►
From YouTube: Kubernetes kops office hours 20201001
Description
Recording of the kops office hours meeting held on 20201001
A
Hello,
everyone
and
welcome
to
out
of
band
cops
meeting
to
kick
off.
Hacktoberfest
we
are
cops,
is
doing
oktoberfest.
We
have
a
bunch
of
suggested
ideas,
including,
I
guess,
cluster
add-ons.
Other
things
bring
your
own
ideas.
I
also
have
got
on
my
raspberry
pi,
so
I
want
to
try
playing
with
that.
A
Adding
support
for
that,
but
the
the
goal
of
octoberfest
is
effectively
to
encourage
new
participants
and
to
encourage,
I
guess,
steady
participation
from
people
that
are
existing
for
people
that
need
a
framing
to
build
out
something
it's
to
encourage
that
and
sort
of
try
to
achieve
it
within
the
month.
I
guess
of
october.
A
A
I
should
also
say
sorry,
so
this
we're
also
trying
to
experiment
with
different
times
for
the
cops
office
hours
meeting,
which
has
a
bi-weekly
meeting
at
12
00
at
noon,
eastern
every
other
friday,
and
so
we
are
experimenting
with
different
times
and
we'll
likely
try
to
do
different
times
for
any
additional
meetings.
A
We
do
for
oktoberfest
trying
to
see
which
times
work
for
people
both
in
terms
of
days
of
the
week
and
in
terms
of
time
zones,
and
so
we
started
with
the
9am
eastern
on
the
the
1st
of
october,
which
feels
appropriate,
but
we
will
undoubtedly
investigate
others.
A
This
is,
I
guess
this
is
a
also
a
kubernetes
meeting
and
so
therefore
is
subject
to
our
kubernetes
code
of
conduct,
and
I
guess
we
will
put
this
also
on
youtube
on
the
cluster
lifecycle
channel,
assuming
that
proves
appropriate.
So
please
do
be
mindful
of
our
code
of
conduct.
We
don't
have
too
many
people
yet,
but
if
it
we
do
get
a
lot
more
people,
please
do
start
using.
I
guess
we
have
a
friend
of
people.
A
So
let's
try
to
use
the
hand
button
if
we
get
conflicts
on
people
talking.
Otherwise,
we're
expected
to-
or
we
want
to
be
welcoming
to
everyone
and
a
good
way
to
get
sort
of
things
going,
is
to
sort
of
do
introductions.
I
guess
I
recognize
most
of
the
people
here,
but
we
can
still
do
a
round
of
introductions.
I
I
propose
we
go
around
and
state,
I
guess
our
name
where
you
work.
If
it's
appropriate,
some
people
have
to
say
it.
Some
people
don't
want
to
say
it
whatever.
A
A
It
can
be
anything
you
want
to,
but
let's
go
around
and
I
propose
I'll
go
first
and
then
I
propose
we
go
through
in
alphabetical
order
by
the
names
you
use
to
register
with
zoom,
so
I
think
cyprian
you'll
be
first,
it's
gonna
be
a
little
odd
because
I
don't
think
the
zoom
ones
are
entirely
sorted,
which
is
a
little
surprising,
but
I'm
sure
people
can
insert
themselves.
It
looks
like
on
my
list.
A
Olay
and
peter
are
out
of
out
of
band,
and
yes,
I
kind
of
suggest,
if
you
just
if
you
don't
want
to
do
an
introduction,
just
just
say,
pass
or
something
and
that's
fine
and
obviously
just
feel
free
to
stay
as
little
or
as
much
as
you
like,
or
if
you
want.
If
you
just
remain
on
mute,
we
will
eventually
give
up
on
you.
A
My
name
is
justin
santa
barbara.
I
work
at
google
a.
I
want
to
experiment
with
the
things
I
mentioned
before,
as
well
as
trying
to
encourage
other
people
and
unblock
other
people
that
want
to
participate.
A
So
the
things
I'm
interested
in
for
the
sort
of
october
scope
are
around
trying
to
get
cluster
add-ons
into
into
cops.
I
think
santoshi,
who
is
here
was
a
or
is
was
a
google
summer
of
co-participant
this
year
and
did
some
great
work
in
cluster
add-ons,
and
I
also
want
to
give
a
shout
out
to
john
myers,
who
did
some
excellent
reviews
on
faster
add-ons
and
some
of
the
challenges
that
that
he
sees
from
the
point
of
view
of
a
cups
user
in
terms
of
using
cluster
add-ons.
A
B
Yeah,
why
not
so
my
name
is
jibriann
hackman.
I
work
at
sematext
as
a
devops.
B
I
started
contributing
to
cops
last
year
and
did
that
steadily,
and
it
was
quite
fun
for
me.
So
hopefully
it
would
be
the
same
for
everyone
for
october
hector
fest.
C
Yes,
howdy,
I'm
david
muckle.
I
work
at
sonos
on
our
core
engineering
cloud
infrastructure
team
and
I'm
just
really
excited
to
start
contributing
to
cops
as
a
first-time
contributor.
I'm
not
entirely
sure
what
there
is
out
there
to
contribute
to
in
that
realm.
But
I
am
excited.
A
Nonetheless,
wonderful,
thank
you
for
for
joining
david.
One
thing
which
occurs
to
me
on
both
of
those
last
ones
is:
we
should
probably
set
up
a
a
slack
channel.
I
guess
we
can
use
cops
developers
for
now
if
it
becomes
unmaintainable
or
if
there
are
topics
that
if
it
becomes
more
of
a
hacktoberfest
topic
and
less
of
a
topic,
we
can
sort
of
start
a
different
channel.
But
let's
do
cops
developers
for
now
and
see
where
we
go
from
there.
I
guess
I
will
skip
fireflies
john.
D
Okay,
I'm
john
myers.
I
work
for
proofpoint
and
cops.
I
started
with
working
on
rolling
update
and
cluster
validation
and
am
now
currently
working
on
more
pki
security
and
node
bring.
A
Up
thanks
john,
I
realize
that
as
people
talk,
they
move
around
in
the
orderless,
which
is
making
this
difficult
libermir.
I
see
you
joined
we're
doing
a
quick
introductions.
If
you
would
like
to
of
your
name
where
you
work
and
any
projects
you
are
interested
in
working
on
cops
or
otherwise
in
during
hacktoberfest.
E
A
Thanks
lumira
and
welcome
mike,
I
think,
you're
next,
if
I
miss
anyone,
please
speak
up,
but
mike,
I
think,
you're.
Next.
F
Right,
I'm
mike
slain,
I
work
also
at
sonos
with
david
and
scott
and
I've
been
contributing
for
a
while
to
cops.
I
did,
I
think,
my
first
contribution.
Besides
some
small
things,
was
internal
load
balancing
for
the
api
servers,
so
that
was
that
was
a
long
time
ago.
We
needed
that
and
yeah.
So
I'm
just
here
to
help
out
wherever
I
can.
People
need
pointers
to
things.
A
Thanks
mike
I'm
going
to
start
taking
notes
by
the
way
we
haven't,
we
have
a
cop's
agenda,
same
principle,
as
I
guess,
with
the
with
the
with
the
slack
channel,
we
have
a
cops
agenda
doc,
I'm
going
to
put
the
cups,
I'm
going
to
put
any
notes
into
that
meeting.
A
Notes
we
might
split
off
into
a
new
doc
should
the
need
arise,
but
for
now
I
will
try
to
keep
track
of
any
projects
that
people
are
interested
in
there
and
please
do
feel
free
to
add
things
if
you'd
like
nadir.
I
think
you
are
next.
G
A
H
Right
so
I'm
michael,
I
work
at
red
hat
right
now,
contributing
mainly
to
the
upper
your
lifecycle
manager,
there's
significant
crossover
with
add-ons
operators
that
been
slowly
trying
to
integrate
for
a
while.
H
A
That's
great,
I
actually
love
that
everyone
is
coming
with
sort
of
open
mind.
I
think
that's
really
really
wonderful.
Unless
I've
messed
up,
which
is
very
likely
peter,
I
believe
you're
next.
I
My
name
is
peter.
I
work
at
a
company
called
tune.
I've
been
contributing
to
cops
for
a
few
years
now
I've
been
a
maintainer
for
almost
a
year.
I've
recently
I've
been
tagging
a
lot
of
issues.
Well,
not
a
lot
a
few
issues
as
help
wanted
or
good
good
first
issue
labels
on
github.
So
hopefully
we
can
use
those
as
a
starting
point.
If
anyone
doesn't
have
anything
particular
they
want
to
work
on
in
terms
of
stuff.
I
want
to
work
on
I'm
hoping
that
we
can
improve
our
end-to-end
test
coverage.
I
Cops,
has
a
lot
of
jobs
in
test
grid,
but
I'm
hoping
to
improve
our
coverage
even
further,
with
specific
features
that
are
not
currently
being
tested.
So
yeah.
A
Thanks
peter
that
will
be
really
great
to
see
scott
mcculloch,
I
believe
you're
next.
J
I
also
don't
really
know
what
I
want
to
get
involved
in
mike
has
been
very
open
about
his
contributing
on
cops,
and
I
I
think,
alongside
david,
was
very
interested
in
seeing
if
there's
any
way
to
get
involved
and
how
to
do
so
and
if
anything
stood
out
in
this
meeting
to
see,
if
there's
any
way
to
kind
of
like
push
that
along
so
I'm
kind
of
just
open
to
seeing
what
there
is
and
how
I
can
get
involved.
A
That's
wonderful
welcome,
welcome
and
thank
you
peter
pointed
out
to
me
that
oh
I
missed
ole
because
ole
spoke,
so
you
move
to
the
top
of
the
list.
I
apologize.
So
should
we
go
back
to
olay
who
I
missed,
I
apologize.
K
Hi,
I
am
willa
marquis,
I'm
a
head
architect
at
sport
radar
test
coverage,
maybe
getting
more
stuff
working
on
openstack.
It's
going
to
be
some
of
the
projects
I
would
be
interested
in,
but
I'm
also
happy
to
assist
in
any
other
projects.
L
L
M
Yeah
hello,
my
name
is-
and
I
just
finished
my
google
summer
code,
internship
with
kubernetes
and
I'm
here,
because
I'm
interested
in
getting
like
cluster
add-ons
into
cups
and
I'm
definitely
open
to
contributing
to
any
other
parts
of
cops
yeah.
Thank
you,
nice
to
meet
you.
A
Wonderful,
thank
you.
Samtoshi
and
yeah
I'll
be
great
to
collaborate
with
you
on
on
that.
Seeing
us
like,
following
up
from
all
the
work
you
did
in
google
summer.
That
would
be
really
great.
A
I
believe
we
have
introduced
everyone,
I'm
just
scanning
through
the
list.
If
I
missed
you,
I
apologize
and
please
speak
up
while
I
scan,
but
I
think
that
is
everyone.
A
A
I
don't
know
what
we
want
to
how
we
best
want
to
do
this
if
there
are
people
that
are
interested
in
pairing
up
with
any
projects
that
they
heard,
other
people
describing
that's
great
and
I
highly
encourage
you
to
do
so.
We
can,
I
think,
coordinate
best
in
the
kubernetes.
Develop
cops
developers,
cops
developer
channel
on
slack,
which
I
encourage
you
to
join.
A
If
you
have
not
done
so,
and
for
those
of
you
that
are
not
perhaps
cops
primarily
primarily
cops,
we
can
let's
use
the
channel
for
now,
and
we
can
split
off
into
a
new
channel
when
we
when
we,
if
we,
if
we
feel
that's
necessary,
because
I
know
that
some
of
you
are
are
more
interested
in
other
other
c
cluster
electrical,
which
is
great,
and
so
we
did
have
some
people
interested
in
cluster
add-ons.
We
had
some
e
to
e
peter
and
olay.
A
Both
cited
e
to
e
the
the
cni
plug-in,
I
think
will
be
will
be
great,
oh
cypriot.
I
actually
didn't
take
notes
on
what
you
said.
You
said
you
wanted.
I
apologize
for
that.
You
said
you
wanted
to
work
on
getting
the
container
runtime
support
into
node.app.
B
And
by
the
way,
I
also
was
planning
for
this
release
to
touch
calico
features,
so
there
are
a
few
newer
features
that
are
quite
interesting
and
I
wanted
to
add
them
to
cops,
not
sure
if
anyone
is
interested,
those
would
be
at
least
two
of
them
would
be
beginner
level.
So
if
anyone
wants
to
join
that,
I
could
help
and
one
could
be
a
bit
more
advanced,
but
not
really
something
extraordinary.
So
there
is
already
an
example
for
it,
so.
A
I
think
one
thing
maybe
we
we
could
do
is
if
there
are
issues
which
we,
which
people
are
interested
in
working
on
as
part
of
this
oktoberfest
topic,
whether
in
cops
or
other
repos,
why
don't
we
label
them
with
a
hacktoberfest
label
and
make
sure
that,
like?
I
will
make
sure
that
I
create
like
covering
issues
for
my
little
raspberry
pi
thing,
for
example.
So
if
there
are
other,
if
there
are
other
topics
that
are
just
not
covered,
please
do
create
a
umbrella
issue.
A
That
was
the
word
I
was
struggling
for.
I
I
will
make
sure
that
we
get
a
get
a
label
added
to
at
least
the
cops
repo
to
add
for
oktoberfest,
and
then
I
will
figure
out
what
the
right
way
is
to
get
queries
going
so
that
people
can
sort
of
see
those
and
coordinate
on
on
github,
which
I
think
is
like
github
is
great
for
asynchronous,
and
I
think
slack
is
is
great
for
synchronous.
A
The
I
don't
know
if
there
are
any
other
topics,
people
would
like
to
other
topics.
People
would
like
to
discuss
in
this
group.
A
I
think
one
of
the
other
things
we
could
do
is
we
could
sort
of
try
to
arrange.
I
guess
breakout
meetings
or
break
out
pairings
might
be
more
likely
for
people
that
want
to
collaborate
or
that
want
to
generally
work
on
anything
or
want
just
sort
of
introductions
to
the
to
the
cops
code
base.
I
think
there
are
a
couple
of
people
that
are,
I
guess,
newer,
I'm
happy
to
do
a
sort
of
introduction
to
the
cops
code
base.
A
One
way
we
it's
it's
hard
to
I'd
like
some
feedback
on
that
like
it's
hard
to
know
how
to
present
what
is
a
large
code
base
to
a
new
person,
whether
it's
better
to
try
to
traverse
it,
whether
it's
better
to
which
risks
sort
of
just
overload,
whether
it's
better
to
add
a
feature
and
sort
of
discover
it
incrementally,
which
is
sort
of
how
I
tend
to
proceed
and
then
there's
a
subset
of
adding
the
feature
where,
like
I
drive
or
someone
experienced
in
the
code
based
drives
versus
pairing
and
letting
someone
less
experience
in
the
code
based
drive
that
can
sort
of
then
ask
questions.
A
D
A
D
Well,
currently,
I've
been
using
the
good
first
issue
tag,
but
one
of
them
isn't
an
issue
yet
so
I
suppose
I
need
to
file
it.
Oh.
A
Yeah,
I
think
the
my
my
issue
with
good
first
issue.
Is
that,
like
it's
not
necessarily
specific
to
oktoberfest
and
like
some
of
them
may
be
older,
I
I
know
that
some
people
have
gone
through
and
cleaned
up,
which
is
wonderful,
and
but
yes
we
can.
I
suggest
we
do
oktoberfest
just
so.
We
sort
of
maintain
some
momentum
around
it
and
it
looks
like
yeah.
We
lost
oh
well.
D
I'll
give
my
ad
it's
issue
719
7119,
which
is
older,
but
I
I
think
it's
a
good
thing
for
someone
to
take
on
as
a
beginning
project,
and
that
would
be
getting
the
aws
autoscale
group
to
actually
drain
nodes
when
it
wants
to
terminate
one,
I'm
just
figuring
out
what
the
system
d
implications
are
and
the
right
mechanism
to
get
that
done.
A
That's,
that's
that's
great.
I
think
one
thing
we
could
do
right
now
is
actually
discuss
like
what
is
entailed
in
that.
Like
also
because
I
see
there
are
some
people
here
that
may
have
experience
from
other
projects
that
might
want
to
chime
in
or
may
be
interested
as
well.
Do
you
want
to
talk
a
little
bit
more
about?
What's
entailed
there.
D
Well,
I
think
part
of
it
is
figuring
it
out,
because
I
understand
there
is
a
way
that
you
know
it
will
do
a
some
systemd
shutdown.
So
someone,
I
think
someone
needs
to
go
in
and
figure
out
what
the
right
unit
file
to
add
in
is
and
what
the
actual
write
mechanism
is
actually
and
then
have
a
piece
of
code
there
which
will
drain
the
node.
A
Nadio,
I
was
thinking
of
you.
Is
there
a,
I
guess,
you'd
probably
guess,
is
there
a?
Is
there
a
drain
on
shutdown
functionality
in
cluster
api
yet
or.
A
That
might
be
a
good
one
to
do.
I
I
remember
there
being
something
called,
I
believe
it's
drano,
but
that
could
it
could
be
a
different
trademarked,
sync
unblocker,
which
I
may
be
finishing
upon.
So
I
apologize
to
anyone
who's
trademark
I
stamped
on,
but
I
think
there
was
a
project
called
draeno.
That
is,
I
think,
a
daemon
set
that
does
some
of
this.
I'm
just
sort
of
searching
for
that
in
the
background,
so
it
doesn't
necessarily
even
have
to
be
a
a
cop,
specific
feature.
F
Actually,
I
tuned
out
for
a
second,
but
so
we
used
durano
but
there's
also
a
kubernetes
project
project,
the
d
scheduler
kubernetes
d
scheduler.
I
don't
know
if
that's
solve
similar
problems.
A
We're
all
busy
typing
in
the
same
two
searches,
I
believe
yeah,
so
it
looks
like
draynow-
is
thank
you
for
the
links
nadir.
I
will
oh
and
mike.
I
will
try
to
paste
them
into
the
the
dark.
The.
A
It's
unclear
whether
that
node
condition
can
come
from
aws
and
the
d
scheduler
I've
heard
of
using
in
the
case
where
you
know,
like
you,
have
two
nodes
and
while
you're
restarting
there's
a
period,
we
only
have
one
that's
healthy,
so
everything
lands
on
that
one
node
and
then
you
know
both
your
your
dns
pods
are
on
that
one
node!
That's
where
that's
the
context.
I've
heard
it
and
I
don't
know
whether
it
does
more.
A
I
don't
know
whether
anyone
here
uses
either
well,
what
uses
drain
out
or
d
scheduler,
or
both
particularly
for
this
problem.
E
But
the
scheduler
seems
to
define
a
policy,
so
if
you
have
pots
that
you
don't
want
on
a
certain
note
is
going
to
execute
on
this
policy,
I
wouldn't
say
that
this
is
exactly
a
draining
process.
However,
I
wanted
to
comment
in
general
about
including
such
the
feature
of
draining
inside
a
project
can
be
tricky,
because
there
are
cases
where
users
might
not
want
to
drain.
G
No,
it's
not
class
api.
I
think
we've
we've
got
some
system
d
tweaks
to
do
around
that,
just
to
make
sure
sure
stuff,
like
mainly
around
kubelet
shutting
down,
but
we
do
have
the
drain
support,
so
you
can
actually
say
you
can
we?
We
have
all
the
train
support
in
terms
of
on
getting
rid
of
a
machine.
G
E
Yeah
because
draining
can
also
block.
So
this
is
the
main
problem
there.
It
has
to
be
a
tweak,
a
knob
for
that.
A
Yes,
I
think,
as
I
understand
and
jump
off
really
correct
me,
the
issue
is
specific
to
unscheduled
auto
scaling
termination.
So
this
is
using
the
aws,
auto
scaler
or
I
guess,
using
spot
instances
where
you
cannot
predict
easily
the
there's
another
link.
Oh,
you
can't
predict
easily
that
a
node
is
going
to
go
away.
It
isn't.
It
is
not
initiated
by
the
by
the
user.
A
It's
initiated
by
aws,
and
so
aws
gives
you,
let's
say
30
seconds
head
heads
up
that
your
node
is
about
to
shut
down,
and
the
question
is:
what
do
you
want
to
do
in
those
30
seconds
and
at
the
end
of
30
seconds
you
shut
down?
Is
my
understanding.
It
isn't
that
you
yeah
well.
D
Spot
you
have
a
time
limit,
but
it's
also
the
case
where
the
asg
decides
that
it's
too
big
and
it
wants
to
to
make
it
smaller.
So
it's
you
know
it's
trying
to
scale
down
the
instance
group,
and
in
that
case
you
pretty
much
have
as
much
as
as
long
as
you
need
you
know,
but
you
do
want
to
get
those
pods
drained,
possibly
respecting
any
disruption.
Budgets.
A
This
actually
sounds
like
a
great
issue.
I've
put
I've
copied
these
links
into
our
dock.
I've
tried
to,
I
put
the
number,
which
I
think
is
9119,
but
I
could
be
wrong,
but
I
I
think
this
is
a
great
one
for
for
someone
to
tackle
yeah
seven,
one,
one,
nine
seven
7119.
I
apologize.
A
If
I
can,
are
there
other
issues
which
people
think
might
be
good
ones
to
discuss,
that
issue
feels
like
it's
a
one,
a
good
one
for
someone
that
understands
a
bit
of
aws
but
doesn't
necessarily
or
clouds,
because
I
think
there's
a
similar
function
on
other
clouds,
but
doesn't
isn't
necessarily
is
newer
to
cops
or
something
like
that
doesn't
require
as
much
in-depth
on
cops
but
does
require
more
sort
of
understanding
of
how
clouds
work
and
what
the
right
thing
to
do.
A
Is
there
as
yeah,
as
was
pointed
out
in
terms
of
the
drain,
potentially
being
a
an
unbounded
and
blocking
operation.
So
it's
not
entirely
clear
what
the
correct
behavior
is.
I
I
have
a
few
issues
linked
in
the
oktoberfest
issue
in
in
the
cops
repo
which
is
9920
at
the
very
bottom
of
the
issue.
There's
a
couple
github
links
they're
both
more
around
the
cops
api.
I
So
if
anyone
wants
to
have
experience
with
that-
and
I
think
that's
a
bit
simpler
of
a
task-
it's
more
kind
of
isolated-
I
think
those
would
be
good
starting
points
and
I
or
anyone
else
or
others,
I
should
say-
would
be
happy
to
provide
assistance
with
those
if
anyone
wants
to
tackle
them-
and
I
think
they're
particularly
useful
to
have-
I
was
just
running
into
some
of
those
issues
myself
this
week.
I
Sure
so
the
first
one
is
cops
issue:
nine,
eight,
four,
five
I'll
paste,
a
link
in
the
chat,
and
so
this
is
currently
we
allow
the
docker
config.
All
of
the
flags
you
pass
to
the
docker
vehman,
that's
defined
at
the
cluster
level,
but
there
is
a
reasonable
ask
to
have
that
be
defined
at
the
instance
group
level,
so
that
different,
auto
scaling
groups
could
be
using
different,
docker
settings,
and
so
essentially,
I
think
that
would
just
be
a
matter
of
adding
the
new
docker
config
field
to
the
instance
group.
I
I
Docker
config,
that
sort
of
thing-
so
I
think,
that's
fairly
straightforward
and
then
the
second
issue
is
6356,
which
is
just
adding
at
least
one,
maybe
more
cubelet
flags.
So
all
of
cubelet's
flags
are
defined
as
separate
api
fields
within
a
big
cubelet,
config
struct
in
the
cops
api
and
so
anytime
there's
a
new
flag.
We
need
to
add
a
new
field
for
it,
and
so
this
is
asking
for
a
housekeeping
interval
flag,
and
so
I
think
we
even
have
documentation
in
the
cops
docs
about
how
to
do
this
exact
task.
I
So
I
think
that's
pretty
well
well
defined.
There.
A
Those
are
those
both
seem
like,
as
you
say,
both
well
defined
and
great,
and
I
have
applied
the
oktoberfest
new
hacktoberfest,
with
halloween
orange
background
label
to
both
of
them.
Thank
you.
I
see
lumiere,
you
have
a
hand
raised.
E
Basically,
I
would
suggest
that
you
try
to
follow
the
issue,
I'm
going
to
find
it
and
move
everything
to
the
couplet
configuration
api
if
possible.
I
I'm
not
sure
justin
if
you're
already
handling
the
corporate
configuration.
A
It's
certainly
something
thank
you
for
bringing
that
up.
It's
certainly
something
that
you
know
the
this
is
all
the
whole
component
conflict.
It's
certainly
something
we've
been
tracking,
I
guess.
Since
the
dawn
of
cops
like
we
have,
we
adopted
the
idea
of
having
a
or
we
embraced
the
idea
of
having
a
com
of
a
component
configuration
in
a
kubernetes
type.
We
did
not
follow
the
component,
config
schema,
largely
because
it
didn't
exist
at
the
time.
A
I
know
john
is
talking
about
an
the
the
next
version
of
the
cops
api
which
you
know
like
would
be
a
great
opportunity
to
converge
these
two
schemas.
I
think
that
would
be
really
good.
We
could
also
talk
about
breaking
them
apart.
The
add-ons,
so
it's
similar
to
the
add-ons
discussion
in
that
today
cops
has
a
cluster
specification
with
a
bunch
of
cluster-scoped
things,
including
cni
configuration
typically
there's
also
instance,
group
configuration,
which
is
the
equivalent
of
machine
deployment
configuration
the.
A
So
in
that
big
clusters,
spec
object
or
spec,
there
are,
you
know,
configuration
for
cni's,
which
probably
eventually
should
go
into
add-ons,
and
there
are
configurations
for
the
core
components
like
cuba,
like
api
server
like
scheduler,
like
cube
controller
manager,
which
I
think
we
definitely
want
to
convert
the
schemas
in
the
long
term,
and
I
think,
there's
a
question
of
whether
we
should
also
think
about
extracting
those
into
step.
A
Separate
objects
which
would
make
the
I
think
that
would
have
a
lot
of
benefits
in
that
the
cluster
object
in
cups
would
become
smaller,
and
it
would
mean
that
the
whatever
tool
you're
using
the
configuration
for
setting
a
field
on
api
server
would
be
the
same
right.
A
You
if
we
can
all
agree
that
cops
and
cube
adm
and
cluster
api
should
effectively
use
the
use
the
api
server
component
config
as
the
primary
means
of
configuration
that
I
think
that
would
be
great
for
a
user
experience
across
the
across
the
tools
and
also
enable
us
to
sort
of
interoperate
a
little
bit
more
than
has
been
practical
in
the
past.
A
Like
obviously,
you
know,
like
cops,
would
like
to
use
cluster
api
at
some
stage
in
the
future,
and
if
we
can
just
say,
oh
well,
look
we
we
only
have
to
bridge
these
like
this
small
set
of
subset
of
fields
that
makes
life
a
lot
easier.
I
don't
know
how
people
they
feel
about
that.
I
Yeah,
I
think
I
think
one
thing
to
consider
is
cops-
is
needing
to
support
a
wide
range
of
kubernetes
versions,
and
so,
if
different
versions
are
using
different
component
config
versions,
we
need
to
make
sure
that
cops
can
support.
Writing
all
of
those.
D
Yeah,
it's
somewhat
unfortunate
because
the
stated
reasons
for
moving
to
the
the
new
config
struct
was
to
make
it
easier
for
cluster
code
like
cops,
and
it
actually
has
the
effect
of
making
it
harder
for
us.
But
there
we
go.
A
If
I
recall
correctly,
kubay
dm
had
some
challenges
with
versioning
of
component
config.
Am
I
speaking
correctly
lumiere
or
is
that
is
that
true,
yes,.
E
I
mean
it's
configuration
versioning.
The
problem
has
existed
like
before
software
existed,
so
we
definitely
have
problems
around
that.
What
we
ended
up
doing
is
we
nowadays
pass
transparently.
E
E
As
you
know,
validation
conversion
is
not
possible.
So
we
just
assume
that
the
version
schema
that
the
user
is
passing
is
compatible
with
the
target
component.
A
This
is
this
is
going
very
well
in
my
mind,
because
now
we
like
this
is
looking
more
and
more
like.
These
are
the
exact
same
problems
we
face
with
cluster
add-ons
and
I
feel,
like
I'm
very
happy
with
the
people
that
are
here
and
like
we
have
almost
accidentally,
gathered
the
the
perfect
set
of
people
to
have
this
discussion
around
like
can
we,
I,
I
think,
not
everyone,
I
think,
has
probably
looked
at
john's
feedback
on
the
initial
integration
of
cluster
add-ons.
I
think
there
were.
It
was
great
feedback.
A
I
think
there
were
questions
around
these
same
issues
around.
I
think,
john,
you
were
more
focused
or
primarily
focused,
I
guess
on
security
concerns
or
manageable
security
manageability,
but
I
think
the
same
like
correctness,
concerns
and
validation
concerns
apply
here.
How
do
we?
How
do
we
not?
How
do
we
validate
the
data
interoperate
with
those
with
like
externally
managed
projects
yet
yet
not
tie
ourselves
to
their
lifetimes,
like
I
think
you
pointed
out,
if
we're
just
gonna
re-embed,
all
the
controllers
back
into
cops
or
back
into
whatever
it
is
like.
A
What's
the
point
of
splitting
them
out
right-
and
I
think
that's
so
specifically,
for
example
like
we
could
of
course
build
in
special
logic
to
validate
the
aws
vpc
cni
configuration,
but
we
would
like
the
awsvpccni
in
the
long
term
to
be
a
an
add-on
or
an
operator
that
can
self-manage
and
self-update
and
do
those
sorts
of
things.
A
How
do
we?
How
do
we
do
that
without
without
baking
in
knowledge
or
too
much
knowledge
of
the
aws,
vpc,
cni,
schema
or
even
operator?
I
don't
know
if
anyone
has
any
approaches
there.
A
A
I
know
that
how
do
people
feel
about
execution
of
containers
like
to
validate?
I
don't
know
whether
anyone
has
debated
that
internally
or
on
their
in
their
projects.
A
So,
for
example,
specifically
for
the
validation,
the
problem
that
lupimir
described
to
it,
we're
going
to
validate
the
cubelet
configuration,
we
could
have
a
docker
container,
whose
job
it
was
to
validate
that
doc,
that
cuba
configuration
and
the
tooling
could
execute
that
docker
container
or
execute
that
code,
somehow
in
some
sort
of
docker
being
a
sandbox,
execute
some
sandbox
code
which
isn't
part
of
the
project
but
is
still
referenced
from
the
project,
and
I
know
that,
like
satoshi
as
part
of
google,
summer
of
code
has
looked
at
like
and
in
cluster
add-ons.
A
We've
also
discussed
this
notion
of
sort
of
one
shot
where
we
run
something
once
to
produce
the
manifests
that
are
that
are
required
and
that
that
sort
of
behavior,
rather
than
necessarily
using
a
full
operator.
I
don't
know
whether
I
feel
like
I'm
wittering.
I
don't
know
if
anyone
wants
to
jump
in
and
save
me.
E
We
have
definitely
discussed
the
idea
of
a
container,
like
I
mean
ephemeral
container.
Doing
some
work,
not
you
know
not
the
ephemeral
container
feature
that
is
coming
to
kubernetes,
but
in
general,
a
container
that
is
doing
a
task
and
then
disappearing.
E
In
fact,
I
believe
I
proposed
to
coaster
api
to
use
the
kubereum
binary
in
the
container
to
up
convert
the
cube,
adm
api
version.
This
idea
was
turned
down.
I
do
not
recall
why,
but
also
if
you
want
to
run
the
couplet
in
a
container
to
validate
its
own
component
config,
you
face
the
problem
where
there
are
no
flags
for
that.
There's
no
like
mode
for
the
couplet
binary
to
validate
configuration,
so
there
has
to
be
something
like
validate.
E
Otherwise
you
are
making
assumptions
that
this
particular
container
is
not
going
to
fail
for
other
reasons
and
not
the
component
config
itself
being
invalid.
Also,
we
propose
to
api
machinery
to
do
the
same
to
the
controller
manager.
Api
server.
I
haven't
spoken
to
scheduling,
but
we
had
this
idea
to
pretty
much
enable
the
validation
on
all
the
core
components.
E
E
You
cannot
extract
it
without
forking
the
component
source
code.
So
I
like
this
idea.
I
would
love
to
see
how
how
this
goes.
If
you
proceed
with
it,
but
we
still
may
have
to
add
these
flags
to
do
the
explicit
validation.
A
Yeah,
thank
you
for
saving
me.
I
think
I
I'd
love
to
hear
from
in
a
minute
from
from
the
deer
about
why
like
whether
he
has
any
insight
into
why
the
classroom.
If
I
wasn't
keen
on
this,
I
think
I
don't
think
it's
unreasonable.
I,
I
am
rarely
in
the
position
of
defending
this
particular
state,
but
I
don't
think
it's
unreasonable
to
say
that
we
can
do
this.
We
can
prove
this
first
in
something
like
cluster
add-ons
and
show
that
it
works
before
we
make
changes
to
kk.
A
I
feel
like
we
can
do
that
validation
if
we
can,
if
we
can
agree
if
we
feel
this
is
a
good
thing
to
do,
we
can
change
kube
builder
or
kubota,
declarative
pattern
or
other
tooling
or
just
right
hand
write
one
to
create
a
validator
validator
that
works
for
some
add-on
and
then
show
that
it
is
useful
and
and
valuable,
and
then
we
can
go
and
say
to
controller
manager
that
they
should
add
this
validate
flag.
A
I
feel
like
that's,
that's
a
reasonable
course
of
action,
rather
than
demanding
they
speculatively
add
it
to
kk.
For
anyone
that
doesn't
know,
kk
is
kubernetes.
Kubernetes,
the
core
repo.
Sorry,
I
apologize
for
using
jargon.
Nadir.
Do
you?
Do
you
recall
why
what
the
feedback
was
from
cluster
api
around
the
idea
of
executing
convenience.
G
And
I
don't
remember
that
discussion,
but
I
suspect,
is
we
didn't?
You
know
really
want
to
shell
out
inside
the
controller
at
the
time.
You
think
that
might
be
less
of
concern
if
there's
a
well-defined
contract
around
validation
and
we
get
programmatic
output
rather
than
having
two
reg
x
paths:
output
from
a
command
yeah.
L
E
A
The
baking,
the
binary
into
the
image
into
the
controller,
as
you
say,
is
a
good
point
and
it
it
it
brings
with
it
like
versioning
challenges
versus
we
have
we've,
we've
pulled
it
out
of
the
code,
but
we
have
not
pulled
it
out
of
the
packaging
problem
as
it
were
so
now
we
have
to.
We
still
have
to
do
a
new
release
of
cluster
api
or
cops
to
produce
a
self-contained
image.
A
I
I
don't
know
if
you
have
thoughts
or
other
people
here
have
thoughts
around
taking
a
dependency
on
docker
or
the
the
the
requirement
to
be
able
to
run
docker
containers.
A
I
know
nikki
red
hat
has
looked
at
this.
A
lot
like
how
to
how
to
is
it?
Okay,
to
take
a
dependency
on
docker
on
in
the
client,
where
we
run
a
cli
tool,
for
example,
which
could
be
in
ci
and
could
itself
be
kubernetes,
and
is
it
okay
to
take
a
dependency
on
on
executing
a
container
in
a
controller
like
the
cluster
api
controllers,.
H
So
I
am
personally
pretty
biased
with
this.
I
think
that
we've
had
a
lot
of
trouble
with
containerized
environments
and
taking
dependencies
on
clyde
with
something
like
docker,
because
it
requires
a
daemon
with
root
privileges,
and
now
you
need
to
run
a
privileged
container
to
run
in
your
build
and
now
that
has
a
whole
slew
of
other
requirements.
H
H
G
Yeah,
I
think
it's
going
to
have
to
be
hard,
no
for
docker
in
a
controller,
given
that
none
of
the
club
image
builder
images
for
cluster
api,
even
ship
with
docker,
it's
container
d,
only,
I
think,
there's
I
don't
know.
I
think
there
was
a
proposal
in
progress
to
move
docking
out
of
kkk
anyway,
so
yeah,
that's
a
no-go
cli,
I
mean
cluster
cuttle,
which
is
the
cli
for
cluster
api
uses
kind.
G
So
I
guess
there
is
some
assumption
around
docker,
but
it's
not
quite
ideal,
but
it
has
to
be
hard
now
for
the
controller.
A
I
think
I'm
next,
though
the
I
don't
wanna
like
overtrain
on
docker
per
se
versus
like
podman
type
thing
like
if
it's
docker
pub,
I
think,
that's
fine.
I
think
I
guess
my
question
I
guess
to
nick
is
like
what
is
the
alternative
to
to
container
execution
for
packaging
or
container
packaging?
Like
does
do
you
have
an
idea
on
how
to
do
it?
A
Is
it
the
case
that
we
would
have
to
like,
for
example,
in
the
case
of
cops
which
might
have
to
install
a
dozen
cni's,
we
would
have
to
package
validating
binaries
for
those
dozens
cni's
in
some
sort
of
cops
distribution
like
build
the
distribution
and
update
it
whenever
there's
a
new
version
of
calico
or
a
new
version
of
selim
or
whatever
it
is.
I
mean
that's
a
fine
answer.
I
just
want
to
know
what,
if
that
that's
sort
of
what
you
would
lean
towards.
H
Go
ahead:
yeah,
okay,
I'm
trying
to
be
good!
I
I
jumped
in
last
time,
so
we've
been
doing
a
lot
of
work
in
trying
to
decide
if
there's
a
good,
just
plain
json
format
or
something
that
can
describe
the
configuration
for
an
operator
on
our
side,
which
is
like
a
set
of
manifest
plus
a
set
of
metadata,
and
those
manifests
have
some
uri
somewhere
that
you
can
go
grab
it
at
and
that
may
or
may
not
be
in
a
container
image
itself.
H
Because
we've
been
running
into
very
similar
problems,
I
the
the
result
of
that
is
not
clear.
Yet
I
don't
know
if
that
helps
at
all.
A
It
does
not
help,
I
think
it's
it's.
I
mean
it's
not
an
answer
really,
but
it's
great
to
hear
that.
Yes,
that
we're
all
sort
of
dealing
with
the
same
thing
and
that
unless
anyone
has
answers,
I
think
we
can
table
this
and
hopefully
make
some
I'd
love
to
make
some
progress
on
this
topic
in
hacktoberfest.
There
are,
there
are
two.
A
I
think
the
two
scenarios
are
very
different,
like
I
think
it's
important
to
keep
in
mind
the
cli
scenario
and
the
controller
scenario
separately
and.
A
I'm
more
confident
in
the
controller
scenario
like
I
feel
like
we
could
spin
up
a
job
or
a
pod
temporarily,
and
somehow
do
that.
Then
I
am
in
the
cli
scenario.
So
it's
interesting
to
hear
that
ninja,
you
seem
more
confident
in
the
sea
of
life,
so
just
go
for
it.
G
Docker
directly
inside
a
controller
according
to
creating
a
new
part
and
getting
results
from
that.
That's
fine
and
we
do
that,
for
example,
in
summerboy,
which
is
the
conformance
checking
thing,
that's
exactly
how
silver
operates.
The
only
other
thing
we
also
need
to
take
into
consideration
is
air
gapped
environments.
G
A
Absolutely
right,
nick,
I
see
your
hand
is
still
up.
I
don't
know
if
that's
a
hold
over
from
prior
raising
and
I'm
going
to
assume.
Yes,
I
know
that
that
exact
topic,
the
air
gap
topic,
came
up
in
john's
security
management.
Manageability
list
of
concerns
I
think
there
are,
there-
are
better
f.
I
think
I
I
sort
of
punted
that
can
down
the
road.
A
I
don't
know
how
john
feels
about
that
by
sort
of
saying
like
yes,
we
should
absolutely
solve
this,
but
it's
very
not
specific
to
to
cluster
add-ons.
You
know
we
want
to
make
sure
that
those
same.
Can
those
same
concerns
apply
to
all
the
images
right
we
want
to.
We
want
to
make
sure
that
the
images
are
all
mirrored
somewhere
for
your
kubernetes
version
that
that
you
have
control
over
it,
that
you
can
see
what
images
they
are
that
you
can
validate
them
there
are.
There
are
some
projects
that
do
this.
A
I
don't
know
whether
people
have
anyone
here
is
successfully
or
on
less
successfully
using
any
of
the
mirroring
tools
that
are
available
or
any
of
the
container
policy
or
container
image
policy
image
pull
policy.
Any
of
those
sorts
of
things.
A
Sounds
like
it's
early
adopter
time.
So,
yes,
I
agree.
It's
very
like
there
are.
There
are
a
bunch
of
problems
there
and
container
d
makes
the
I
think
container
d
makes
the
make
some
of
it
easier
in
that
you
are
able
to
set
up
mirrors
with
container
d.
It
doesn't
talk
about
how
that
those
mirrors
get
populated,
and
it
doesn't
talk
about
how
we
can
be
sure.
A
I
I
had
originally
scheduled
this,
for,
I
think,
an
hour
and
a
half
to
end
at
10
30..
I
don't
know
if
there
are
other
topics
that
people
want
to
bring
up.
A
Our
traditional
time
is
about
an
hour,
and
that
seems
to
be
what
most
people
can
focus
for
or
make
time
in
their
busy
calendars,
for
if
there
are
not
other
topics,
I
propose
that
we
that
we
can
follow
up
in
in
the
cops
developer
slack
channel
initially,
and
we
will
likely
create
a
hacktoberfest
channel
if
people
feel
that's
of
interest
for
these
crosstopic
ones.
A
I
I
don't
know
what
the
right,
whether
people
have
used,
I'm
wary
of
creating
a
sort
of
ephemeral,
discussion
channel,
but
maybe
that's
what
slack
is
like.
You
know
like
these.
If
we
have
a
great
discussion
around
how
to
like
implement
cluster
add-ons
in
cops
and
and
cluster
api
and
cube,
adm
like
we
don't
want
to
disappear
on
november
1st.
E
You
give
it.
This
is
initially
a
cops,
no
october
event,
maybe
should
be
the
cop
shadow
for
now.
Are
you
proposing
something
like
a
generic
channel
with
the
kids
like.
A
I
wasn't
sure
what
the
right
way-
I'm
I'm
I
don't
know
I
I
want
to-
I
want
to
somehow
encourage
participation.
I
don't
want
to
require
people
to
work
on
cops
related
things
if
they
don't
want
to,
but
I
also
want
to
say
we
have
to
start
somewhere.
So
I
like
the
idea
of
starting
with
cops
developers
and
in
slack
and
we
can
farm
out
from
there.
A
I
guess,
like
certainly
the
cluster
add-ons
one
we
can
move
like
all
the
projects
have
channels,
so
we
can
presumably
redirect
to
the
most
appropriate
channel.
E
A
A
A
How
do
people
feel
about
like
pair
programming
and
streaming
that
sort
of
thing
or
programming
and
streaming
that?
Do
you
think
that's
good
or
not
good?
Does
anyone
try
doing
that
and
then
I'll,
let
everyone
go
back
to
their
daily
lives.
E
I
personally
think
this
is
great
and
also,
I
think,
the
the
code
base
walkthroughs
they
die
out
or
not
are
also
great.
We
should
definitely
have
more
of
that
in
the
communities
project.
In
fact,
we
should
host
pretty
much
weekly
events
of
this
type,
because
this
is
a
great
way
to
get.
You
know
the
the
itches
going
in
people
like
hey.
I
really
want
to
get
involved
in
this
particular
package.
It's
interesting
to
me
yeah.
E
I
think
this
is
a
very
good
idea
and
if
cops
pirates,
something
like
that,
I
will
definitely
try
to
approach
other
six
and
people
to
look
at
this
idea.
A
Cool,
well
I
mean
that's
something
I
I
think
might
be
fun
I've.
Always
I've
always
done
something
like
that.
I've
always
learned
stuff
from
it.
So
I
feel
like
I
can.
I
can
maybe
look
at
doing
some
sort
of
streaming
type
event.
I
think
the
only
peter's
reminded
me
we
should
decide
on
a
time
for
our
next
meeting.
We
can
also
do
that
in
another
doodle
or
in
actually
did
we
have
a
did.
We
have
a
wednesday
peter.
Do
we
have
a
wednesday
time
slot
that
was
otherwise
voted
very
highly.
I
I
Eastern
there
was
also
three
votes
for
wednesday
at
10
a.m.
A
I
can
do
I
can
I
can
do
a
tuesday.
I
can
do
either
slot.
I
we
wanted
to
not
be
too
close
to
the
friday
slot.
I
could
do
tuesday
at
9am
and
we
can
figure
out
what
we
do
in
that
slot.
I
don't
imagine
we
want
everyone
to
go.
I
don't
mention
everyone
wants
to
show
up
every
every
couple
of
days
for
a
chat,
but
I
think
it
could
be
fun
if
people
I'm
sure
that
we'll
find
something
to
do
it
could
be
broadcasting
coding.
A
It
could
be
pair
programming
for
coding.
It
could
be
digging
down
into
one
of
these
topics,
so
I
I'm
gonna
mentally
reserve
that
slot,
at
least
unless
anyone
disagrees-
and
we
can
make
that
the
next
time
and
we
we
will
leave
the
agenda
undecided.
So
if
anyone
wants
has
anything
they
want
to
discuss
at
that
time,
tuesday
9
a.m,
eastern
they
are.
A
A
This
meeting-
and
I
think
next
steps
are
to
see
you
all,
hopefully
on
the
cops
developer
channel,
where
we
will
form
out
fun,
ideas
and
and
talk
about
these
things
and
possibly
split
into
like
it
sounds
like
there's
a
a
cluster
add-ons
discussion
that
will
likely
come
out
of
this.
A
If
anyone
else
has
any
closing
thoughts,
please
say
so
otherwise
thank
you
all
for
attending
and
hope
to
see
you
throughout
the
month
of
october
thanks.
I
think
this
is
great
awesome.
Thank
you.