►
From YouTube: Kubernetes Kops Office Hours 20180302
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:
it
is
Friday
March
2nd.
This
is
cops
office
hours
where
we
discuss
all
things.
Cops.
Any
question
answered
any
topic
discussed.
We
do
have
a
couple
of
items
on
the
agenda
if
you
haven't
yet
done
so,
please,
to
put
your
name
on
the
agenda.
I
will
be
pasted
the
link
into
the
agenda
for
anyone
joining
sorry
to
chat
for
anyone
joining
now,
and
otherwise
we
can
just
get
right
into
it.
A
B
That
PR
has
already
been
cherry-picked
that
I
put
in
for
the
machinery
rules
into
the
1
9
release
branch
I
was
sweet
so
that
that
means
which
one
so
we
so
the
API
machinery
rules
were
messed
up
or
let
me
let
me
back
up
the.
What
are
they
the?
Let
me
look
at
the
PR,
so
I
can't
say
exactly
what
was
done
when
I
was
putting
in
the
tests
so
that
we
can
validate
API
machinery.
I
noticed
that
API
machinery
was
not
correct.
B
A
Well,
I
mean
I,
think
we'll
do
an
alpha,
I'm
gonna.
Do
it
like
this
weekend,
if
I
think
there's
like
four
or
five
issues
that
came
up
which
are
relatively
simple?
What
effects
I
don't
like
to
try
to
attribute
importance
to
them
like
the
there
was
one
about
where
we
something
about
gnats
and
elastic
IPS
I
noticed
like
subnets,
are
always
DHCP
options.
I
did
gateways
and
round
tables
are
always
dirty
on
shared
and
tagged.
A
So
some
things
were
just
sort
of
simple
ish
to
fix
that
the
other
that
Meriden
alpha
or
the
other
thing
that
merits
an
offer
is
the
ami.
So
building
an
ami
and
putting
into
alpha
2
is
on
my
topic
for
this
weekend
and
yeah,
and
so
they
nearly
build
the
API
machinery
that
we
get
and
the
I
think
you
know
I
think
one
of
the
things
I
wanted
to
talk
about
is
how,
if
you
I
wish
we've
tried
to
in
the
past,
so
cops
urbanized.
A
My
point
now
is
released,
I
guess
in
December
ish,
and
we
are
now
in
March
just
and
so
there's
a
big
delay
between
the
kubernetes
release
and
becomes
release
them.
So,
to
an
extent
that's
been
deliberate.
In
the
past,
we've
tried
to
say
that
you
know
when
you,
when
a
cops
release
comes
out,
we
consider
it.
It
should
work.
A
Is
this
sort
of
a
fun
way
to
put
it,
but
we
consider
it
like
ready
to
go
into
production,
or
at
least
you
know
your
station
type
environment?
We
don't
we
don't
we
don't
do
it
with
issues,
we
don't
do
it
with
one
I
know
of
kubernetes.
One
another
option
we
could
do
is
so
that's
uncomfortable
people,
but
so
that's
uncomfortable
people.
That
then,
are
like
well
I
want
to
use
criminals.
1-9
I
have
a
deaf
cluster
I.
You
know,
I
accept
the
risks
that
thing.
A
C
A
B
I
would
rather
release
with
cops
one
ten
one
when
you
mean
with
criminals,
110
I
mean
what
does
that
mean,
because
any
critical
bug
fixes
that
they
fix
in
110
that
they
find
in
110
should
be
in
110.
One
like
I
did
with
what
happened
with
the
one
which
was
at
1/8,
where
they
opened
up
off
once
yeah
I
mean
I,
think
I
that
happens,
I
mean
it
happened.
I
just
don't
want
our
users,
no
matter
what
you
tell
people
they'll
be
like.
Well,
no
you
released
it.
So
it's
you're
able
to
run
in
production.
B
A
B
A
Release
yeah
I
was
thinking.
We
can
do
it
a
couple
of
weeks
after
and
we
basically
do
it
more
mechanically,
and
we
say,
let's
pick
a
number,
let's
say
Corrigan,
we
say
like
two
weeks
after
the
release.
We
will
do
our
one
1000
or
when
we
do
our
our
first
release
of
110
I,
don't
number
it.
We
do.
Our
first
reasons
cops
110
and
it
we.
A
B
A
C
My
biggest
thing
is
more
releases
to
be
honest,
we
can
tag
it
and
call
it
whatever
we
want,
so
my
biggest
thing
would
be
even
if
it's
just
in
my
dev
clusters,
if
I
could
just
try
different
versions
here
and
there
I
I,
think
I.
Think
there
you're
right
both
of
you
were
right
that,
like
people
come
into
the
channel
like
the
day
after
the
release
and
they're
like
I'm
running
cops,
you
know
1/9
with
you
know,
with
kubernetes
110
and
you're
like
wait.
A
second
1
9
is
not
even
out
yet.
C
How
are
you
doing
any
of
this?
You
think
you're
doing
stuff
you're
not
doing
so
I.
Don't
know
you
know
the
more
people,
the
more
alphas
more
betas
make
it
easier
for
Justin
to
cut
them,
probably
as
all
valid
things
beginning
and
called
it,
you
know,
call
them
whatever
we
call
them.
That's
that's
my
main
thoughts.
D
Yeah
echo
that
I
think
guarding
the
not
wanting
to
be
a
gatekeeper
too
much
to
kubernetes,
like
we
don't
do
it
with
our
projects.
We
tend
to
lag
behind
too
much
actually,
but
I
could
see
companies
that
want
that
are
using
cops
to
rule
out
kubernetes
that
want
to
get
the
latest
version
out
as
soon
as
possible,
so
they
can
just
start
testing
it
right
and
have
a
couple
months
of
bacon
on
their
side.
D
E
B
A
Well,
not
necessarily
don't
one,
but
maybe
we
we
do
exercise
some
judgment
in
terms
of
the
GA,
and
we
can
review
that
in
three
months:
six
months,
whatever
it
is,
but
I
like
the
idea
of
cutting
alphas
and
betas
and
got
a
naming
scheme
around
that
early,
so
that
we
are
both.
We
are
both
clear
with
people
that
that
it
is
not
entirely
problem-free,
necessarily,
but
also
that
they
are
able
to
do
it
in
bacon
themselves.
A
C
I
think
my
my
biggest
thing
would
be
as
long
as
we
provide
clarity
on
what
alpha
and
beta
mean.
Then
it
solves
a
lot
of
that
issue.
Alpha
maybe
alpha
is
just
alright.
We
have
a
release.
This
might
work
with
you
know.
Kubernetes,
one,
nine
or
one
10
beta
might
be.
Okay,
we're
fated
might
be
what
we
call
alpha
now
in
a
way
and
then
once
it's
really
released,
you
know,
then
we're
really
saying
it's
prod,
ready,
I
I
really
actually
like
that
that,
like
I'm
gonna
I,
we
lag
behind
as
well.
C
B
What
you
said
very
prudent
Mike,
we
just
need
more
releases.
I've
had
an
email
threat
going
with
a
couple
of
or
I
started
with
Tim
and
Phil
Phil's,
one
of
the
steering
committee
member
Tim,
and
you
guys
probably
arena
start
an
email
with
him
again
about
how
to
get
an
automated
release
process
for
us.
So
that
takes
work
off
of
Justin's
hands
and
we
can
do
this
and
have
a
release,
but
that
it's
probably
slow
roll
to
get
it
that
automated.
B
A
Mostly
automated
process
on
my
site,
mostly
the
challenge-
is
knowing
when
to
push
the
button
right
and
being
like.
Can
we
get
to
a
point
where
we
believe
that
there's
nothing
outrageous
in
there,
but
we've
merged
in
at
least
all
the
PRS
that
are
ready,
merge,
ready
and,
like
we've
done
a
good
pass
of
getting
in
the
fears
that
emerge
ready
and
not
like
ongoing
conversations,
but
I
like
I,
think
we
have
an
actual
item.
I.
Think
that's
good
I'd
love
to
get
to
an
automated
release.
I
think
there.
A
There
are
people
challenges
as
what
a
second
ago
challenges,
but
I
think
I
mean
what
this
could
mean
is
it
means
that
we
start
110
alpha
almost
now,
which
ie
will
be
confusing
for
me,
but
I
well,
I
will
think
about
what
that
actually
means
and
whether
actually
that's
fine
so
I
it
might
be
possible.
It
will
be
a
weird
messaging
thing,
and
so
maybe
maybe
as
soon
as
we
do
cups,
one
I
know
beta
that
we
do
cups
110
alpha.
That
might
be
a
less
crazy
state
of
affairs,
but.
A
C
B
B
A
Releases,
yes,
so
we
will
have
cut
an
alpha
at
this
point
like
there
is
technically
a
1/9
branch,
we're
likely
to
do
10,
20
30
POS
before
before
release
I'd,
guess
right
and
those
will
be
both
merged
onto
the
1/9
branch.
Well,
the
immersion
to
master
and
then
will
I
will
typically
like
fast-forward
merge
onto
the
1/9
branch.
A
B
A
If
it's
an
important
feature,
then
we
will-
and
we
have
enough
time
until
the
next
release-
and
we
will
like
I
mean
everybody
says,
offers
like
from
day
two
right
or
from
day
one
right.
They
stay
released.
The
first
alpha
build
or
something.
F
In
the
last
meeting
that
we
had
right
like
last
Friday,
we
spoke
about
Amazon
Linux
being
used
as
an
ami
or
and
then
we
wouldn't
have
to
worry
about
spectra
and
meltdown
and
certain
components
so
does
cops.
One
night
depend
on
things
like
the
am
I
being
updated
and
those
things
or
do
we
release
them
in
a
in
parallel,
but
they
are
not
related
at
all.
A
Would
I
think
that's
a
great
idea:
I
mean
I.
Examining
the
scope
of
what
we
do
is
is
exactly
we
should
be
doing
and
I
I
love.
The
idea
of
using
ours
and
Linux
I
think
we
should
start
by
to
an
extent.
So
we
had
to
do
it
am
I
in
the
past
because
of
just
the
timelines
of
releases
and
things
like
that.
That's
less
true!
Now
the
sort
of
you
know
the
the
kernels
that
ship
in
the
main
distros
are
now
good
enough
that
that
we
can
now
run
them,
which
is
good
and
I.
A
Would
love
to
get
on?
To
you
know,
say:
Debian
stable
is
good
enough.
Amazon
Linux
is
good
enough
and
all
these
sort
of
things
which
were
not
far
off
I
would
love
to.
Yes,
we
should
definitely
support
Amazon
Linux.
That
would
be
great,
then
also
we
would
might
be
able
to
say
you
know
if
you
want
to
use
ena
Amazon
Linux
is
great
like
that.
Would
work
that
would
be
great
I.
A
Think
that's
a
good
idea
and
the
so
I
actually
put
that
in
the
agenda
is
a
good,
a
good
item,
the
other
one
that
I
causes
me.
A
lot
of
strife
is
network
drivers
network
manifests,
and
we
are
in
an
unfortunate
situation
at
the
moment
where
a
version
of
cops
embeds
the
network
manifests,
and
so,
if
calico
or
whatever
it
is
I
want
to
pick
up
a
flannel
decides
they
have
a
new
network
provider
version
that
fits
an
important
bug.
A
The
only
way
to
get
that
out
of
student
cops
release,
which
is
a
fairly
bad
position
for
us
to
be
in
not
least
because
everyone
has
to
update
their
no
has
to
update
their
cops
releases.
So
that's
a
little
crazy
and
I'm
hoping
we
can
do
the
long-promised
add-on
manager,
which
actually
we
saw
some
progress
on
people
like
suggesting
adding
manifests
crime.
We
got
that
into
the
Alpha
last
minute,
dropping
and
I'm
hoping
to
look
at
whether
we
can
actually
move
our
manifests
in
some
way.
A
Such
that
you'll
be
able
to
specify,
for
example,
that
new
that
new
manifests
for
flannel,
which
would
be
they
could
be.
What
you're
saying
right
so
we're
saying
in
the
same
way
that
an
ami
and
Imus
Analytics
ami
is
maintained
by
someone
else
they
did
by
someone
else.
You
reference
it
here,
we
say
panel
go
get
the
you
know
up,
they
took
your
flannel
and
that's
not
really
our
concern
so
Amazon
the
nexus
updated
finalized
up.
They
did
and
or
more
pointers
and
less
of
a
bottleneck
or.
A
A
A
A
B
So
hold
on
hold
on
I
wasn't
quite
done
yet.
I
I
understand
your
concern
about
the
CNI
providers
baked
in.
If
we
release
more
often
I
think
having
those
manifests,
updated,
faster,
I,
don't
think
is
going
to
be
as
much
a
challenge
and
users
are
able
to
override
the
manifest
that
we
have
as
add-ons
already
I.
Believe
that
what
do
you
mean
you
just
install
a
new
manifest
you
update
it,
it's
not
going
to
downgrade
the
manifest
if
the
version,
if
that
on
number,
is
greater.
True.
A
B
A
Have
a
cop
we
still
have
to
do
a
cop's
update
and
our
users
have
to
update
cops,
which
is
not
you
know
it
becomes
problematic,
I.
Think
I,
don't
know
how
people
feel
about
like
right.
Now,
like
the
good
side
of
us
not
releasing
cops.
Very
often,
is
you
don't
necessary,
have
to
update
cops
that
often
right
where's
now
we're
talking
about?
You
would
have
to
update
cops
weekly
type
thing.
B
A
Third,
the
third
main
topic
item
was
the
machines
API,
which
is
sort
of
the
third
main
theme,
and
the
machines
API
is
essentially
bringing
bringing
of
the
notion
of
a
launching
a
machine
as
a
into
the
API
machinery
into
the
community
safe
of
machinery,
so
that
you
can
specify
you
can
launch
a
machine
by
creating
an
API
object
and
it
joins
your
cluster
and
then
on
top
of
that
primitive
they.
So
that's
like
a
pod
and
then
on
top
of
that
primitive.
A
They
build
two
additional
primitives,
which
are
machine
sets
and
machine
deployments
which
are
like
replica,
sets
and
deployments
the
idea
being
that
then,
first
of
all
machine
sets
and
machine
deployments
would
be
shared
amongst
all
the
various
tooling.
So
a
machine
said,
the
machine
deployment
would
work
the
same
way
on
gke
as
on
cops
as
on
dks,
as
on
aks
as
on
their
metal
as
I'm
anything
infantry,
and
it
would
do
the
correct
upgrade
procedure.
All
that
sort
of
one
saying
the
other.
A
The
other
thing
is,
it
would
then
be
better
integrated
with
tooling,
like
cluster
autoscaler,
so
there
are
so
that
there
wouldn't
be
this
like
disconnect.
I
guess
between,
like
which
fields
are
owned
by
cops
and
which
fields
around
by
the
autoscaler
type.
In
that
it's
it's
a
big
thing,
it
seems
to
be
going
well,
it
looks
like
a
fairly
yeah.
It
looks
like
it's
going
well
and
it
works
well
and
I
have
like
some
prototypes
of,
like
you
know,
bring
out
machines
on
VMware
and
things
like
that.
A
So
it's
it's
certainly
impactful
and
like
we
have.
It
duplicates
a
lot
of
the
logic
we
have
in
terms
of
our
rolling
updates,
and
we
want
to
be
sure
that
we
are
coordinating
with
it
and
likely.
We
will
end
up
contributing
a
lot
of
our
logic
to
thee.
Certainly
our
requirements
to
the
machines
API.
So
that's
I,
think
you
know
I
think
that's
the
other
item
that
I
I
think
you're
like
most
important
on
our
from
my.
A
A
miles
110
yeah
we're
further
roadmap
yeah,
that's
a
great
like
other
any
good
topics
that
people
think
you
know
cops
should
generally
be
addressing.
I
have
another
one
which
I
didn't
put
in
there,
which
which
is
like
network
splitting
out
networking,
not
a
different
project
but
like
into
a
different
object,
so
that
it's
easier
to
share
and
maybe
more
like
better
to
find
where
the
boundary
is
between
networking
and
your
compute
type
things,
which
I
mean
the
machines
API.
A
A
The
NCD
one
I
think
I
wanted
to.
If
people
haven't
read
it,
this
will
likely
start
happening
pretty
soon.
The
we
have
to
get
everyone
at
city,
3
before
city
4
comes
out
the
we
have
to
make
sure
that
so
the
I
wrote
an
Etsy
team
manager
I
think
it's
called
yeah
entity
manager,
which
basically
is
a
sort
of
like
the
sed
operator
from
core
OS,
but
doesn't
rely
on
a
central
controller,
so
it
can
work.
It
works
a
little.
It's
a
little
easier
to
understand
in
terms
of
bootstrapping
type
scenarios
or
failure
type
scenarios.
A
The
it
defines
a
backup
strategy
and
essentially
I,
want
to
introduce
very
soon
isolate
out
a
piece
of
that
which
may
see
just
backs
up
EDD
in
that
backup
format
and
very
soon
I
want
to
get
that
turned
on
I.
Think
I
think
it's
already
optional.
I
think
you
can
turn
it
on.
It's
not
very
documented
how
but
I
want
to
get
it
documented
and
easy
to
turn
on
like
a
flag
and
then
on
by
default,
the
idea
being
that
once
you've
back
up
into
that
format,
you
can
restore
from
it.
A
E
A
A
Was
it
didn't
the
it's
in
the
roadmap
anyway,
but
the
ICD
manager
is
here
and
it
does
it
automates
the
restore
it
or
make
the
backups
you
don't?
If
you
use
sed
manager,
you
don't
miss
the
restore
I,
don't
have
a
standalone
restore
component
yet,
but
that
could
be
interesting
as
well
for
people
that
want.
Then.
E
E
A
B
To
get
a
little
bit
background
first
time
what
this,
what
the
topic
is
well,
basically,
there
are
multiple
users-
or
this
has
been
a
long
long
request
from
a
lot
of
different
folks
for
cops
not
to
create
in
an
inner
gate,
Internet
gateway
and
not
update
the
route
table.
Or
this
was
it
the
route
table
in
egress
value
for
subnet.
So
we
don't
add
a
route
task
and
we
don't
create
the
Internet
gateway.
So
those
two
things
you're
teeing
me
up
here.
B
B
B
A
I
was
thinking
would
create
if,
if
there
was
a
subnet
that
that
was
not
marked
but
needed
an
in
that
gateway,
then
we
would
provide.
We
would
create
it
if
we,
if
there
was
no
michi
creative,
you
would
not
created
so
essentially
provide
it
or
next,
whatever
name.
This
is
means.
Don't
do
this
I
want
to
do
it
in
a
gateway.
I
want
to
do
a
VPN
gateway.
B
I
want
to
do
it.
I,
gotcha,
I,
gotcha,
I,
understand
what
you're
saying.
I
got
a
question,
though,
because
we
don't
have
any
other
object,
I,
think
or
I.
Don't
think
we
have
any
other
object
like
that:
AWS
object
that
we
don't
create
because
we
don't
need
it.
So
what
we're
doing
is
we're
taking
the
API
we're
putting
in
business
logic
and
not
being
explicit
in
the
API
that
we
don't
require
this.
B
B
B
A
It
feels
to
me,
like
you,
would
want
a
subnet
where
you
basically
say
on
a
per
segment
level.
We
don't
create
things,
we
don't.
We
don't
deal
with
your
networking
for
this,
something
that
you're
gonna
take
care
of.
That
subnet
I
feels
like
that.
Would
cover
a
lot
of
the
use
cases
and
I,
don't
think
I,
don't
know
why
we
would
create
a
gateway
if
we
didn't
need
it
I
think
Rodrigo.
Do
you
guys.
B
B
B
A
To
me,
I
I
feel
like
the
only
reason
we
don't
have
that
logic
right
now
is
because
you
always
need
an
IG
w
today,
because
even
when
you're
using
that
gateways,
you
actually
still
need
an
IG
w.
You
don't
need
you
don't
need,
always
need
an
IG
w
today.
For
that,
but
the
configurations
we
define
you
do
yes,
this
is
a
feature
we
need.
A
A
B
A
Anyway,
I
feel
like
it
would
be
a
we
cannot.
We
can
look
at
this
again.
I
feel
like
it
would
be
nice
to
have
something.
That's
more
just
like
leave
my
whole
subnet
alone.
So
let's
look
at
what's
left
other
than
if
I
just
said
to
you
suggest
as
never
required
gateway.
If
we
send
that
required
gateway,
let's
find
out
what
additional
changes
are
made
to
the
subnet,
etc.
Yeah.
B
Maybe
we
can
not.
Everybody
go,
go
and
work
this
out,
but
you
know
take
this
offline
but
you're
a
PA
he's
not
making
sense
to
me
because
I'm
looking
at
the
route,
the
route
is
the
public
route
table
and
the
Internet
gateway,
which
has
nothing
to
do
with
the
private
networks
and
has
nothing
to
do
with
an
at
gateway.
A
Let's
look
at
the
terraform
output
when
this
field
field
flips?
How
about
that
I?
Don't
know
if
there's
no
way
to
do
a
test
on
this,
which
is
unfortunate.
Why
don't
you
flip
temporarily
flip
one
of
the
one
of
the
other
other
integration
tests
so
that
we
can
see
the
effect
on
the
terraform
output?
I'll
just
add
this
to
the
complex
integration
test.
Sure
yeah,
because
then
we'll
see
exactly
what
changes
I
mean
I,
just
don't
know
if
the
complex
actually
already
uses
one
of
these
other
ones
or.
B
Yeah
I'll
find
one
of
the
ones
that
uses
that
or
doesn't
know
so
this
would
be
used
with
NAT
with
or
without
okay.
The
use
case
that
I
have
is
public
networked.
We
don't
have
that
in
the
at
the
bank.
We
don't
use
NAT
gateways,
we
just
don't
want
to
create
a
9gw.
Everything
is
public
networks,
but
it's
just
it's
on
a
private
IP
space
and
there's
no
public
IPS
on
the
instances.
Then
everything
goes
back
through
Direct
Connect
check
act.
B
B
Now
I
just
wanted
to
say
thanks
real
quickly
to
Robin,
Mike
and
Rodrigo.
You
guys
are
helping
out
great
with
PRS,
appreciate
it
there's
a
month
left
in
the
mentoring
program,
so
it's
almost
done
and
I'm
yeah
still
assigning
them
to
them
and
get
them
knocked
out.
It's
great
to
have
you
guys
review
it
first.
Thank
you
so
much.
It's
really
helpful.