.
B
I
don't.
I
don't
think
we
need
vk
and
I'm
not
sure
what
vk
brings
to
us.
That
would
be
greater
than
kept
c
doing
it.
A
Interesting,
I
guess
I
my
understanding
was
that
you
have
the
control
plane
which
still
runs
on
real
vms,
and
then
you
have
your
networking
and
then
you
just
have
like
virtual
notes,
which
are
like
some
of
your
nodes,
but
all
of
them
not
all
of
them.
So
you
can
also
have
real
nodes
side
by
side
with
virtual
nodes.
A
So
it's
not
like
what,
if
you
use
vk,
so
it's
not
really
like
a
different
provider.
It's
just
like
a
type
of
node.
That's
in
your
cluster.
B
C
So
I
guess
I
see
virtual
keyword
is
kind
of
responsible
for
running
the
work,
like
the
pause,
the
workload,
whereas
what
you're
kind
of
suggesting
is
running
the
nodes
as
like
a
node
is
like
the
api
instance
is
a
node,
and
then
you
run
containers
inside
of
those
that
dock,
like
docker
and
docker
kind
of
thing.
B
C
Well,
I
do
have
a
friend
who
or
a
colleague
that
did
this
probably
a
year
and
a
half
ago
or
two
years
ago,
where
they
got
an
entire
kubernetes
cluster
running
fully
in
aci,
and
I
think
you
wrote
a
blog
post
on
it.
It
might
be
a
good
starting
point
to
even
understand
what
the
limitations
are.
B
Yeah
I
was
talk,
I
had
a
short
conversation
with
pm
over
there
and
they
had
they
had
brought
up
that
they
had
spun
up
clusters
before
with
aci
and
yeah.
I,
if,
if
you
could
point
out
where
that
is
that
that
post
would
love
to
read
through
it,
that
would
be
super
helpful.
A
B
I
I
think
it
depends
right.
So
if
you,
if
you're
running
nodes
like
all
the
time
like
if
you're
always
running
nodes,
you
probably
want
to
mix,
I
think
you
would
want
a
mix
of
like
here
is
my
base
set
of
my
reserved
instances
and
then
maybe
above
that,
here's
my
spike
and
when
I
have
traffic
spiking,
I
spike
with
aci,
and
then
I
come
back
down
to
like
my
reserved
set
almost
almost
like
auto
scaler
would
be
smart
enough
to
spin
up
hey.
B
I
need
to
respond
quickly
and
then
replace
with
like
more
permanent,
lower
cost.
B
I
don't
know
that
starts
to
get
really
complicated,
but
it
would
be
so
cool.
I
I'll
link
in
some
pricing
details.
A
Yeah,
do
you
want
to
create
an
issue
on
the
repost
that
we
have
that
idea
somewhere
and
then
we
can
think
about.
A
Snyder
cool
all
right:
let's
keep
going
hey
janet
by
the
way.
It's
really
good
to
see
you
here.
We
have
another
discussion
on
fleeky
tests.
Let's
do
it
either.
D
Well,
there
is
the
pr
about
fabrizio
and
cluster
api,
that
is
about
the
kcp
thing
that
we
suspect
might
help
our
kcp
flaky
test.
But
I
guess
once
we
have
once
we
have
that
they're
supposed
to
do
an
rc,
I
think
today
or
tomorrow
I
can't
remember,
and
then
we
can
get
that
it
just
got
this
pr
just
got
married
yesterday,
but
I
guess
we
can
wait
for
the
rc
and
and
try
to
test
with
that
and
see
how
that
helps
with
the
kcp
upgrade
test.
D
I
did
a
pr
to
test
out
the
removal
of
the
joint
retry
of
cuvidium.
It
seems
to
be
working.
Fine
except
the
same
clickiness
that's
been
happening
around,
I
think
from
most
of
the
cases
I've
seen
it
seems
to
be
in
the
ipv6
test.
D
A
Great
yeah
and
then
there's
also
been
some
more
flakiness.
I've
noticed
not
just
on
ipv6,
but
also
the
multiple
control
plane
cluster
on
the
pr
tests.
Since
the
new
kubernetes
version,
pr
merged.
A
Think
it
might
be
a
coincidence,
maybe
not
so
I
think
there
are
a
few
different
things
going
on
here,
so
it
would
be
really
great
if
we
can
get
a
repro
of
ipv6,
so
we
can
see
what's
going
on
there,
because
I
suspect
they
might
be
specific
to
ipv6.
If
it's
a
flake,
that's
happening
there
and
then
in
terms
of
the
one
that
merged
we
might
be
able
to
pull
it
in
already
or
we
can
just
wait
for
the
rc.
A
But
I
think
we
should
really
start
testing
cappy
master,
so
we
can
set
up
a
new
job
that
runs
the
cappy
end-to-end
test
with
the
like
latest
commit
in
capi.
How
does
that
sound
like
as
a
periodic
job.
D
A
Oh,
I
see
what
you're
saying
I
guess
we
could
do
something
where
you
like
check
out
the
repo
and
then
you
build
the
image
from
there
and
then
you
check
out
kevzie
and
use
the
image
yeah.
It
would
just
be
a
little
more
complex
of
a
job
definition
but
yeah
for
now
short
term.
We
should
use
the
rc
but
I'll
try
to
look
into
adding
a
new
job.
D
B
So
the
metrics
would
not
the
tracing.
Perhaps
would
the
thing
is,
I
don't
think
we're
deploying
out
the
jenkins
sidecar
when
we're
running
the
tests,
so
we
very
easily
could
deploy
out
that
jenkins,
sidecar
or
not
jenkins.
I'm
sorry,
jager.
B
The
other
jay,
so
the
jager
sidecar,
could
definitely
in
the
jaeger
operator
could
definitely
be
deployed
out
in
our
testing
environment.
B
We
would
then
have
to
call
out
and
get
get
the
traces
or
visualize
the
traces
right
or
or
we
could
put
the
traces
like
somewhere
else
like
have
a
another
cluster.
That's
running
that
we're
actually
exporting
traces
to
that's
kind
of
like
a
more
permanent
cluster.
I
don't
know
we.
We
need
to
talk
through
that.
I
think.
D
E
B
So
if
you're,
the
only
time
the
tracing
components
get
deployed
right
now
is
through
tilt,
but
it's
not
hard
to
do
it.
If
you
look,
there's
there's
just
some
telemetry
components
and
if
you
run
customize
on
that
directory,
it'll
produce
all
the
components
for
the
controller,
as
well
as
the
telemetry
components.
B
A
Okay,
the
other
issue
that
we
should
really
fix
that
could
help
us
with
investigating.
This
is
right
now,
if
you're
like
we're
getting
the
workload
cluster
logs,
but
whenever
one
of
these
clusters
fails
to
provision
it
usually
times
out,
and
so
it
doesn't
fetch
the
logs
at
the
end.
So
we
end
up
getting
logs
for
all
the
successful
clusters,
but
not
the
failed
one,
which
is
the
most
important
one
to
get
the
logs
from.
So
if
we
fix
this,
we
could
have
logs
in
in
two
week
runs
in
pro
for
the
failures.
D
A
All
it
anything
else
on
testing.
D
So
the
retry
one
I
retry
join
when
I
just
keep
like
running
the
test
a
bunch
of
times.
I
don't
seem
to
see
failures
related
to
that
retry.
It
seems
to
be
the
same
clickiness,
so
I
think
we
should
keep
running
it
a
few
more
times
and
keep
it
until
we
get
the
new
copy
one
and
test
with
that.
But
just
does
that
make
sense
to
everybody.
D
A
Okay,
before
we
move
on
to
nelson
planning,
does
anyone
have
any
topics
they
want
to
bring
up
any
questions?
A
A
Okay,
so
I'll
start
sharing
the
milestone.
Can
you
see
my
github
yeah.
A
All
right
so
we're
technically
past
zoo
for
zero,
four
ten,
which
we
were
hoping
to
release
this
week
since
we're
still
waiting
for
cappy
release.
We're
delayed
on
this
one
as
well,
which
is.
A
Mustang,
oh
yeah,
we
should
discuss.
Are
we
doing
0
for
11?
Are
we
starting
with
alpha
4?
What's
because
we
had
seven
zero,
four
that
might
be
the
last
one?
It
might
not
be.
D
Bro
we
can't
pick
up.
We
can't
pick
up
the
vr
for
four
types
yet
correct.
A
D
A
So
I
think
in
the
meantime,
we
shouldn't
block
ourselves
from
moving
forward
with
features
and
releasing.
I
think
so.
A
Okay
sounds
good,
so
in
terms
of
the
due
dates,
we
usually
do
once
a
month.
Obviously
zero.
Four
ten
was
a
bit
longer
and
zero.
Four
eleven
will
probably
be
longer
too
there's
also
thanksgiving
week
in
the
us
in
two
weeks.
A
F
A
Okay,
so
let's
just
go
through
these
and
see
which
ones
we
want
to
keep
in
here
that
we
think
we're
gonna
get
done
by
next
week,
which
ones
we
want
to
move
back
to
next
and
which
ones
we
want
to
move
to
zero
four
eleven:
okay
enhancement
proposal,
clouds,
resource
reconciliation,
david's.
What's
the
plan
on
that
one.
B
I
it
has
continually
been
bumped
down
like
the
priority
list
like
it.
It's
always
like
number,
two
or
number
three
and
something
else
comes
up
and
then,
like
I,
I
think
we
put
it
in
here
with
the
hopes
of
getting
it
in
with
telemetry
coming
in
ahead
of
it,
and
we
prioritize
that
and
I
think
right
now,
the
top
priority
is
to
get
machine
cool
version
upgrade
working
as
opposed
to
like
getting
the
enhancement.
B
I
don't
feel
bad
about
kicking
the
enhancement
down
a
little
further.
I
don't
know
what
do
others
feel
like.
A
I
think
we
just
haven't
had
the
urgency
to
do
it
so
far,
because
we
don't
have
the
implementation
yet
because
we're
waiting
to
see
what
happens
with
aso
and
like
so
for
those
who
don't
know
azure
service
operator.
It's
like
a
project
for
basically
azure
operators
to
reconcile
like
infrastructure
resources,
which
is
very
interesting
for
us,
because
it's
it
would
kind
of
do
what
we
were
hoping
to
use
here.
But
it's
not
to
the
point
where
it's.
It
doesn't
support
all
the
resources
that
we
want,
yet
that
we
need
for
kubernetes
cluster.
A
But
I
know
they're
working
on
that
and
we
had
talked
with
the
team
who
maintains
it
at
their
office
hours
and
they
would
want
to
like.
Have
us
be
early
adopters
potentially,
and
in
that
case
they
would
try
to
prioritize
like
the
resources
that
cabsie
needs,
and
so
I
think
like
when
we
get
to
that
point.
Where
we're
like
trying
to
spike
on
that,
then
we
need
to
get
the
proposal
done.
But
until
then
it's
kind
of
like,
like
you,
said
david.
It's
always
going
to
be
second
priority.
A
So
we
can
keep
it
in
the
mail
zone
and
look
at
it
and
bump
it
down,
or
we
can
just
move
it
to
next
for
now
and
if
it
comes
up
earlier
than
plan,
that's
fine.
D
We're
waiting
for
updating
the
proposal
before
we
can
merge
stuff.
I
don't
know
if
you
want
to
include
this
in
zero
four
ten,
considering
just
like
a
really
big
thing
that
we
would
be
merging
if
we
merge
it
in
like
the
last
few
days.
So
I
think
it's
done
from
my
perspective.
I
just
fixed
the
owner
reference
thing
and
we
just
need
danish
to
take
a
look
at
it,
make
sure
we're
using
it
properly
and
david
to
update
the
proposal.
D
D
A
That
was
my
my
guess,
but
I'm
not
100
sure.
D
No,
I
meant
I
asked
about
that,
because
the
original
proposal
had
mentioned
that
we
need
a
change
in
cluster
color
move,
but
then
we
kind
of
completely
changed
our
approach.
So
I
don't
know
if
that's
still
a
thing
or
not
that's
what
I
was
trying
to
ask
about,
but
I
can
test
that
I
haven't
tested
that.
D
I
would
say
bump
unless
you
think
it's
the
persian
to
get
it
in
for
antenna
and
then
I
can
like
focus
on
it
and
try
to
do,
but
I
prefer
to
bump
it
okay.
What
do
you
think.
D
B
B
Fair,
so
it's
pretty
big
on
that
on
the
topic
of
testing,
we
could
just
make
another
service
principle
so
that
the
job
doesn't
actually
have
to
create
that
service
principle,
and
then
we
can
use
that
one
in
our
tests,
similarly
to
how
we've
been
using
it
in
the
controller
so
far
like
we
have
one
in
the
environment
for
the
controller,
we
could
very
easily
create
another
service
principle
and
use
that
as
an
identity
within
that
cluster.
D
Yeah,
that
was
the
part
that
was
getting
stuck,
is
like
me
and
like
the
before
the
actual
test,
creating
that
identity
like
the
actual
identity
in
azure,
so
that
you
can
create
azure
identity
returns
for
it.
A
D
A
I
can
help
with
that.
I
have
access
to
the
subscription,
so
I
can
help
you,
okay,
perfect.
Okay,
when
running
work,
clothes,
oh,
that
one
we
also
have
been
bumping
james.
I
don't
know.
If
you
can
hear
me,
you
probably
can't
see
my
screen,
but
I
don't
know
if
you
remember
there's
this
issue
where,
like
a
single
control,
plane,
node
takes
like
oh,
it
takes
15
minutes
for
the
load
bouncer
to
provision.
A
A
C
Yeah,
it's
going
to
be
totally
in
the
controller
manager.
I
think
it's
a
pretty
low
priority,
because
the
likelihood
of
running
a
single
control,
plane
node,
is
pretty
small
and
that's
the
kind
of
use
case
that
happens
when
you
have
multiple
control
plane
nodes
once
you
once
you
have
that
it
triggers
the
leader
election,
which
causes
the
cash
to
to
refresh
regularly.
So
that's
why
we're
not
seeing
it
like
in
the
end-to-end
test.
I
think.
B
I'm
sorry
just
just
to
be
clear
that
was
in
azure
cloud
provider.
A
C
B
Pessimism
versus
optimism,
I
think
it
I
think
it's
most
likely
to
land
in
411.
I
just
just
based
on
how
long
it's
taken
to
to
solve
it
thus
far.
B
Oh,
no
just
next
or
411,
or
next
it's
it's
super.
It's
super
important
that
the
fix
is
probably
going
to
be
uncappy,
though.
C
A
A
A
D
B
This
shouldn't
be
any
change,
so
the
identity,
the
the
identity
structure,
has
already.
B
Yeah
yeah
exactly
so,
if
if
the
secret
is
a
reference
to
either
that
string-
or
I
mean
really
the
certificate's,
just
a
string
too,
so
it's
just
fill
that
secret
reference
with
the
certificate.
Okay,
yeah
just.
F
D
D
Or
after
I
think
it
seems
to
be
working
so
far,
I
ran
the
test
like
four
or
five
times.
We
can,
I
think,
keep
it
for
like
the
whole
next
week
when
we're
testing
the
new
rc
and
see
what
happens,
but
it
seems
to
be
fine,
so
I
would
say
we
can
merge
it
in.
A
I'm
leaning
towards
keeping
like
waiting
until
the
next
one,
just
because
once
we
release,
we
release
the
templates
and
hard
to
go
back
so
that
way
it
gives
us
more
time
to
get
some
mileage,
because
these
facts
aren't
very
common.
They
just
happen.
Sometimes
it's
possible.
We
got
lucky
so
far:
okay,
okay,
all
right!
So
for
now
we
just
have
the
flaky
jobs
in
the
current
one,
which
is
great.
We
can
focus
on
that.
D
A
Oh
wow,
weird,
okay,
this
is
definitely
next
anything
else.
A
Okay,
I
wonder
if
we
should
move
this
one
david,
the
azure
machine,
pull
notes
fail
to
join
after
15
minutes
before
upgrade
like
the.
I
think
this
should
be
zero
for
eleven
and
then
upgrade
should
be
next
and
then
stretch
for
zero.
Four,
eleven,
no.
B
That's
what
you
meant
by
upgrade.
I
was
thinking
even
when
you
you
upgrade
after
15
minutes.
It's
still
the
bug,
the
after
15
minute
bug
is
its
own
thing.
The
feature
ford
cordon
and
drain
with
you
know.
E
B
Upgrade
or
max
search
or
whatever
we
want
to
do.
That's
that
that's
more
of
a
feature
right.
A
Yes
and
that's
blocked
by
the
bug,
indeed,
okay,
oh,
I
would
like
to
add
this
one
to
zero
four
ten.
Actually,
because
I
the
pr
is
ready,
it's
just
waiting
for
an
approve
and
I
wanna
get
that
merged.
No,
there
are
no
objections.
It's
those
conformance,
ci
job.
D
A
C
So
one
of
the
things
I'm
working
on
and
I'm
pretty
close
is
the
windows
I
missed.
When
you
said
the
this
milestone
is
gonna,
be
completed,
hoping
to
have
a
pr
open
the
next
day
or
two.
C
A
0411.:
okay,
anything
else
that
we
want
to
prioritize
for
zero.
Four
eleven.
A
The
terminal
error
reconcile
a
req,
I
think
cheyenne.
You
were
working
on
that
or
there's
already
a
pr
for
it.
Should
we
put
that
in
the
mouse
zone.
E
Yeah,
so
I've
opened
a
work
in
progress.
Pr
so
like.
I
need
a
confirmation
on
the
approach
and
there
is
also
like
a
dependency
on
cappy,
where
they
were
discussing
about
adding
conditions
related
to
terminal
errors.
A
Yeah
awesome,
so
should
we
put
that
in
zero
for
11
for
now,
then,
since
it's
pro,
it
might
not
merge
in
the
next
two
days,
yeah
sure.
D
E
I'm
not
sure
entirely,
there's
still
like
a
discussion
going
on.
I
don't
know
what
the
exact
timeline
is.
A
A
A
A
No,
I
think,
that's
zero.
Actually,
I'm
gonna
move
that
to
zero
five,
because
that's
waiting
for
the
work
or.
E
A
Oh
yeah,
this.
A
I'm
just
checking
the
ones
that
have
people
assigned
to
them.
What's
that
why,
why
are
you
assigned
to
this
david?
Are
you
working.
A
Oh,
I
think
it
was
because
multi-tenancy.
B
A
A
4
11,
for
now,
I'm
just
leaving
it
as
next.
Okay,
all
right,
I
think
we're
close
to
done.
A
A
A
Yeah,
no,
I
think,
that's
different.
I
think
that's
blogging
authenticating
in
kubernetes,
using
your
azure
creds.
B
Yeah,
that
is
our
back
with
aad
behind
it
yeah
aad,
all
of
the
things
like
how
many
different
ways
can
like
aad
cut
through
or
our
different
cluster
configurations
right.
B
Go
ahead
that
identity
documentation
is
going
to
be
like
enormous.
A
I
think
there's
probably
room
for
a
few
more
things,
but
we
can
keep
it
as
this
for
now
and
then
as
we.
If
anything
comes
up
in
the
next
week
or
two
through
slack
or
anything
like
that,
we
can
add
it
on.
How
do
you
feel
about
that.
A
Yeah,
oh
yeah,
there's
cubecon
too
next
week.
Yes,
next
few
weeks
are
probably
gonna
be
slow.
I
think
next
office
hours,
which
is
supposed
to
be
in
two
weeks,
is
on
thanksgiving
thursday,
which
is
a
us
holiday,
so
we're
probably
gonna
skip
that
one
or
I
don't
know
if
people
who
aren't
in
the
us
want
to
still
do
it,
but
david
and
I
are
going
to
be
out
for
sure
in
jameson's
dwellings,.
A
Okay,
so
let's
just
cancel
that
one:
do
we
want
to
meet
again
in
a
month
or
do
you
want
to
do?
Should
we
do
like
a
intermediate
office
hours
we
can
do.
We
can
see
if
there's
need
for
it
in
three
weeks,
because
next
week.
A
A
Okay,
if
not
thanks
for
your
patience
as
we
went
through
this
nelson
planning
and
we'll
see
you
all
next
month,
have
a
good
q
con
if
you're
attending
kubecon
and
enjoy
the
u.s
thanksgiving
holidays.
If
you're
in
the
us.