►
From YouTube: kubernetes kops office hours 20190329
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.
This
is
cops
office
hours,
it
is
March
29th
2019.
We
have
I'm
your
moderator
and
facilitator.
Just
in
Santa
Barbara
I
work
at
Google.
We
have
an
agenda.
The
link
should
be
pasted
in
the
chat.
There's
a
couple
of
items
on
there,
including
a
great
graphic
which
I
think
we
can
get
off
with
a
minute.
A
A
Don't
want
to
share
the
shirt
for
the
video
so
that
people
can
see
but
yeah.
It
is
fairly
clear
that
we
are
getting
more
likely
at
releases.
I,
don't
know
what
happened
with
1:8
I
guess.
We
really
said
like
three
days
before
the
Alpha
three
days
before
I'm
like
okay
and
remember
doing
that,
but,
okay
and
I
guess
we
should
actually
get
back
to
that
sort
of
cadence.
A
A
Upstream
issues,
river
is
more
deliberate,
but
I
think
112.
We
don't
have
that
to
point
to
it's
just
it's
just
on
us
and
yeah.
We
should
we're
at
me
and
we
should.
We
should
address
that
I
think
it
makes
sense
in
happier
news.
We
did
release
112
alpha
2
and
alpha
3
I.
Didn't
you
offer
one
because
I
messed
it
up
this
week,
and
so
we
are
hopefully
getting
closer
on
the
112.
So
I
think
we're
going
to
talk
about
some
bugs
on
that.
A
minute.
A
One
of
the
things
we
could
do
is
do
the
113
and
114
alphas
and
betas,
basically
at
some
point
immediate
at
some
point
where
it
makes
sense
and
the
112
early
cycle,
so
I
didn't
want
to
do
all
three
of
exactly
the
same
time,
because
I
think
it
would
just
get
too
confusing
with
the
cherry-picks
but
I
think
once
if
we
get
112
to
beta,
then
we
can
do
113
alpha
and
possibly
114
at
the
same
time,
I
think
that
actually
is
reasonable.
It
was
yeah.
A
The
the
big
thing
is
113
turns
off
at
CD,
3,
sorry,
that's
et
to
shelter.
Everyone
estimate
this
before
no
I'm
kidding
the
it
turns
off
at
City
too.
So
that's
that
that's
sort
of
the
one
where
we
have
to
be
really
careful,
I
think
that's
where
we're
finding
a
lot
of
the
bugs
in
112
but
as
far
as
I
know,
there's
no
like
significant
changes.
There
were
some
api's
that
are
being
only
deprecated
but
I.
A
Yeah
there
is
nothing
there's,
no
like
significant
thing.
We
could.
We
could
accelerate
those
and
bring
them
down
to
I,
guess
more,
where
we
on
the
one
seven
timeline
where
we
have
like
a
one
or
the
one,
eight
timeline
for
alpha,
where
we
release
alpha
and
beta
little
before
release
of
kubernetes
and
the
one
seven
timeline
where
we
then
mark
the
stable
release
around
the
month
after
when
we
feel
it
is.
A
The
ecosystem
is
actually
stable,
so
kubernetes
and
CNI
providers
and
you're
not
going
to
blow
up
I,
don't
know
if
that
makes
sense
to
people
as
that
as
the
target,
and
then
we
could.
We
could
actually
release
the
Alpha
even
like
three
days
before
is
sneaking
under
the
wire,
but
we
could
release
the
alpha
like
much
earlier
like
we
could
release
a
115
alpha
I
mean
in
theory
now,
but
in
practice,
when
they
cut
the
first
RC
would
probably
make
the
most
sense
for
kubernetes.
C
Yeah
I
love
the
idea
of
getting
more
in
line
with
them
and,
like
you
know,
I
think,
there's
enough
people
on
this
call
that
that
have
been
around
for
a
while
and
seeing
that
layout.
We
purposefully
keep
things
back,
but
we
can
only
say
that
so
long,
we're
so
far
behind
you
know
and
I
think
it
works
really
well
to
say,
like
alpha
equals
RC
for
kubernetes,
kubernetes
and
beta
equals.
You
know,
release
that
way.
We're
like
hey.
A
That's
a
great
I
think
that's
a
great
thing
to
get
to
it's
actually
very
I.
Think
that's
great
I
think
it's
also
very
mostly
mechanical,
but
like
clear,
clear
and
objective.
A
A
A
E
First,
one
is
a
terraform
bug
which
I
think
Rodrigo
you
grabbed
assigned
to.
You.
I
took
a
what
it's
a
second
bug
in
that
extensions
policy
issue
that
got
at
it.
This
one
is
it's
just
not
base64
encoding.
The
user
data
when
that
feature
is
enabled
I
took
a
look
through
the
code
base
last
night
and
I'm
starting
to
find
where
that
is,
families.
E
E
But
that
one
brings
up
I
think
interesting.
We
should
probably
talk
about
how
we
test
some
of
that
stuff
a
little
bit
better
from
like
a
integrations
for
terraform
confirmation
and
how
we
like
the
first
bug
beyond
fixed.
Was
it
failed
with
just
a
terraform
plan?
This
one
is
only
telling
an
apply
time
so
it'll
like
it,
takes
an
actual
full
integration
sort
of
tested
to
catch
that,
but
I
think
both
of
those
are
worthwhile
like
at
least
doing
a
plan,
part
of
it.
E
One
of
the
masters
went
away
when
it
came
back
up,
it
refused
to
join
and
the
master
was
rejecting
it
and
the
it
was
a
TLS
error.
That's
in
that
in
the
logs
in
that
ticket
I
tested
that
without
for
three
and
I
haven't
been
able
to
replicate
it.
So
I
don't
know
if
it
was
a
fluke
or
if
it
was
an
edge
case
that
I'm
just
not
hitting
in
this
right
I
figured
that
Justin.
You
probably
wanted
to
look
at
that
one,
a
little
bit
more
yeah.
E
A
On
the
that's
about
the
two
nice
in
order
like
I'm
the
first
one,
we
actually
have
a
lot
of
testing
most
are
a
lot
of
our
testing,
actually
uses
the
terraform
output
and
verifies
that
the
terraform
output
is
sort
of
like
the
golden
image.
The
old
name
is
testing,
so
we
check
it
again
like
string
test
it
against.
You
know
what
we
expect,
which
checks
that
it
hasn't
changed
unexpectedly
and
that
it
looks
exactly
as
we
expect
it
to
it
doesn't
check
out.
The
terraform
itself
is
valid
yeah,
which
is
so
yes,
we.
E
A
E
A
Yeah,
the
other
thing
we
can
do
is
I
guess
when
we
actually,
when
we
add
a
when
we
add
a
new
set
of
terraform
outputs
or
targets,
I
guess
they're
called
like
a
new,
a
new
task.
We
should
actually
test
it
at
least
once
manually,
using
using
the
yes
using
in
that
scenario
and
I
guess.
The
other
thing
we
could
do
is
we
could
have
an
e
to
e
test.
A
It
actually
says
terraform
to
bring
out
the
cluster,
which
I
don't
think
we
have
today
that
that
wouldn't
catch
everything,
but
it
would
catch
whatever
scenarios
we
tested.
So
we
catch
catastrophic
type.
Things
like
I
think,
for
example,
there's
a
there's,
a
PR
out
which
says
the
next
version
of
terraform
has
a
stricter
syntax
around
escaping
or
something
like
that,
and
then
I
guess
that
each
each
s
would
be
great
to
have
there,
because
that's
gonna
change.
A
Yeah
I
think
yeah
at
least
at
least
I'll
run
a
manual
test
on
that
output.
I'll,
take
a
look
at
as
well
and
see
like
while
I'm
there,
whether
I
can
just
fix
it
at
the
same
time.
That's
four:
six,
seven,
no,
six,
seven,
zero
zero
and
then,
on
the
other
one,
a
question
on
that
which
is
the
SED
up
or
the
SED
master
issue.
E
A
E
A
Of
randomly
not
as
part
of
the
rolling
update,
rolling
update
was
successfully
gone.
Yeah.
Okay,
I
will
definitely
want
to
look
at
that.
Thank
you
for
the
bringing
that
one
up
the
there
is
something
around
caching
of
host
names,
but
I
actually
thought
we
worked
around
it
like
we.
We
had
like
passed
some
extra,
so
I
will
definitely
look
at
that.
Thank
you
for
the
heads
up
on
that
yeah
I.
C
Have
one
that
I'm
working
on
a
fix
floor
that
I
had
approved
the
knee
I'll,
be
tagging
thing
that
just
wasn't
gonna
work
at
all,
and
so,
if
you
have
cluster
tags,
it
continuously
tries
to
apply
them
to
the
ELB.
It
might
apply
to
the
EOB.
But
then
it
won't
ever
remove
ones,
and
you
know
it's
it's
just
it
was.
It
was
half
complete,
so
I
have
a
fix,
I'm
just
fixing
the
tests,
but
they
should
be
done
today.
Awesome.
A
I
mean
the
options
are
we
can
always
revert
if
it's
new
functionality
we
can
roll
forwards
and
try
to
fix
it.
We
can
feature
flag
it
if
it
like.
We
can,
if
it
just
like
sort
of
works,
but
doesn't
really
work
like
works
for
some
people,
but
that's
another.
That's
a
middle
ground
between
pervert
and
and
fixing,
but
yeah.
If
you're
able
to
fix
it.
That's
great.
Is
there
a
issue?
Never
for
that.
There's.
A
Okay,
great
I'll,
look
at
that
yard
looks
like
it
is
six
703
right,
yeah
just
keep
it
growing,
Justin
the
the
issue
with
the
calico
upgrade
yep.
Yes,
the
that
was
actually
my
fault.
It
was
I
thought
it
was
something
complicated.
It
isn't.
It
was
as
part
of
that
apps
v1
change,
yes,
a
heads-up
to
everyone,
so
we're
moving
from
absolutely
one
beta
one
four
deployments
and
stateful
sets
to
apps
v1
and
there
is
a
breaking
breaking
change.
The
the
selector
the
selector
is
required
in
apps.
A
V1
I
had
dropped
the
selector
when
I
like
rebase,
because
I
was
also
moving
from
like
upgrading
calico
to
CR
DS
and
the
other
liceo
DS,
but
the
PR
is
crossed,
I
dropped,
the
selector,
and
so
all
that
was
happening
was
the
112
zero
alpha,
2,
calico
c
rd
was
just
in
that.
So
that's
what's
going
wrong.
So
how
do
you
already?
Have
you
already
fixed
that
then
and
put
it
in
yes,
one
twelve
zero
alpha
three
has
the
six,
which
is.
A
A
Okay,
but
I
want
six
six
three,
six
thanks!
Oh
thank
you,
oh
and
so
I
also
fixed
the
the
other
issue.
We
we
found
on
success,
three
six,
which
is
if
you
hadn't,
see
a
legacy
s
CD
manager
using
HTTP,
where
you
sort
of
opted
into
HTTP,
and
then
you
try
to
use
sorry
if
you
have
a
legacy
at
CD
mode
and
you
opted
into
H
into
TLS
and
then
you
try
to
use
Etsy
D
manager
at
city
manager
refuses
to
import
it
because
it
says
I
don't
recognize
TLS.
A
A
How
do
people
about
getting
to
beta
well,
it
sounds
like
we
have
a
round
of
runner
a
round
of
fixes
and
we
should
get
those
in
getting
to
beta
and
then
at
when
we
do
112
112
zero
of
beta.
We
also
cut
113
alpha
and
114.
Alpha
I
think
I
can
I
can
maintain
the
cherry-picks
on
three
branches.
I
have
a
script
which
shape
which
I
paste
it
in
somewhere,
so
that
other
people,
if
read
and
wants
to
cherry-pick,
feel
feel
free
to
steal
that
script.
Weird
regression,
it
was
on
one
of
your
yeah
yeah.
A
A
A
A
But
if
I
don't
wants
an
easier
one,
it's
it's
floating
around
I
support
ease,
exactly
stricter,
but
much
simpler.
Okay
and
okay.
I,
don't
know
if
we
have
anything
else
on
the
oh,
we
have
another
I,
don't
echo
Shane!
That's
your
issue.
We
just
talked
about
I.
Don't
have
anything
else
on
the
agenda
that
we
should
talk
about
or
whether
everyone
would
like
for
the
odd
minutes
back.
E
C
A
A
Think
that's
fair,
but
I
think
the
the
other
things
are.
We
aren't
building
our
other
binaries
for
arm
either
I,
don't
know
whether
it's
a
top
priority
to
build
the
cop
CLI
forearm,
for
example,
I'm
sure,
once
we
I'm
sure
it
will
not
be
too
hard
once
everything
else
is
done,
but
the
for
the
am
is
I
think
we
can
I
would
recommend
using
like
a
stock,
a
stock
AMI
either
Amazon
system,
Linux
or
Debian
stock
or
Ubuntu
or
CentOS,
or
whatever
your
preferred
image
is
and
then,
but
we
would
still
have
to
do
that.
A
A
A
We
already
build
cops
for
different,
OSS
and
I.
Think
it's
just
a
similar
but
slightly
different
flag,
just
to
say,
built
for
arm
or
cross
build
to
arm.
We
have
protic
you,
which
is
a
darker
saved
in
so
I'm
guessing.
We
would
actually
that
that
one
might
be
not
too
hard
either
that
one
would
be
built
and
then
exported,
but
the
other
one
we
need
is
we
need,
like
a
CD
manager.
For
example,
we
need
the
arm
version.
A
If
there
are
any
other
changes,
we
need
to
do
to
pull
like
the
arm
version
of
cube,
API
server.
We
have
to
plumb
those
through
so
I,
guess,
I
guess.
The
next
step
would
be
to
see
how
QAPI
server,
for
example,
does
it
and
we
should
just
mirror
whatever
QAPI
server
does
like
I?
Don't
if
anyone
knows
whether
cubic
bi
server
has
different
names
on
arm
or
whether
it
uses
multi
arch.
A
C
A
A
D
E
A
Yeah
it
feels
like
we
have
four
open
ish
forward
like
areas
and
yeah.
Those
four
seem
good.
It
also
means
that
people
are
trying
it
and
I
think
that
is.
It
is
reasonable
to
go
to
beta
beta
what,
if
people
think
that's
fine
and
then
doing
beta
1,
then
that
will
trigger
me
to
do
bait
out
113,
alphas
and
114
alpha.