►
From YouTube: Kubernetes Kops Office Hours 20180608
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
A
A
So
we
were
looking
pretty
good
on
110
on
our
I
think
we
failed
to
make
progress,
but
we
haven't
fallen
further
behind.
So
we
didn't.
You
can
wear
this
on
our
on
earth.
Perpetual
next
recent
cops
sort
of
around
the
next
recent
communities,
but
we're
gonna
try
to
make
progress
on
that
at
C,
D
manager,
I
think
was
the
big
blocker
this
time
and
that,
hopefully,
will
be
better
but
anyway
hoping
to
cut
the
110
alpha
train
this
weekend.
A
So
if
there's
any
PRS
in
particular
that
you
want
to
get
in
or
that
are
important
that
I've
overlooked
or
anything
like
that,
please
mention
them
now
or
make
sure
they
are
marked,
as
110
PRS
marked
as
110,
ideally
with
blocks
next.
If
they
are
that
important
I'm,
definitely
gonna
do
a
pass
on
all
the
110
PRS
and
hopefully
get
in
anything
that
we
can
get
in
or
should
get
in
and
otherwise
I
think
Andrew.
You
had
a
query
about
the
version
of
110
yeah.
B
A
The
look
is
there
a
particular
one,
I
think.
Actually
I
was
looking
for
and
I
couldn't
find
it
so
I'm
guessing
it
just
merged.
There
was
a
PR
which
I
said
we
could
merge.
You
had
an
issue,
but
I
guess
someone
has
merged
it.
But
if
not,
please
let
me
know
and
I
will
make
sure
that
we
Peters
on
the
phone
I'm
fairly.
Certain.
A
A
D
B
B
A
Yes,
let
me
mention
so
the
what
we
have
in.
We
have
these
channels
concepts.
There
are
two
files
channels,
all
four
channels
stable.
Each
version
of
cops
can
can
either
Rep
can
recommend
a
version
of
kubernetes,
and
so
we
tend
to
put
in
the
latest
versions
of
kubernetes
as
they're
released
into
the
Alpha
Channel
first
and
then
promote
them
once
we
feel
that
they're
stable
into
the
stable
Channel.
So
typically
when
I
do
a
release.
A
F
A
A
D
D
So
I
understand
it
like
it's.
We
can't
like
it
I
think
it
should
be
recommended
just
because
I
don't
think.
First
of
all,
like
digit
ocean
is
still
alpha,
so
it's
not
gonna
I.
Don't
think
you
should.
We
should
block
versions,
because,
because
of
that,
but
also
like
I
understand
that
we
can't
force
anyone
not
to
use
those
versions
but
I
feel
like
as
long
as
the
default
is
a
version
that
works
and
it's
stable
for
everyone
that
that
would
be
great,
so
yeah
recommended
for
me,
that's
good
and
we
we
certainly
could.
A
B
B
B
D
B
Me,
let
me
put
it
a
different
way.
Sorry
I
interrupted
you,
but
let
me
I
asked
a
base
made
a
bad
statement.
Cloud
provider
as
in
cops,
do
yeah
all
right,
so
it
does
not
work
with
these
two
versions
and
cops.
Do
I
would
recommend
putting
code
into
cops
to
stop
people
from
using
the
wrong
versions
with
digitalocean
only
sure
where
else
it's
gonna
it's
going
to
create
support
tickets,
for
you
did
yeah.
A
D
C
Pronounce
my
name
is
way
both
Patil
and
I
wanted
to
introduce
myself
I
work
at
Microsoft
on
Azure
stack,
and
it
was
just
last
night
that
I
started
browsing
the
repo
and
I
think
the
problem
that
cop
solves
actually
is
very
pertinent
to
address.
So
I
was
like,
if
you,
if
you
guys,
if
I
may,
take
like
five
minutes
to
just
give
it
a
little
intro.
Okay,
so
I'm
sure
stack
is
basically
a
hybrid
cloud
solution
from
Microsoft
and
more
of
an
on-premise
solution
and.
G
C
Of
that,
as
a
mini
as
your
on
premise
and
I've
been
working
on
the
ACS
engine,
which
is
the
provisioning
of
kubernetes
on
Azure.
Having
said
that,
the
admin
persona
for,
for
you
know
deploying
managing
and
upgrading
the
cluster
is
more
of
an
in-house
thingy.
You
know
people
have
gone
overboard
in
creating
custom
scripts
to
just
do
that
behind
the
scenes,
because
it's
there
is
no
customer
actually
getting
to
you
know
do
that.
C
However,
for
Azure
stack,
it's
actually
pretty
important
that,
since
we
are
giving
it
to
a
customer,
we
need
support
an
admin
persona,
something
that,
like
a
corpse
utility
there,
you
know
once
the
customer
is
deployed,
because
the
the
cluster
on
Azure
stack.
They
want
to
manage
it
by
themselves,
and
so
for
us
for
me,
because,
specifically,
there
is
a
vested
interest
to
actually
start
contribute
contributing
towards
cops
and
I.
C
See
that
there
is
no,
as
your
support
to
begin
with
and
and
I'm
I,
understandably,
because
I
think
there's
a
lot
of
overlap
between
what
ACS
engine
does
and
what
cops
does
and
I
reached
out
to
a
few
ACS
engine.
Folks,
saying
guys,
why
are
we
not
on
cops
and
it's
like
very
low
on
their
priority
list,
given
that
there's
so
much
overlap?
Having
said
that,
for
us,
it's
actually
pretty
important
that
we
actually
get
on
on-site
now.
I
just
want
to
give
a
little
bit
of
comparison.
C
But
having
said
that,
if,
if
my
team
manages
to
bring
cops
on
board
on
Azure
stack,
it
will
be
just
a
flip
switch
to
go
back
to
us
or
just
because
they're
using
so
much
everything
latest.
We
we're
currently
to
give
you
a
bit
of
a
status
right
now
on
Azure
stack
variable
to
run
kubernetes
one
point:
seven,
eight,
nine,
with
with
ACS
engine
provisioning
it
and
we
are
just
freakin
using
the
IES
capability.
So
this
primary
compute
storage
and
a
little
bit
of
Sdn
network
from
the
azure
resource
provider.
C
Besides
computer
and
storage
and
I
started
browsing
the
repo
and
I
see
that
there
is,
you
guys,
have
embedded
the
AWS
SDK
and
Google
Cloud
SDK,
so
the
first
step
is
probably
to
embed
the
azure
CLI
SDK
and
then
at
least
start
making.
Calls
to
you
know
see
if
we
can
connect
to
storage,
create
blobs,
create
configurations
and
whatnot
and
then
sort
of
move
further
from
that
again,
I
really
I
would
I'm
not
really
too
much
prepared
for
this
I
just
figured.
There
is
a
call
happening
and
I
forget
just
I'll
log
in
and
I.
C
A
A
However,
I
can
I
think
I
think
what
you've
described
in
terms
of
these,
if
you're
able
to
try
it
out
on
AWS
or
GCP,
just
to
see
the
way
that
the
files
are
basically
stored
on
s3
or
GCS,
and
that
that
is
you
sort
of
buy
them
machines
as
they
boot
up
and
I.
Think
that's
the
case
also
for
do
right,
Andrew
yep!
So
that's
that
is
indeed
the
first
requirement
and
if
you
so,
the
a
good
place
to
start
would
be
to
write
a
sort
of
simple
VFS
implementation
for
I.
C
B
C
So
currently
this
is
on
my
own
I
was
trying
to
actually
get
a
few
people
from
ACS
engine
team,
but
because
they
said
this
is
lower
priority.
So
my
goal
right
now
is
to
really
provide
some
proof
of
concept.
That
cops
is
a
good
alternative
and
then
sort
of
take
it
to
my
management
saying
guys.
We
should
be
doing
this
and
and
then
sort
of
get
more
leverage
and
have
more
people
work
on
it.
C
A
Two
things:
if
I
may,
one
of
them
is
there's
an
effort
you
may
be
interested
in,
which
is
the
cluster
API
or
the
machines
API,
which
is
because
the
tricky
bit
is
sort
of
provisioning.
The
or
one
of
the
two
key
pieces
is
provisioning.
The
like
the
cloud
objects
right
and
the
machines
API
is
trying
to
bring
that
with
the
cluster.
It
guys
try
and
bring
that
into
the
kubernetes
api
ecosystem
and
trying
to
adopt
that,
and
so
that
that
may
be
a
good
way
to
proceed,
but
sort
of
maybe
a
bit
more
medium
term.
A
A
shorter
term
tactic
is
you'll,
see
when
you,
when
you
get
into
the
code,
and
maybe
we
should
schedule
a
call
to
talk
through
a
tour
abusing
like
that,
but
you'll
see
that
cops,
creates
a
model
and
then
has
tasks
which
are
similar
to
terraform
tennis.
Ok,
those
tasks
can
either
execute
directly
in
cops
or
it
can
generate
text
it
can
generate
terraform
on
AWS
can
generate
CloudFormation.
A
C
There
was
one
vector
that
I
was
thinking
which
was
to
not
do
a
whole
lot
of
reimplementation
and
sort
of
use.
Acs
engine
again
I
I
was
trying
to
get
familiar
to,
but
I
definitely
see
the
value
and
I
think
that
will
be
a
easier
sell
for
me.
Rather
than
saying
here's
a
completely
right
and
yeah.
B
It's
gonna
be
kind
of
it's
gonna,
be
kind
of
challenging
to
integrate
ACS
engine
into
the
cops
model.
We
use
an
async
task
model.
I,
don't
think
it's
you
know
realistically,
like
Andrew,
how
many
total
hours
can
you
say
like?
Did
it
take
you
100
coding
hours
or
80
coding
hours
to
knocked
out.
Do
what
do
you
think
yeah,
so
Huck's,
safe.
D
Because
it
was
kind
of
like
I
know
it's
over
like
over
like
a
year,
yeah
like
I
was
kind
of
like
it
wasn't
like
free
time,
so
I
just
kind
of
little
bits
at
a
time
when
I
ever
had
time
so
I
can't
say
but
I
just
linked
in
the
chat.
The
umbrella
issue
that
so
there's
a
digitalocean
support,
umbrella
issue
and
I
pretty
much
linked
that
issue
to
all
the
work
I've
done.
So
you
can
kind
of
resolve
some.
C
I
think
I
probably
am
varies
in
a
similar
boat
as
Andrew.
It
seems
like
and
I
definitely
want
to.
You
know,
get
more
leverage
by
just
proving
something
real,
quick
and
sort
of
showing
that
value
and
really
for
us
it's
if
the
customer
there
is
a
customer
asked
to
do
this,
then
I
will
get
enough
leverage.
The
only
reason
is
right
now,
since
we
have
an
alternative
of
ACS
engine,
there
is
not
a
much
of
leverage,
but
it
to
me.
A
I
think
I
think
rendering
engine
works
the
same
way
as
Tara.
For
me,
I
think
rendering
to
a
chess
engine
may
be
possible
and
maybe
well
certainly
easier
than
writing
the
directs
yeah,
the
giraffe
exclusion,
which
is
basically
we
implementing
ACS
engine
and
that
might
also
make
the
ACS
engine
team
a
little
more
amenable
to
like
the
approach
right.
C
Yeah,
well,
there
is
good
and
bad
sizing.
We
don't
I.
Azure
stack
doesn't
have
to
support
the
entire
as
your
story
and
vice
versa,
and
which
is
why
it's
I
think
both
can
still
live
on,
there's
aks,
which
actually
has
a
heavy
dependency
on
ACS
engine,
and
so
it's
not
like
if
I.
If
we
had
an
alternative
ses
and
it
will
render
worthless
it's
just
that,
for
they
don't
have
enough.
Motivation
to
you
know,
go
overboard
on
ACS
engine
just
for
as
your
stack,
so
I
am
looking
out
for
Azure
stack
from
both
fronts.
C
A
A
A
That
yeah
I
think
I
think
we
need
to
support
stretch
and
Amazon
Linux,
I
think
yeah.
Hopefully,
ideally
one
day
we
would
everyone
would
be
using
an
image
we
don't
have
to
support,
but
I
was
in
Linux
is
CentOS
and
the
image
we
bilities
Debian
based.
So
that's
a
pretty
big
switch
for
people
to
you
know
it's
pretty
big
change
to
force
on
people.
So
let's
take
this
in
a
measured
way.
Do.
B
A
B
At
least
having
problems
with
the
is
it
the
NVE
ball,
the
law
controller
nvme
hi?
Is
anybody
running
like
em,
fives
or
C
fives
on
the
call
I'm
running
in
them
for
the
nodes,
not
the
Masters?
Okay,
and
are
you
having
any
problems
with
stretching
I'm
using
quarter
less?
You
don't
count
them
dance.
Of
course,
you're
using
Corollas.
A
H
G
A
I
get
locked
into
the
VPN.
There
is
a
note
in
that
in
the
channels
that
we
we
break
tests.
We
can't
do
it
because
the
tests
right
now
so
that
that's
actually
the
bigger
it's
a
sort
of
minor
hurdle.
It's
not
even
a
real,
it's
not
a
real
test
failure,
but
it's
the
way
that
the
that
we
pause,
something
to
do
with
connection
tracking
or
something
has
changed
and
like
different
kernel
versions.
We
just
need
to
fix
this.
Hey
Rodrigo,
you
running,
see
5m5.
B
A
B
A
Hope
1:11
can
be
a
shorter
cycle,
in
other
words,
not
a
three-month
cycle,
so
we
can
actually
advance
a
little
bit
on
the
kubernetes
release,
we'll
see
if
we,
if
we
achieve
that,
but
that
is
my
hope.
The
mCP
manager
is
supposed
to
the
default.
So
that
will
be
a
big
require.
A
lot
of
dte
testing
for.
A
For
it
for
one
ten,
its
opt-in
for
one
eleven,
it
will
be
the
default,
and
does
that
include
birds
from
two
to
three
upgrades
will
will
all
right
so
once
we
make
it
the
default,
then
we
can
default
at
CD.
Three
upgrades
will
be
possible.
It
may
not,
depending
on
the
status
depend
on
what
you
de
testing
of
the
upgrades.
It
may
not
necessarily
be
as.
A
E
Kind
of
have
a
quick
one:
I
want
to
get
the
name
tags
on
the
auto
scaling
group
name,
us
she's,
not
propagate,
so
I
kind
of
need
to
create
a
secondary
nametag
specific
for
cops
that
will
propagate
to
the
instances
so
that
we
can
do
this
discovery.
I
actually
started
thinking
into
that
and
kind
of
pick,
my
head
against
the
wall,
but
hoping
you
guys
can
slack
for
more
details
on
those.
You
know
I,
don't
understand,
maybe
open
an
issue
making
discussing
issues.
E
So
we
have
a
process
that
goes
through
all
instances
and
make
sure
that
who
they
belong
to
and
blah
blah
and
part
of
that
is
they
actually
get
their
own.
Naming
like
each
instance
has
its
own
specific
name
and
because
we
have
our
own
baking
process
for
the
instances
that
automatically
happens
and
whatnot
unfortunately
cops
breaks
that
because
we
propagate
the
name
tags
for
the
issues
and
I
think
I'm,
not
the
only
one.
That's
getting
this
issue
I've
seen
a
couple
of
their
people
open
tickets
on
cops
for
the
same
similar
issues
of
numerous.
So.
E
E
E
A
B
F
F
G
D
A
C
F
A
B
B
F
F
F
B
A
A
F
B
C
A
G
Justin
I'm
curious,
you
mentioned
earlier
the
Machine
API.
Do
you
branch
up
in
with
that
effort
underway,
or
is
that
just
something?
That's
you
daydreaming
about
right
now
or
what's
the
story
so.
A
A
There
is
a
very
old
branch
of
cop
somewhere
that
uses
it,
but
it
it's
pretty
old
at
this
point,
so
I
need
to
rebase
and
get
that
going.
It's
not
really
the
the
API
itself
isn't
fully
even
alpha
yet,
and
the
cops
branch
was
sort
of
from
earlier.
If
you're
interested
in
working
on
I
can,
if
you
putting
on
Zak
I,
can
try
to
find
it
for
you,
I'm.
G
A
The
there
the
mission
the
cluster
API
has
to
sort
of
portions
there's
a
cluster
object
and
a
machine
object
to
me.
The
object
is
very
compelling.
The
Machine
object,
they
I
sort
of
think
up.
Think
of
that
as
like
a
pod
and
then
they
build
on
top
machine
sets
and
machine
deployments,
and
the
idea
is
that
then.
Currently
we
have
some
logic
in
the
you
know
cop
CLI,
to
do
a
rolling
update,
for
example,
and
there's
some
the
autoscaler.
A
The
cluster
autoscaler
has
some
logic
to
look
at
how
auto
scaling
groups
work
and
figure
it
out
from
there.
The
the
big
idea
is,
if
we
put
those
the
machine
concept
into
the
kubernetes
api
that
you
can
drive
everything
through
Vinay's
communities,
api,
you
can
do
a
rolling
update
through
the
conveyance.
If
here
you
can
rescale
through
the
kubernetes
api
cluster,
auto
spoiler,
so
hopefully
smart
people,
but
other
things
you
can
do
through
this
API
as
well.
A
So
essentially
it
is
putting
the
cops
instance
group
concept
into
your
cluster
so
that
you
can
manipulate
on-the-fly,
but
it's
also
a
standard,
a
communities,
standard
objects
that
will
work
across
all
old
tooling.
We
use
cops
or,
as
your
or
you
know,
AWS
sir
I
don't
see
any
any
tuning
because
I
don't
read.
G
A
A
There's
definitely
interesting
work
on
the
the
bringing
up
and
specifying
a
machine
in
the
in
the
API
is
not
a
terribly
hard.
What's
a
little
bit
more
complicated
as
specifying
the
configuration
of
that
machine,
the
machines
API
currently
basically
has
a
big
extension
block
in
there,
where
you
can
put
in
whatever
provider
specific
stuff
you
want.
So
we
could
effectively
embed
the
cops
instance
group
in
a
machine.
That's
one
way
to
do
it:
okay,
it's
that
sort
of
the
fuzziness.
So
it's
it's
relatively
straightforward
to
implement
the
basics,
but
yeah.
G
A
B
A
Like
we've
learned
our
lesson
about
adopting
things
before
they
are
beta
at
least
b16
annotations
that
we
use
their
alpha.
That
was
painful,
and
so
the
we're
not
gonna
we're
not
gonna
force
anyone
to
use
it,
but
it
would
be
great
to
have
like,
as
Andrew
said,
you
know,
the
functionality
of
of
auto-scaling
groups
were
the
equivalent
on
any
cloud
or
on
bare
metal.
A
So
you
know,
we've
talked
for
a
long
time
about
getting
bare
metal
support
up
and
running,
and
one
of
the
problems
there
is
obstructing
out
the
the
hardware
as
it
were,
for
bare
metal
and
the
machines.
Api
is
a
nice
way
to
do
that,
so
that
was
where
I
was
starting
with
the
machines.
Api
is
thinking
well,
if
we
do
it
for
very
metal
sure
that
will
be.
That
would
be
alpha
in
cops.
B
B
D
A
B
H
B
B
A
In
theory,
it's
just
everything
we
put
in.
We
have
to
make
sure
we
is
supportable
in
the
machines
API
which
it
should
be
because,
conversely,
the
machines
API
wants
to
so
we
have
a
real-world
user
that
has
submitted
a
PR
that
adds
forward
functionality
that
other
people
say
is
useful
as
well,
and
the
machines
API
also
wants
to
have
the
same
level
of
support
for
real-world
use
cases
to
do
so.
B
I
mean
the
main
problem
puter
with
that
is
that
he's
duplic
making
new
instance
groups
objects,
and
we
don't
want
that.
We
kind
of
want
a
history
like
with
replica
sets
with
employments
right.
You
know
how
you
can
roll
back
replica
set
or
roll
forward
replicas
I'm.
Sorry
replica
set
history
is
tracked
within
the
deployment
object.
That
makes
sense
so
that
we
need
to
have
a
new
ASG,
but
a
not
a
new
instance
group.
A
Yeah
so
and
when
and
if
the
machines
API
arrives,
then
it
will
have
that
three-tier,
notion,
I,
guess,
I'm
guessing
a
year,
it's
not
even
alpha
yet
right,
there's
a
like
a
circular
dependency
right,
the
we
need
to
we
need
to
get
into
real-world
usage
before
it
can
and
make
sure
that
it
it
works
before
we
want
to
declare
it's
stable,
but
we
might
take.
We
also
don't
want
to
put
it
into
cops
until
it's
locked
or
more
locks.
A
So
that's
why
I
think
the
way
to
break
that
might
be
to
say
well
bare
metal
or
something
like
that
right
or
some
other
platform
that
we
currently
don't
support.
The
ward
wants
to
adopt
it
like
sto,
chooses
to
adopt
it
right,
but
we're
not
gonna
force
it
on
DL
and
like
we
can.
We
can
add
support
for
a
brand
new
platform,
yeah.
D
B
A
B
A
Think
it'd
be
great
to
have
cops
proven
out.
I
think
you
know,
I've
done
some
I've
done
some
public
like
PLC's,
that
do
elements
of
that
and
trying
to
figure
out
that
it's
mostly
about
figuring
out
whether
whether
we
clicked
in
the
idea
well
figure
out
what
what
approach
we
should
take
and
circling
the
idea
and
how
customized
fits
into
that
customized
is
a
nice
way
to
do
patches,
I,
guess,
you've,
just
read
them
or
overlays
on
a
manifest.
A
It
sort
of
solves
the
problem
that
you
know
particularly
there's,
always
a
user
out
there
somewhere
in
the
world
that
wants
to
customize
every
field
and
if
you
allow
a,
if
you
have
a
template
model,
you
end
up
with
every
field
in
your
template
being
another
variable
declaration.
You
know
it's
like
a
different
language
and
so
patches
and
manifests
the
patches
and
overlays.
Allow
you
to
use
the
same
language
to
express
those
changes
which
is.
A
B
The
Justin's
working
with
basically
replacing
channels
or
two
channels,
yeah
our
channels,
add-ons
manager
with
a
kubernetes
level,
add-ons
manager
which
would
take
care
or
basically
be
open
source
supported.
So
we
would
be
able
to
get
rid
of
parts
of
channels
for
stuff
like
DNS
CNI.
Anything
else
that's
installed
so.
A
I'm
really
excited
about
frankly
Justin
but
I
mean
just
set
expectations
that
I've
been
looking
at
this
problem
for
a
long
time
even
before
I
joined
Google.
So,
yes,
it
is
certainly
it
is
great
to
be
at
Google
and
have
more
time
to
work
on
it,
but
there's
also
more
requirements
now
than
just
going
and
writing
something
cool
I.
C
A
A
Cx
you'll
be
able
to
opt
into
at
CD.
The
plan
is
over
to
opt
in
to
exit
city
manager
in
1.10
and
what
that
will
allow
you
to
do
if
you
choose
to
opt
in
which
is
not
a
sort
of
good
idea.
But
if
you
choose
to
opt
in
which
you
know
fine
for
a
test
cluster
that
you
can
then
move
from
between
STD
versions
and
change.
The
size
of
your
at
CD
cluster
and
your
ICD
cluster
size
is
currently
tied
to
your
number
of
masters.
A
So
I'd
certainly
been
a
common
request
for
people
to
go
from
at
least
one
to
three
masters
is
very
common
and
then
people
that
decide
their
masters
need
to
be.
They
need
more
masters
than
that
for
performance
reasons
to
go
to
five
or
some
like
that.
I
would
typically
recommend
scaling
up
rather
than
scaling
out
beyond
three,
but
each
to
their
own
as
a
perk.