►
From YouTube: Kubernetes Sig Node 20180227
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
Full
implementation
details
on
how
the
new
notes
created
how
to
update
the
note
in
the
implementation
and
kind
of
the
different
printers
that
get
passed
through
and
what
talks
the
Karina's
API
server,
and
things
like
that.
So,
if
you
guys
want
to
go
ahead
and
just
read
it
through,
it
should
only
take
a
couple
minutes
and
then
we
can
talk
about
it.
A
A
B
A
Kind
of
why
we're
coming
into
this
group
too
in
this
forum,
because
we
haven't
thought
about
everything
so
right
now
the
way
that
implementation
is
working
I'll
just
be
completely
honest.
The
way
we're
doing
is
whenever
a
new
feature
comes
through
for
a
CI.
So
as
your
container
instances,
for
example,
we
just
got
support
for
a
DNS
name.
Labels
like
once
that
comes
through.
We
directly
put
it
into
virtual
couplets,
there's,
not
a
lot
of
like
futuristic
design
happening.
A
Yet
networking
is
a
huge
one,
because
I'm,
not
a
hundred
percent
sure
how
pods
that
are
in
your
cluster
are
going
to
talk
to
pods
that
are
in
a
different
service
like
in
hyper
service
or
Azure
service
or
whatever,
and
then
there's
another
question
for
what
what's
a
provider
even
like?
What
should
we
accept
what
should
be
in
the
repository?
What
should
it
so
there's
a
lot
of
those
kinds
of
questions
right
now
that
we
need
to
work
through?
A
So
that's
where,
like
once,
we
come
towards
networking
like
I'll
mean
others
can
help,
drive
us
back
and
then
bring
it
to
you
guys
and
be
like
this
is
kind
of
what
we're
thinking
for
the
design
and
then
hopefully
talk
through
it
because
I
think
we'll
need.
But
but
this
is
just
like
a
starting
point.
A
I,
don't
think
so.
We
do
have
a
concept
of
it's
possible
to
install
multiple
connectors
for
different
configurations,
so
I
and
I'm
just
gonna
use
AC
Isaac
as
an
example,
because
it's
the
one
that
I
work
most
closely
with,
but,
for
example,
with
a
CI
you
have
to
put
in
like
resource
constraints
that
are
congruent
to
what
we
have
within
our
service
itself.
So
you
could
have
different
connectors
that
have
different
constraints
or
specs
like
that.
A
You
could
also
have
a
connector
that
deploys
into
different
regions,
so
you
can
have
one
connector
that
deploys
out
into
ecos.
You
can
have
another
one
that
deploys
out
into
the
West
us
and
you
can
start
to
get
like
multiple
connectors
in
it
in
a
cluster
and
kind
of
the
dollop
scenario
is
what,
if
you
just
don't
have
VMs
in
your
cluster
at
all,
you
don't
have
any
in
Egypt,
VMs
and
you're
just
running
off
with
containers.
A
D
A
A
Yeah,
basically
being
able
to
just
abstract
above
like
the
VA
infrastructure,
so
people
don't
have
to
see
that
or
deal
with.
It
is
the
way
that
we're
looking
at
it.
So
it
is
different,
but
like
the
cluster
configurations,
that's
where
it
can
become
a
little
bit
confusing
me
and
Robbie's
joined
Robbie
I,
hello,
hi,
right,
okay,
everyone!
A
This
is
Robbie,
so
he's
one
of
the
main
maintainer
x'
of
the
project,
so
I
want
to
introduce
him
because
I
think
he'll
be
able
to
come
to
every
week
every
weekly
meeting
and
he
has
more
of
like
technical
details
of
what's
going
on.
He
understands
that.
Do
you
want
to
ask
your
question
again
in
the
media,
Robbie
cancer,
sure
I,
guess.
D
A
E
E
G
G
E
H
Know
how
you
follow
coupon
idea
scheduling
all
those
kind
of
things,
because
this
guy
to
me
you
need
to
have
the
total
capacity
I
locatable,
and
then
you
have
the
common
requester
right
against
I.
Don't
you
have
the
tender
against
that
I'm,
not
sure
this
is
how
you're
going
to
you
follow.
Kubernetes
countouring
model
scheduling
model
and
any
model
can.
C
I
answer
that,
because,
like
I
had
to
deal
with
this
with
what
I
did
you
could?
Basically,
it's
basically
good
comms,
an
infinite
resource
node
like
when
I,
when,
when,
when
Burns
demoed
ACS
and
the
general
meeting
I
asked
about
like
how
it
was
implemented,
and
he
said
basically
they
are
running
individual
containers
or
that
pods
within
individual
VMs.
So
basically
it
doesn't,
it
doesn't
have
you
can
think
of
it
as
a
node,
which
has
no
resource
limitation.
F
There's
a
lot
of
ways
you
can
satisfy
this
model,
though
right
like
you're,
basically
discovering
the
virtual
cubelet
as
like
a
meta
node
or
an
auto
cloud.
Autoscaler
and
like
you
could
go
in
dynamically
provision
of
virtual
node
or
a
fake
note
in
the
kubernetes
api
construct
itself.
That
has
a
max
pod
constraint
of
one
and
it's
advised
exactly
to
meet
a
pod
size
request
and.
F
F
Don't
mean
what
we
think
they
mean
just
introduce
confusion
or
user
error
like
months
down
the
line
so
like
trying
to
make
something
that
is
a
node
not
be
a
node
feels
like
a
little
concerning
versus
like
another
model
where
you
could
have
something
that
fulfills
like
the
vision
of
a
virtual
qubit,
but
it
presents
literal
node,
remember,
cube
control,
plane
that
might
have
like
scheduling
constraints
imposed
on
them.
So
you
might
only
support
one
pod.
You
might
support
only
specific
set
of
taints
for
that
certain
pods
only
land
on
them.
E
B
C
I
H
The
cute
differentiator
here
yeah,
they
didn't
really
actually
they're
an
actor,
know
the
management
and
they're,
not
just
management.
One
part.
They
are
not
a
fake
there.
It's
the
new
node
join
I'd,
caster
and-
and
they
are
not-
and
also
under
note,
we
we
will
stand
off
those
kind
of
things.
We
still
need
the
Cuban
eight,
the
know
the
manager
to
report
back
custom
manager
to
here
it
is
the
usage,
all
those
kind
of
things,
and
so
it's
quite
different.
Each
one
have
their
own
role
and
acting
there
as
themself.
H
C
C
Don't
translate
necessarily
easily
to
this
model,
especially
the
way
I
did
it
the
way
Microsoft
is
doing
it
it's
possible.
They
can
translate
better,
especially
because
I
was
using
just
a
regular
qubit
with
CRI,
so
things
like
attaches
were
being
dumb
would
be
dumb
like
if
you
were
using
an
EBS
volume,
it
would
be.
It
would
tell
Cuba
to
do
the
attach
like
that.
C
Cri
has
no
knowledge
of
this,
so
it
fell
down
for
me
but
like
in,
if
you
have
sort
of
your
own
full
cubelet
doing
these
things,
like
you
can
say:
oh
you
want
me
to
attach
volume,
x4,
pod,
Y
or
because
it
eventually
be
mounted
for
pod.
Why
I
can
basically
proxy
this?
The
issue
is,
is
that
you
can
end
up
with
sort
of
race
conditions.
Cuz
attach
is
used
as
a
locking
mechanism
to
prevent
things
from
trying
to
get
to
get
both
and
as
a
scheduling,
I
guess
barrier.
C
F
To
meet
on
the
gap
you'd
run
into
ISM.
If
you
want
to
drive
really
dense
clusters,
then
you
would
have
the
appearance
of
like
a
very
large
number
of
nodes
that
need
to
drive
checkpointing
operations
back
to
the
control
plan.
So,
like
I,
don't
know
what
the
scale
limits
you
were
trying
to
push
with
the
virtual
culet
itself,
but
like
if
you
you're,
going
to
drive
like
a
thousand
pods
behind
a
virtual
cupid
interface.
If
you
did
manifest
those
all
as
unique
notes,
then
you've
kind
of
overwhelmed
aren't
scaling
characteristics.
B
Agree
that
that
would
be
one
of
the
big
challenges
to
trying
to
scale
one
pot
per
node
model,
but
I
think
that
that
there
are
issues
that
could
be
solved
there
and
any
progress
or
it's
making
that
situation
better
would
benefit
the
whole
of
kubernetes
right.
We
would
solve
scaling
problems
for
the
non
one
pod
case.
It
would
generally
improve
performance
versus
this
approach,
which
is
kind
of
like
putting
that
effort
into
a
lot
of
the
cubelet,
and
there
will
kind
of
always
be
these.
H
I
H
F
J
Like
what
they
are,
I
also
have
a
question,
so
I
mean
it
seems
like
all.
This
could
be
done
unilaterally
outside
kubernetes
whoa.
What
what
needs
to
be
in
base
just
I
mean
I,
guess
why
I
mean
seems
like
someone's
good.
You
should
just
go.
Do
this?
What
needs
to
happen
inside
inside
kubernetes,
kubernetes,
I,
guess
and
then
like
just
penis
into
like
a
stable
of
cubelet
yeah
and
we're
gonna
have
to
version
that
and
things
now,
because
we
have
multiple
cubelets
and
people
don't
want
to
rewrite
their
virtual
cubelets.
Every
time
we
bump
kubernetes.
K
I
agree
with
this:
you
know
that
CRI
should
be.
You
know
also
the
host
for
virtual
machines,
but
we
haven't
really
got
that
far
along
in
the
API
on
CRI.
For
you
know,
for
PM's
is
that
is
that
the
root
is
this.
The
result
is
virtual
cubelet,
the
result
of
cry
not
you
know
not
handling
virtual
machines
and
first-class
way.
C
So
I
did
implement
it
with
like.
If
you
look
like
I've
shared
it
before
I'll
share
the
link.
Again,
it's
not
up
to
date.
It's
the
last
commit
I
did
to.
It
was
one
seven
time
frame.
Well,
like
you
can
do
this
via
the
CRI,
though
there
are
hacks
it
like
I
had
to
hack
it's
a
bit
package
to
get
it
done
so
I
I
guess.
F
A
A
F
Sponsor
repos,
then
core
repos
so
like
to
scope.
This
discussion,
I
can't
tell
like
if,
if
the
project
maintainers
are
asking,
you
know
to
let
sponsor
this
work
and
I,
don't
know
like
as
a
group.
If
that's
something
people
want
to
do
or
not,
I
don't
have
to
have
a
deeper
discussion
on
that,
but,
like
generally
speaking,
to
go
back
to
sus
question
like
should
we
be
thinking
about
how
how
you
want
to
change
team
itself
to
support
your
use
case
and
like
if
those
things
were
identified?
K
A
We
can't
come
back
so
basically,
what
are
you
guys
missing,
like
the
customer
use
case
right
now,.
I
A
Ok,
ok,
oh
yeah,
I
understand
that
the
sig
standpoint
cuz
I,
also
like
don't
know
from
a
customer
standpoint
like
what
is
going
up
in
a
sick
note
and
what's
more
important,
but
from
a
virtual
COO
at
some
point
and
why
it's
being
implemented
and
why
we
really
think
it's
important
is
the
fact
that
people
now
don't
have
to
manage
VMs
in
their
cluster
and
that's
the
way
that
we're
thinking
of
it
and
like
they
don't
have
to
do
upgrades
or
updates
or
patching
or
whatever
for
their
agent
notes,
because
that's
all
hidden
from
them
all
they
see
are
their
pods
and
that's
all
they
have
to
deal
with.
A
A
L
K
L
H
So
sick
action
people
talk
to
me
and
suggest
they
talk
to
the
note,
if
the
note,
because
the
first
start
it
is
the
change
note
and
also
change
accumulate,
so
they
think
about
want
to
stick
know
the
community
start
the
first
to
say
this,
how
we
potential
collaboration-
oh
oh,
like
collaboration,
and
if
the
signal
to
think
about
this
is
not
to
signal
the
one
to
sponsor,
then
there
may
be
the
virtual
Cooper
net
project.
I
have
to
look
into
some
other
sponsorship.
L
H
I
want
to
answer
your
earlier
question
and
the
human
society.
Yes,
when
we
walk
honest,
yeah
I,
remember
your
demo
and
I
even
remember
you
came
to
to
our
site
and
give
the
demo
the
when
we
start
a
signal
to
say
I,
we
do
talk
about
some
potential
after
say:
I
could
be
the
remoter
see.
I
think
it
involves
the
email
to
communicate.
Also
mention
that.
But
it's
not
there's
a
lot
of
tons
of
the
problem,
and
so
we
didn't
make
any
progress
on
that
one
and
so
so
back
to
the
but
the
profit.
H
It
is
the
profit
it
is
when
you
do
them
on
with
the
end,
we
encourage
that
one.
It
is
because
Cooper
neck
is
open-source
project.
We
want
to
encourage
of
innovation.
Now
you
ask
why
we
push
back
this.
Here's
the
really
because
when
we
talk
about
this,
when
it
has
become
to
the
community
or
group
for
sick
group
or
sponsor
party,
then
we
need
to
think
about
is
the
real
use
kisses
by
capital?
H
Is
the
real
production
use
kisses
and
it's
the
real
user
waiting
for
this
solution
and
if
this
is
aligned
with
kubernetes
api?
Today's
concept
and
the
design
and
the
architecture,
so
that's
why
we
have
to
think
about
this
one.
So
it's
not
really
push
back,
which
is
question
because
we
think
community
the
use
cases-
and
we
do
understand
that-
maybe
the
component
use
case,
but
is
that
strong
enough,
like
the
people
mention
that
not
imagine
the
air
can
we
can
we
fix
that
problem
through
some
other
way
in
standard?
H
Next
we
have
to
change
the
whole
architecture,
design,
change,
the
control
plan
and
the
change'
caster
api
and
the
tubes
and
the
chino
the
api
to
fit
that
the
model
and
there's
the
many
things
that
management
am.
Is
that
the
common
ideas
first
design
philosophy
here?
So
a
lot
of
things
needed
a
single?
It's
not
like,
though
we
tried
to
push
back
is
just
me.
We
need
to
ask
ourselves
question
so.
C
C
A
Another
use
case
for
us
is
scaling
really
fast,
so
we
have
the
way
that
ACI
works.
We
have
all
these
warm
pools
of
clusters,
they're
running
kubernetes
in
the
backend,
but
basically
we're
able.
F
A
F
The
architecture
you
have
to
me,
like
the
delta,
is:
how
quickly
can
you
bring
a
node
to
a
cute
cluster
breakfast?
How
quickly
can
you
spawn
your
container
on
one
of
these
warm
pools
so
like
if
you
have
like
node
pools
that
are
already
warned,
can
they
just
dynamically
shift
in
a
line
which
cluster
they
serve
or
are
you
trying
to
have
these
node
poles
serve
many
clusters
like.
A
C
To
the
point
would
be:
is
that
you're
only
paying
for
the
exact
resources
needed
for
the
VM
that
you
are
using
as
opposed
to
a
if
you,
if
a
person
himself
would
scale
this
out
like
himself
or
have
sort
of
warm
nodes
waiting
in
the
wings
to
be
used?
They
would
have
they
would
be
paying
that
cost
24/7,
for
when
it's
not
needed.
Is
that
correct,
yep.
F
B
It
seems
like
the
first
step
would
be
to
develop
a
formal
specification
for
the
cubelet
and
all
of
the
cubelets
behavior,
which
is
not
something
we
currently
have.
So
you
know
iterating
over
every
feature
and
and
really
getting
down
into
the
details
of
how
those
are
expected
to
work,
and
you
know
if
there
are
optional
features
than
declaring
Moses
so,
but
like
really
defining
what
the
expectations
of
a
node
in
the
cluster
is.
A
E
D
D
F
F
E
To
the
cluster,
it's
just
an
alternative
and
who
example
like
completely
mad
for
secrets,
or
these
information.
The
virtual
couplets
will
stretch
from
the
kubernetes
from
the
api
server
and
the
standard
to
the
provider
and
the
provider
will
decide
how
to
use
that
either.
Point
on
for,
like
the
a
CI,
translates
all
down
into
a
secret.
E
F
F
Logging
right
layer
for
my
container
read
only
or
could
I
could
I
exhaust
everything.
I'd
want
to
write
to
you
on
my
copy-on-write
layer
and
then
induced
like
local
disk
pressure
and
then
just
how
all
these
things
like
come
back
up
to
the
cluster
control
plane.
I
think
these
are
the
tension,
points
that
that's
him
myself,
Don
and
probably
David
and
Seth
are
raising,
but
and
and
that's
we're
having
a
formal
I
expect
because
it
PI.
H
H
A
F
A
A
Pots
within
their
cluster
virtual
pots
or
virtual
on
couplets,
basically
so
that
right
now
it's
targeted
as
unpredictable
burst
workloads,
and
it
could
be
like
that
for
a
while
there
I
mean
you
could
just
be
running
all
your
long-running
services
on
your
agent
and
your
cluster
and
then
burst
out
through
virtual
couplet
when
you
need
to
so
that
could
definitely
be
a
use
case.
That's
just
forever
goes
it
really.
F
F
So
they
have
any
other
things
you
want
to
discuss
on
this
topic.
I
guess
it'd
be
good
to
get
more
details,
I,
guess
and
then
probably
good
to
understand
how
what
like,
if
there's
a
formalized
goal
of
having
the
sake
sponsored
of
some
kind
like
we
should
probably
talk
through
what
that
process
would
look
like,
but
there
any
other
topics
for
for,
like
today's
discussion,
I
cover.
C
Yes,
I
have
a
hackish
microphone
set
up,
there's
been
a
like,
I
guess,
confusion.
People
have
like
that.
They
don't
necessarily
see
how
I
turn
ends
like
kubernetes
concepts
like
necessarily
easily
fit
in
such
as
like
daemon
sets
being
a
prime
example,
but
the
thing
I
would
point
out
to
think
about.
C
Is
that
like
this
could
apply
to
if
you're
dealing
with
heterogeneous
clusters
right,
if
you
had
like,
like
docker
like
a
normal
doctor,
shim
node
and
window
and
a
Windows
node
all
in
the
same
cluster
you're,
not
going
to
run
a
daemon
set
across
the
whole
cluster
so
away?
It's
really
viewing
this
as
sort
of
a
heterogeneous
node
that
yeah
we,
you
might
be
running
things
in
a
daemon
set
or
other
ways,
but
they
just
don't
apply
to
this
type
of
virtual
node.
C
And
it's
it's
thinking
in
terms
of
we
like
I
guess
when
we
approach
this,
we
should
think
of
it
in
terms
of
heterogeneity,
because
I
think
in
practice,
kubernetes
is
going
to
have
to
move
in
that
direction.
We've
it's
been
today
told
I,
guess
recently
very
homogeneous
clusters,
but
I
think
you
know
I,
don't.
F
Think
that
that's
true
I,
think
the
mental
leap
for
me
is
like
no,
that
wants
to
prevent
present
to
container
runtimes
at
the
same
time
or
like
a
node
that
wants
to
present
to
OSS.
At
the
same
time
like
you
can
support
heterogeneous
clusters.
Just
fine
and
yes,
users
have
to
like
learn
to
target
their
daemon
sets
appropriately
to
the
cluster,
but
you
haven't
like
foundationally
changed
the
meaning
of
the
net
of
the
node,
primitive
and-
and
this
is
basically
challenging
us
to
change
the
meaning
of
the
node
primitive.
Yes,.
H
I
agree
and
also
I
I
know
earlier
people
I
believe
people
bored
of
the
daemons
side.
It's
not
I
mean
on
everything
inside
they
just
actually
mean
them
function
noted
here
those
basic
or
functionally
active,
represented
by
those
demon
side.
For
example,
you
don't
need
a
Koopa
proxy.
It's
not
all
the
kubernetes
cluster
needed
to
know
proxy,
but
those
cannot
have
the
routing
functionality
how
you
are
going
to
get
it.
H
So
that's
why
we
asked
of
ask
and
also
the
and
if
the
node
have
the
problem,
how
you're
going
to
report
that
no
the
problem
is
the
anyway.
You
are
going
to
report
those
no
to
addition,
and
we
just
want
to
understand
those
kind
of
things.
So
that's
kind
of
that
we
try
to
define
our
notice
back
for
neck
last
a
couple
years
and
how
we
manage
military
sauce,
how
we
management,
know
the
Cartesian
and
and
if
there's
the
problem,
how
we
diagnose
those
issues
and
the
providing
reliable
support
to
the
customer.
B
This
is
a
little
bit
what
I
was
getting
at
with
the
talking
about
a
formal
specification.
You
said:
maybe
we
could
say
that
demon
sense,
you
don't
expect
to
run
demon
sets
on
a
virtual
node
right
now.
The
expectation
is
that
demon
sets
run
on
every
node
in
the
cluster
and
so
I
think
we
need
to
decide
whether
that's
a
feature
that
is
acceptable
to
to
break
in
a
conforming
communities
and
they.
H
And
another
thing
it
is:
the
issue
is
the
first,
my
reaction
less
so
this
the
today's
proposal.
I
would
just
think
about.
Okay,
we
have
the
conformance
test
and
don't
even
talk
about
caste
level
of
the
compound
pastor
and
just
know
the
Iowa
confirm
tester
how
are
going
to
want
to
get
those
things,
and
so
we
have
the
node
level
confirm
test
and
I
believe
a
lot
of
is
going
to
broker
ISO
so
how
we're
going
to
to
validate
to
provide
conform
for
this
way.
H
So
that's
why
I
think
that
I,
just
like
the
like
what
Tim
said,
we
need
our
formal
virtual
kubernetes
back
here
and
then
we
understand
it
as
well.
What
do
you
require
change,
what
it
is
resume
and
then
we
are
going
to
come
back
to
discuss
this
one
and
direct
I,
don't
know.
What's
the
problem
new
process
for
those
for
this,
your
committee
to
design
for
that,
okay,
/,
sick,
a
grouper
sponsor
project,
how
we're
going
to
go
through
process.
So
but
at
least
I
think
we
need
to
say
the
virtual
coupon
is
back
right.
H
F
F
F
So
like
it's,
it's
transparent
and
we
can
use
it
as
a
rubric
going
forward,
but,
like
I,
think
that
the
the
clear
speck
and
then
having
our
individual
charter
identified
would
probably
be
Purex
to
then
figure
out
what
to
do
with
with
this
project
moving
forward
and
maybe
maybe
dawn.
We
should
set
a
goal
for
signo
to
get
our
charter
done
in
the
next
like
four
weeks
or
so
sure.
F
F
The
only
thing
I
know:
I
apologized,
on
I-
tried
to
get
material
together
today
for
Tim
to
give
feedback
on
the
sidecar
elevated
privileges.
There's
a
rather
spirited
internal
email
thread
here,
but
had
getting
all
of
their
perspectives
on
the
matter.
I.
Imagine
every
company
you
had
a
similar
one,
and
so
I
will
try
to
get
that
done
for
tomorrow
and.
F
It's
more
like
the
tension
is
yes
every
case.
Where
it's
been
called
out.
You
could
ask
the
component
that
requires
it
to
be
written
rewritten
to
not
require
it,
but
then
it's
a
matter
of
like
is
that
actually
viable
or
not
there's
about
six
or
seven
use
cases.
We've
had
and
I
can't
say
that,
like
we
firmly
feel
one
way
or
the
other
one,
if
it
should
be
proceeding
so
in
the
dock,
I'll
just
list
the
pros
and
cons,
we've
discussed
and
then
Tim,
you
could
collapse
that,
where
you
I'm.