►
Description
Want to view more sessions and keep the conversations going? Join us for KubeCon + CloudNativeCon North America in Seattle, December 11 - 13, 2018 (http://bit.ly/KCCNCNA18) or in Shanghai, November 14-15 (http://bit.ly/kccncchina18).
kops Intro – Justin Santa Barbara, FathomDB (Any Skill Level)
About Justin
Justin is one of the kubernetes sig-aws leads and started the kops project, so loves to talk about how to install and operate kubernetes, or on all things kubernetes-on-AWS or on other clouds (particularly GCP, having just joined Google!)
Join us for KubeCon + CloudNativeCon in Barcelona May 20 - 23, Shanghai June 24 - 26, and San Diego November 18 - 21! Learn more at https://kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy and all of the other CNCF-hosted projects.
A
Hi
everyone,
thank
you
so
much
for
coming
on
a
Friday
afternoon
of
Kubik
on
its
I
know,
it's
off
I
think
people
are
think
about
going
home.
So
thank
you
for
for
coming
in
here.
I
am
gonna,
give
you
an
introduction
to
cops
and
an
update
on
the
project.
We're
gonna
start
from
nothing
and
then
go
through
like
some
of
the
internals
and
a
little
bit
of
road
map.
A
So
if
you're
more
advanced
apologize
to
the
beginning,
if
you're
more
of
a
beginner
I
apologize
for
someone's
stuff
at
the
end,
but
I
wanted
to
just
give
you
a
very
quick
I
want
to
start
with
a
very
quick
idea
of
you
know
if
you
haven't
seen
it
like
what
is
the
cops
experience?
So
if
we
want
to
create
a
cluster,
you
do
cops,
create
cluster,
the
name
of
the
cluster
and
these
zones
in
which
you
want
to
do
it.
A
A
Better,
yes,
okay,
I,
actually
recently
joined
Google,
so
I
realized
that
I
probably
should
be
creating
it.
Not.
Let's
do
I'm
gonna
get
to
the
top
of
the
screen
again,
not
in
the
Amazon,
your
C's
1
B,
but
in
Google's
a
u.s.
East
1
B,
so
that
will
create
it
in
Google
instead
of
in
Amazon
and
actually
have.
If
we're
gonna
do
that,
let's
have
some
fun
and
do
a
oh
there.
A
It
was
so
we
are
gonna,
do
an
H,
a
cluster
with
master
count
3
and
in
zones
us
each
one
B
or
C,
1
C
and
us
each
one
D.
So
we
will
start
that
going
and
then
we'll
get
into
the
talk
and
we'll
come
back
in
a
couple
of
minutes
and
hopefully
fingers
crossed
everything
will
be
working.
So
you
can
see
it's
thinking
about
it.
It
infer
that
that
line.
A
There
says
that
it
detected
that
the
cloud
was
GCE
based
on
the
zone
we
specified
and
now
it's
what's
doing,
is
it's
basically
building
a
a
model
similar
to
how
terraform
works
and
you
can
see
all
the
tasks
is
what
we
call
them
that
it's
going
to
create.
So
a
bunch
of
key
pairs,
some
instance
blades
for
GCE
some
instance
group
managers,
which
are
like
auto-scaling
groups
on
AWS,
some
firewall
rules
anyway,
a
whole
bunch
of
stuff,
and
we
can
now
do
update
cluster.
A
Yes,
it's
actually
going
crate.
Pardon
me
yeah
what
I
do
test
just
cops.
Cops
update,
cluster
test
gates,
the
name
of
our
cluster.
It's
actually
going
create
it
in
the
cloud
and
before
it
was
just
saying
what
it
would
do,
think
think
think
think
and
it's
gonna
run.
Those
I
think
it
was
sixty
odd
tasks
that
it
had
to
run.
That's.
A
Issuing
some
certificates,
the
internet
is
a
little
bit
slow
I
apologize
for
that.
Well,
we
can
leave
that
running
and
we'll
come
back
to
that
in
a
second
I'm,
pretty
sure
it's
going
to
succeed
once
the
intern
is
happy
so
overview.
What
the
talk
is
gonna
be
quick
introduction
to
cops
and
a
demo
a
little
bit
about
some
more
advanced
usage
of
cops.
A
So
if
something
does
go
wrong,
you
know
you
have
some
idea
of
what
to
do
and
sort
of
the
vision
for
for
cops
for
the
next
I
guess,
one
to
two
years,
what
we're
sort
of
what
the
big
areas
are.
So
we
looked
at
the
UX
around
creation.
You
create
a
cluster.
That,
essentially,
is
the
cop
CLI
tool
is
supposed
to
feel
a
lot
like
coop
cuddle.
A
So
when
you
coop
code
or
create
secret,
for
example,
it
goes
if
that's
a
sort
of
facade
or
that's
sugar
on
top
of
creating
an
API
object
or
a
set
of
API
objects
and
cops.
Create
cluster
is
sugar
on
top
of
creating
our
API
objects.
So
a
cluster
and
a
set
of
instance,
groups
is
what
we
called
on
the
cluster
being
the
general
configuration
and
a
set
of
instance,
groups.
Being
an
instance
group
manages
a
set
of
instances,
so
we
have,
for
example,
in
an
H,
a
cluster
configuration.
A
We
have
three
instance
groups,
one
for
a
master
in
each
zone
and
then
one
instance
group
for
the
nodes-
and
you
could
add
at
different
instance.
Groups,
for
you
know,
nodes
that
have
GPUs
things
like
that,
so
cups.great
cluster
cups
update
cluster,
which
actually
goes
and
does
the
actual
creation
of
the
objects.
This
is
a
little
bit
not
like
kubernetes
in
that,
because
these
resources
cost
money.
Unlike
most
kubernetes
resources,
we
don't
immediately
create
them
and
then,
in
a
couple
of
minutes
you
are
able
to
use
coop
cut.
A
A
A
When
we
go
and
see,
let's
see
how
we're
doing
Wow,
okay,
yep
cool
we're
scrolling
problems.
So
yes,
so
in
the
background
we,
while
that
was
while
I
was
blabbering
on,
we
finished
the
cups
update
cluster
command.
So
we
went
and
created
a
bunch
of
instances,
I
suspect.
So
we
should
have
created
three
masters
and
two
notes:
yay
three
masters
master,
they're,
interleaved
and
then
the
two
nodes
in
the
middle
and
if
we're
really
lucky
this
mean
this
may
be
a
little
too
fast,
but
we'll
see
hasn't
got
there.
Yet,
oh,
yes,
it
has
okay.
A
Well,
we
actually
just
hit
it.
Our
three
masters
have
joined
the
cluster,
but
our
notes
have
not
yet
joined
the
cluster.
So
we'll
come
back
to
that
in
another
minute,
but
I
wanted
to
show
you
the
which
we
look
at
first,
let's
look
at
how
you,
let's
get
the
upgrade
scenario
a
little
bit.
So
if
you
wanted
to
change
so
again,
a
lot
like
coupe
cuddle,
coupe
cuddle
edit,
would
be
the
way
you
would
interact
with
these
objects.
So
cops
get
cluster
will
list.
The
is
like
coupe
cuddle
get
objects.
A
You
can
see
the
cluster
we
created
their
cups
edit
cluster
is
the
equivalent
of
coupe
cuddle
edit
cluster
here
is
the
yellow
object
for
the
cluster,
and
you
can
see
that
we're
running
1/9
something,
and
we
might
want
to
change
that,
for
example,
2,
1,
4.
Sorry,
you
can't
see
cuz
my
cursor.
We
might
want
to
change
that
to
194,
and
then
we
would
do
a
save
it.
We
would
do
a
cops
update
and
a
cops
rolling
update
there,
and
that
would
be,
although
be
required,
to
update
from
193
to
1
and
4.
A
For
example,
what
we
just
did
was
the
equivalent
of
coupe
code
or
edit
on
a
deployment
which
isn't
you
know
it.
It
works
great
in
kubernetes,
but
it
isn't
necessarily
the
way
that
we
recommend
most
people
manage
their
clusters.
So
a
better
way
to
manage
it
is
out
of
order.
A
better
way
to
manage
is
a
sort
of
the
declarative
way
of
managing
your
cluster,
where
you,
some
people,
call
it
get
ups,
you.
A
Basically,
you
store
all
your
yamo
files
in
in
git
and
then
you
make
changes
in
that
git
repository
through
a
you
know,
your
normal
PR
process
and
you
go
through
you
review
those
changes
to
your
infrastructure.
You
sign
off
on
them,
you
merge
them
and
then
a
CI
or
CD
system
would
apply
those
changes
for
you.
So
if
we
look
at
what,
for
example,
cups
get
cluster,
oh
does
cops,
get
cluster
will
dump
the
full.
A
The
full
state
of
I,
think
I
mean
cops,
get
test
arcade-style
ogre.
This
dumps
the
full
gamal
for
all
the
objects
in
your
cluster.
So
there
we
are.
We
see
a
bunch
of
instance
groups
and
our
cluster
speck
at
the
top.
If
I
pipe
that
to
my
cluster
channel,
then
I
can
save
that
and
now,
if
I,
for
example,
wanted
the
way
we
would
normally
do
an
update
in
a
sort
of
more
production
environment.
A
So
we
have
very
quickly
covered
a
lot
right.
There
I
don't
know
if
I've
lost
everyone,
I'm,
sorry
if
I
have,
but
what
we've?
What
we've
seen
right
there
is.
We
have
cops
that
creates
API
objects.
That
should
feel
a
lot
like
interacting
with
coop
cuddle.
We
were
able
to
create
clusters
on
your
infrastructure,
B,
AWS
or
GCE
very
quickly.
A
You
can
use
cops
edit,
which
is
a
bit
like
coop
cuddle
edit
to
do
interactive
changes
to
your
cluster
and
we
have
a
workflow
that
is
designed
for
integrating
into
get-ups
or
more
sort
of
structured
environments.
So
that's
if
you
wanted
to
you
know,
have
an
SRE
type
review
the
changes
to
your
custom
before
you
made
them.
Some
environments
are
even
more
locked
down,
because
this
is
sort
of
a
black
box
and
when
I
run
cops,
update,
will
show
you,
my
cups,
update
cluster,
so
we'll
show
you
some
other
changes
that
are
gonna
happen
here.
A
We'll
show
you
the
changes
that
are
gonna
happen
here,
as
we
see
them
thinking
so
there
we
are.
So
you
can
see
that
we
were
going
from
1.
9
3
is
the
last
one
we
actually
applied
going
to
1
9
5,
and
you
can
see
that,
for
example,
the
instance
templates
which
control
the
configuration
of
those
instance
groups
has
changed,
and
you
can
see
that
the
instance
group
managers,
which
is
the
the
actual
groups
of
those
instances,
is
pointing
from
the
old
one
to
the
new
configuration.
A
A
The
reason
we
didn't
use
terraform
initially
was
because
at
the
start,
terraform
had
the
TF
or
still
has
the
TF
state
file,
and
it
also
was
not
as
good
at
importing
existing
resources.
So
one
of
the
things
here
is
that
we
don't
have
a
strong
and
a
requirement
around
like
maintaining
a
TF
state
file,
it's
more
similar
to
kubernetes.
A
In
that
regard,
and
so
it
says
here,
terraform
output
has
been
placed
into
out
terraform
and
if
we
find
out,
you
can
see
that
here
we
have
a
pretty
big
terraform
template
with
the
same
resources
right
there.
So
as
this
chain,
as
as
your
configuration
changes,
you
can
choose
to
output
to
terraform
and
get
the
changes
which
a
lot
of
I
guess,
more
ops
or
more
controlled
environments,
really
like
one
of
the
sort
of
the
best
way
to
use.
A
This
is
to
put
this
into
a
TF
module
and
we
have
some
output
variables,
which
well
particular
on
AWS,
it's
more
important
to
output,
things
like
the
subnet
IDs
and
the
VPC
ID.
So
on
AWS
those
appear,
there's
less
need
for
it
on
Google,
where
the
the
objects
are
referenced
by
name,
but
this
works
well,
where
you
put
it
into
a
TF
module,
and
then
you
check
it
in
with
your
application
and
then
your
other
terraform
things
can
reference
your
your
cops,
managed
cluster.
A
We're
on
time
we're
doing
good
yeah
we're
doing
it
all
right.
That
was
a
very,
very
quick
and
then
so
you
go
in
there.
Utf
plan
UTF
apply,
and
it
would
create
your
your
cluster
for
you
in
that
way.
So
we
have.
We
have
raced
through
the
cops
user
experience
like
what
it
is
like
to
run
a
soar
on
a
cluster
using
using
cops,
let's
see
so
where
are
we
as
a
project
like
what
is
happening
in
cops
then?
So
today
we
work
great
on
AWS
and
on
GCE.
A
Gc
is
technically
still
feature
gated,
but
we're
just
trying
to
figure
out
whether
IP
aliases,
which
are
a
sort
of
relatively
new
feature
of
kubernetes
and
google
cloud,
whether
IP
aliases
should
be
turned
on
by
default,
or
not
it's
sort
of
hard
to
switch
between
them.
So
that's
what's
holding
up
GC,
but
there's
some
excellent
work
going
on
right
now.
Thank
you.
Andrew
for
our
digital
ocean
support
is
coming
in.
What's
the
stay?
Where
would
you
say
it?
Is
it's
two
alpha
beta
soon
there
we
are
so
digital
ocean.
A
A
I
mean
I,
just
I'm
trying
to
do
a
sort
of
race
through
everything
to
give
everyone
a
grounding
in
this.
No
matter
like
where
you
are
the
very
quickly
how
it
works.
We
use
the
kubernetes
api
machinery
for
the
court
objects,
so
the
cluster
and
the
instance
group
objects
you
saw
and
the
way
like
cops
edit
was
able
to
behave.
Sort
of
similarly
to
coop
code
or
edit
that
all
works
because
we're
using
the
kubernetes
api
machinery,
so
one
of
the
things
that
we
are
hoping
will
happen
or
I'm
hoping
what
happened.
A
I'm
gonna
work
on
is
getting
creating
a
server
version
of
cops.
So
right
now
cops
is
a
CLI
tool,
but
because
we're
using
API
machinery
objects,
we
can
actually
create
a
cops
server
and
interact
with
cops
in
that
way.
That
would
mean
that
right
now
you
can
have
a
team
and
you
can
all
share
an
s3
bucket
or
a
GCS
bucket.
This
would
be
a
even
more
advanced
where
you
can
start
to
apply
our
back
rules
or
namespaces
or
whatever
you
whatever.
A
You
can
do
with
kubernetes
all
of
those
sort
of
things
now,
so
we
create
those
objects
where
we
manage
those
objects
or
you
manage
those
objects
and
then
the
the
cops
tool
when
you
run
cops
update
it
Maps
that
there's
a
model
layer
which
sort
of
transforms
the
the
objects.
Cluster
and
instance,
group
objects
into
tasks
which
are
the
units
that
will
go
and
create
a
firewall
rule
or
a
order
scaling
group
on
AWS
or
a
managed
instance
group
on
GCE,
so
the
model
layer
is
what
sort
of
translates
from
the
abstract.
A
This
is
the
cloud.
This
is
the
configuration
of
the
cluster
that
I
want
into
actual
cloud
man
like
the
manifestation
in
cloud
objects.
So
it's
the
one
that
knows
like.
What's
the
what's
the
right
way
to
run
a
cluster
on
AWS
or
what's
the
right
way
to
run
a
cluster
on
GCE
and
then
each
task
is
sort
of
responsible
for
reliably,
creating
or
Gordon
updating,
changing
the
cloud
objects.
So
each
task
has
retry
logic.
Each
try
has
each
task
has
like
mutation
logic
to
update
it,
can
update
and
write
to
terraform
or
cloud
formation.
A
That's
that's
sort
of
the
structure
of
it.
If
you
ever
get
into
the
code
and
the
tasks
by
the
way
are
when
you
saw
the
the
dry
run,
output,
the
the
that
when
I
ran
cups,
update
cluster
without
yes
and
we
saw
the
simulation
of
what
was
going
to
happen.
Those
are
the
tasks,
that's
a
diff
at
the
task
level.
So
when
you're
using
the
CLI,
all
the
CLI
is
doing
is
generating
those
tasks
and
then
essentially
running
them
all.
A
In
sequence,
we
have
a
sort
of
clever
ich
clever,
like
the
directed
tree
or
directed
graph
algorithm,
to
sort
of
figure
out
the
correct
sequence
in
which
to
run
them
all,
and
each
task
sort
of
in
a
kubernetes
inspired
way
reconciles
the
the
actual
state
of
the
cloud
object
as
it
finds
it
to
the
state
which
is
desired
and
so
cops
creates
the
instances,
there's
cloud
objects,
and
then
it
exits
the
picture.
It
doesn't
run
anymore.
A
A
Another
good
thing
is
that
it's
self-healing,
one
of
the
one
of
my
favorite
party
tricks,
is
to
like
turn
off
all
the
nodes
that
you
have
all
the
instances
that
you
have
and
on
AWS
and
auto
scaling
group
will
cause
them
to
basically
relaunch
the
same
thing
on
GCE
with
the
managed
instance
group.
So
you
don't
at
3:00
a.m.
in
the
morning
when
you
have
a
disaster,
you're
not
running
the
cup
CLI
tool
again,
because
we're
not
in
the
picture
anymore.
The
bad
thing
is
that
we
probably
should.
A
But
currently
we
don't
even
check
success,
so
the
cups
to
exits
like
very
after
about
30
seconds,
but
it
takes
around
four
minutes
for
your
for
five
minutes
for
your
cluster
to
come
up,
and
so,
if
you
don't
have
enough
quota,
for
example,
for
a
machine
to
launch
it's
a
little
bit
awkward
to
figure
out
exactly
why
so.
We're
gonna
have
to
fix
that
in
the
UX,
but
I
think
we
have
over
cups
of
validate
commands.
A
So
we
just
need
to
like
spawn
that
one
and
then
watch
what
happens
the
more
interesting
or
where
things
go
wrong.
More,
therefore,
is
how
it
works
on
the
node,
so
each
instance
that
boots
up
we
want
to
have
it
join
the
cluster,
either
as
a
master
or
as
a
node.
So
the
instance
boots
up
and
runs
a
startup
script
on
AWS.
A
That's
the
user
data
on
GCE
I
think
it's
also
called
the
user
data,
maybe
the
instance
data,
but
there's
some
startup
script
mechanism,
which
is
which
is
run
at
boot,
and
the
role
of
that
script
is
only
to
download
a
component.
We
call
node
up
and
notif.
Shaab
is
essentially
to
normalize
the
image
by
that
I
mean
install
any
packages
that
are
required
and
missing.
A
I
think
we
can
still
get,
for
example,
because
technically
get
is
requirement
of
kubernetes
in
terms
of
the
get
volume
source
get
volume.
We
installed
the
correct
version
of
docker,
for
whichever
version
of
kubernetes
you're
using
each
version
of
communities
is
tested
with
one
or
two
versions
of
docker.
So
we
try
to
make
sure
it's
the
right
one
we
install
cubelet.
Obviously
we
install
the
control
plane
manifests
and
we
install
a
component
called
protic.
You
and
protic.
A
You
Bob
pretty
cue
buttons
on
the
master
and
pretty
cubes
job
is
to
deal
with
EDD
so
critically,
we'll
go
and
find
the
correct
volumes
for
Etsy.
Dee
will
start
at
CD
and
install
at
some.
Other
system
manifests
like
cube
DNS,
so
this
is
sort
of
the
protocol.
A
shavette
CD
is
sort
of
the
big
trick
which
that's
the
that's
how
it
works.
A
Cops
uses
DNS
for
discovery
that
was
sort
of
one
of
the
big
design
decisions,
discovery
that
the
two
sort
of
main
elements
of
discovery
are
that
each
node
has
to
be
able
to
know
where
the
master
is
in
a
cloud
type
environment.
That's
not
obvious,
and
the
other
one
is
that
the
HCD
nodes
need
to
be
able
to
discover
each
other
and
I'm
a
chai
master
scenario
or
in
a
live
over
the
at
CD
scenario.
That
approved
pretty
good.
A
But
if
you
want
to
use
real
DNS,
it
was
a
bit
of
a
hurdle
for
some
people,
so
you
saw
that
I
created
test,
arcade
start
local,
which
is
a
new
one.
We
have
got
called
gossip
DNS,
which
is
sort
of
an
easier
way
to
to
get
going.
It
doesn't
it's
not
as
reliable
as
a
real
DNS
name,
but
it
is
we
do
test
it.
It
seems
to
work
fine,
but
it's
just
not
the
preferred
option,
but
it's
certainly
much
easier
to
get
going
with.
A
The
best
option
is
to
use
a
real
DNS
name
that
requires
setting
up
your
NS
records,
which
is
the
challenge
sometimes
on
AWS.
You
can
use
a
private
hosted
zone
in
route
53,
which
means
that
your
it
doesn't
actually
have
to
be
a
real
DNS
name
anymore
anyway,
so
that
you
do
cops,
create
cluster
D
and
s
equals
private.
If
I
recall
correctly,
but
DNS
for
discovery
was
a
sort
of
big
design
session
for
cops
and
they
worked
out.
Okay,
I
think
would
be
how
I
put
it.
So
what
is
what
is
there?
A
Cops
is
sort
of
a
big
thing
with
a
lot
of
pieces
built
into
it
and
I
think
everyone
wants
there's
a
lot
of
efforts
happening
in
macroon,
A's,
installation,
space
and
I
think
for
cops.
We
have
a
lot
of
good
ideas
and
some
ideas
that
maybe
didn't
work
as
well,
and
so
we
want
to
sort
of
break
up
cops
into
the
relevant
pieces
so
that
we
can
share
the
pieces
and
reuse
pieces
that
other
people
have
used.
A
So
the
prototype
component
we're
going
to
break
it
up
into
sort
of
a
separate,
ed,
CD
manager
and
atom
manager.
Thinking
about
the
DNS
and
discovery
manager.
How
discovery
works
note
up?
We
want
to
basically
integrate
with
cube
ATM,
which
does
a
fairly
good
job.
Where
does
a
good
job
of
the
of
the
equivalent
functionality
for
a
node
up,
but
we
also
have
some
extra
pieces
we
do
like
cube.
Atm
does
not
install
docker,
so
we
need
to
figure
out
who's
going
to
do
that.
A
Etsy
D
manager
is
sort
of
a
big
one.
I
don't
want
to
spend
too
long
on
this,
but
basically
we
are
working
on
getting
SCD.
Yellow.
Most
users
today
of
cops
are
using
SED
to
exit
e3
is
deprecated
in
will
be,
Depp
will
be
removed,
it
is
deprecated
and
at
CD
two
will
be
removed
in
communities
1.13,
so
we
have
to
help
our
existing
users
get
from
su
2,
2,
SC
3
and
we're
sort
of
leading
the
charge.
A
I
guess
we're
championing
like
how
we
get
there,
particularly
in
an
H,
a
scenario,
and
so
ed
city
manager
is
gonna,
be
the
piece
that
that
does
that.
So
in
1/9
we
added
backups,
which
are
sort
of
the
first
step
to
move
from
the
protocol
at
City
Manager
to
the
standalone,
ed
city
manager
in
cops
110,
which
is
not
released
yet
you'll
be
able
to
opt
in
to
ed
city
manager
and
we're
gonna,
make
it
CD
3
the
default
for
new
clusters
and
in
111.
A
A
That
should
work
the
same
way
on
every
cloud,
a
machine
set
which
is
like
a
replica
set
and
a
machine
deployment
which
is
like
a
deployment.
So
soon
you
will
be
able
to
manage
instances
in
the
same
way
that
today
you
manage
your
pods,
so
through
deployments
and
each
cloud
or
each
provider
will
only
have
to
write
a
machine
controller
and
the
machine
deployment
will
work.
The
same
way.
A
Thinking
about
we're
thinking
about
in
cups
like
what
does?
How
should
discovery
work
is
DNS,
the
best
mechanism
can
we
move.
There's
a
great
project
called
external
DNS,
which
we
mainly
use
and
other
mechanisms
for
discovery
include
things
like
IP
tables,
so
I
think
Able's
are
the
way
that
services
work
in
kubernetes
and
can
we
adopt
those
as
in
more
reliable
way
or
any
of
of
doing
discovery
that
require
you
to
go,
get
a
real
DNS
name
and
add
all
management.
A
I
think
is
the
last
thing
that
we're
sort
of
looking
at
spitting
out
so
cups
includes
some
I.
Guess
you
call
it
extras
things
aren't
like
rook
what
they're
absolutely
required
to
be
part
of
kubernetes
itself.
We
want
to
find
a
balance
between
where
we
currently
are,
where
we
bake
them
into
cops
and
it's
quite
hard
to
change
them
versus
a
world
where
you
have
to
like
figure
everything
out
yourself.
A
So
it's
nice,
if
you
think
about
the
UX
that
we
started
with
it's
nice,
that
it
was
cops
great
cluster
I
think
it's
nice,
it's
I
think
it's
nice
that
it
was
cops,
create
cluster
test
up
gates
that
local
in
the
zone
I
didn't,
if
specify
like
a
DNS
version
and
a
CNI
version
and
a
cloud
controller
manager
version
and
all
those
things.
Now
you
can
specify
network
plugins
there
today,
but
we
want
to
make
that
more.
We
will
make
it
easy
to
use
while
not
unpredictable,
while
not
necessarily
having
to
bake
everything
into
cops.
A
So
that
is
what
add-on
management
should
be
I,
think
and
that's
what
we're
trying
to
figure
out
how
to
how
to
reconcile
those
sort
of
conflicting
requirements.
I
think
we
have
time
for
Q&A
I've
raced
through
that
I
apologize,
but
shall
we
do
other
questions
if
you,
if
there
are
questions
I,
think
we
can
just
stand
up
and
I
will
repeat
the
question
or
just
shout.
A
A
A
A
We
put
in
there
so
that
the
nodes
know
what
to
configure
and
that's
about
pulling
as
much
logic
as
possible
out
of
node
up
and
into
a
more
predictable
way,
so
you
can
sort
of
see
what's
happening
and
that
that
will
help
us
to
with
the
add
ons
effort
to
make
it
more.
You
know
predictable
and
easier
to
customize.
It
would
be
we're
working
on
splitting
those
two
concepts
up
because
they
don't
logically
then
it's
convenient,
but
they
aren't
required
to
necessarily
be
to
say
they
shouldn't
be
required
to
be
the
same
bucket.
A
The
question
was:
how
do
we
deal
with
expiring,
TL
acid?
The
answer
is
what
we
can
do,
Cecilia's
with
the
the
glib
answer
is
we
said
a
10
year.
If
explorations
the
10,
not
so
glib
answer.
Is
we
I
think
we
can
now
do
a
TLS
rotation
but
honestly
I
think
so
I
think
our
TLS
rotation
right
now
is
disruptive.
We
honestly,
we
don't
yet
I
think
we're
close
in
that
you
can
you
can
today
you
can
delete
a
key,
create
a
new
one.
A
A
Yes,
the
question
was
to
envisage
drivers
age,
cops,
being
managed
being
used
to
manage
clusters
on
managed
solutions
such
as
eks
or
gke
I
I
think
that
would
be
great
I
think
it
I
think
there
are
some
differences
in
the
models
used.
So
when
we
edited
the
cluster,
there
were
a
lot
of
options
in
there
and
you
don't
even
see
like
a
lot
of
the
options
are
just
left
emitted.
If
they
are
default,
so
it
can
be
tricky
to
match
the
full
specification
into
the
things
that
eks
or
GK
or
as
yours,
managed
service
supports.
A
So
the
hope
is
that
when
we
all
manage
our
machine
through
machine
deployments,
that
that
will
be
a
you
know,
cops
will
use
a
machine
deployment
and
keep
it
and
will
use
machine
to
and
KS
will
use
it
machine
deployment
and
everyone
will
use
a
machine
deployment
and
that
might
be
a
sort
of
cops
like
experience
without
necessarily
being
requiring
you
to
use
cops.
It's
not
so
we
could
maybe
do
it
I'm
not
yet
convinced
I.
A
You're
saying,
is
it
a
so?
The
question
was:
is
our
day
to
operations
like
is
it?
Is
it
simple
for
health
management
and
process
management?
And
what
do
we
do
after
this?
And
the
answer
is
that
we
don't
really
get
as
involved
in
that
it's
more
about
the
management
of
the
cluster
itself,
so
the
upgrades
and
lifecycle
of
the
of
the
cluster
so
I
knew
a
new
cluster
version
or
a
new
qpn
S
version,
or
something
like
that
that
that
is
more
what
it
manages
and
I
think
you
know
installing
a
some
sort
of
management.
A
Your
management
tool
of
choices
is
where
we
we
sort
of
hand
that
off
so
we
don't
really
get
as
involved
in
that
I'm,
not
sure.
If
that's
the
right
thing
to
do,
but
it
seems
like
we
try
to
set
it
up
in
a
good
way
and
then
yeah
be
nice
if
it'd
be
nice,
if
some
good
add-ons
were
available
that
set
you
up
well
for
day
two,
but
we
currently
don't
really
do
that.
There's
one
at
the
back
and
then
okay.
A
Yes,
so
the
question
which
is
very
politely
put
given
the
given
the
state
of
affairs
is
essentially
like.
Currently
a
cube
version
and
a
cops
version
and
a
cube
recommends
were,
is
tied
to
the
cube
version,
kubernetes
version
and
there
can
be
a
delay
and
that
delay
can
be
large
and
can
we
make
that
shorter,
effectively?
I
think
so.
You'll
notice
that,
like
we're
on,
we
haven't
released
cops
110,
which
means
that
criminals
110
is
not
officially
supported.
A
The
cobra's
110
works
with
the
current
like
master.
We
we
e
te
test,
so
it
is
working,
but
the
what
we
said
up
to
date
up
to
like
the
beginning
of
the
year,
is
that
we
release
cops
110
when
kubernetes
110
and
all
the
CNI
providers
and
everything
is
working
and
stable,
and
we
would
be
happy
to
run
it
in
production.
What
what
we
discussed
in
the
office
hours
for
cops,
which
is
every
other
Friday,
it's
noon,
Eastern
9
a.m.
Pacific.
A
But
you
know
much
earlier
in
this
cycle
by
cops.
112
112
alpha
should
be
released
very
early
in
the
cycle.
How
about
that
we're
working
out?
So
that's
that's
sort
of
the
the
trade-off
and
the
early
release
of
alphas
and
betas,
or
how
we're
trying
to
address
that
and
give
you
the
choice
instead
of
saying.
No,
you
can't
do
that.
A
A
A
The
output,
for
example,
which
you
saw
from
the
cops
update,
wasn't
entirely
obvious
about
what
was
gonna
I
didn't
think
it
was
entirely
obvious
what
was
going
to
happen
in
terms
of
what
is
what
exact
change
will
be
making
to
that
instance.
Template
right,
the
terraform
output
is,
is
clearer
and
is
better
and
a
lot
of
the
ops
people
that
a
lot
of
people
know
how
to
read
that
output
and
are
familiar
with
it.