►
From YouTube: Kubernetes SIG Node 20201110
Description
Meeting Agenda:
https://docs.google.com/document/d/1j3vrG6BgE0hUDs2e-1ZUegKN4W4Adb1B6oJ6j-4kyPU
B
It's
a
tradition
for
every
meeting:
right:
okay,
hello,
it's
november,
ninth
or
tenth,
it's
a
signal,
meeting
pr
status.
So
today
we
have
like
we
have
two
weeks:
statistics
not
one
week,
because
last
week
was
skipped
and
59
pr's
were
created
and
46
were
merged.
So
it's
a
pretty
good
number
keeping
in
mind
that,
like
some
of
them
were
closed,
so
we
again
above,
like
closing
immersion,
more
prs
than
creating,
which
is
a
good
trend.
B
Let's
keep
it
up
and
I
really
want
to
emphasize
that
we
have
29
lgtm
prs,
some
of
them
lgtm
for
a
long
time,
and
we
really
need
to
have
somebody
to
take
a
look
and
start
approving
those
prs.
We
also
have
some
prs
and
test
area.
We
discussed
it
on
ci
group
yesterday
and
the
week
before
that
doesn't
have
enough
attention.
B
So
whoever
interested
in
reviewing
test
prs,
which
is
very
important
for
reliability
of
kubernetes,
please
take
it
a
shot
and
I
can
paste
a
specific
url
if
you
want,
but
it's
pretty
hard
easy
to
find.
C
So,
on
the
pending
approval
that
have
lgtm,
I
had
spent
a
fair
amount
of
time
yesterday,
going
through,
like
a
bit
of
the
backlog
and
trying
to
find
everything
that
I
could
approve
a
number.
I
don't
know
if
you
have
stats
sergey
for
things
that
required
approval
in
parts
of
the
kk
tree
that
are
not
just
to
the
sig,
but
I
think
that
that
list
should
be
much
smaller.
I
don't
know
if
you
have
a
way
of
tracking
that.
A
I
also
went
to
through
some
of
the
alrighty
looks
good
to
me
here,
but
there's
some
something
like,
for
example,
runtime
class.
It
is
looks
good
to
me,
but
there's
actually
still
having
a
question
raised
by
reviewer,
but
it's
not
answered.
So
that's.
Why
also
didn't
feel
comfortable
just
approval,
like
the
also
circuit.
You
brought
some
topical
here
today
right,
so
so
so
this
is
why
we
cannot
just
look
at
the
the
pis
looks
good
to
me
and
then
just
approval,
so
so
so
a
lot
of
time.
I
look
at
that.
A
A
If
the
original
author
can,
if
there's
the
answer
and
also
achieve
the
agreement
and
please
write
down
and
the
comment
and
summarize
the
decision
and
it's
much
easier
for
fellow
approver
and
to
give
approval
and
even
like
the
some
open
pricing
at
least
make
a
point
there.
So
this
is
the
open
question,
but
we
decided
and
so
that's
kind
of
much
easier
to
reasonable.
Otherwise,
the
fellow
approver
basically
have
to
really
read
the
whole
code
to
understand
why
this
is
not
not
and
addressed.
B
Yeah,
it
makes
sense,
maybe
you
and
derek
you
can
drop
a
comment
saying
it
doesn't
seem
relevant
or
like
it
still
have
open
questions.
So
it
may
happen
that
some,
like,
I
said
people
believe
that
all
questions
are
answered,
but
in
reality
it's
not
clear
from
my
pr
status.
C
Yeah,
mostly
sergey,
my
comment
was,
the
query
you
were
tracking
here
is
everything
that
was
labeled
with
sig
node,
but
not
exclusively
just
label
with
signature,
and
so
a
number
of
these
things
have
api
changes
that
were
being
held
up
or
spanned
a
couple
other
sigs,
and
so
I
think
from
my
perspective,
I
thought
we
were
the
takeaway
here
for
me
is,
I
feel,
like
we're
getting
in
a
cleaner,
better
spot
and
hopefully
that
those
on
the
call
can
feel
that
way
as
well
and
so
yeah
anyway.
B
Think
how
how
to
track
that?
I
I
see
what
you
like
cherry
picks.
For
instance,
they
are
all
gtm
for
a
while,
but
they're
stuck
on
the
release
managers,
for
instance,
that
kind
of
situations
right.
A
A
Is
so
so
so
so,
but
that
one
I
don't
concern
about,
we
don't
give
approval.
We
give
approval
anyway,
the
fellow
call
it
is
renis
manager.
So
it's
not
the
concept.
My
most
concerned,
it
is
some
is
required
about.
Like
not
answer,
questions
like
the
address
the
question.
No,
no
last
decision
a
comment
made
there.
Another
thing
it
is
like
a
shared
duty
like,
for
example,
api,
and
the
summit
is
like
the
shared
with
the
security,
the
us
seekers,
so
that's
kind
of
the.
A
I
don't
want
like
the
where
people
over
the
problem
and
just
approval,
then
the
things
get
merged
the
better.
Actually,
the
some
corresponding
team
didn't
look
into
and
didn't
agree.
So
that's
the
problem
here
yeah,
but
I
agree
with
direct
overall.
I
think
we
are
getting
better
and
better.
C
Yeah
and
maybe
before
we
dive
into
the
early
topics
here,
I
was
trying
to
focus
on
what
we
wanted
to
close
out
in
120
this
week.
Given
that
there's
code
freeze,
I
want
to
thank
kevin
for
taking
a
pass
at
the
memory
manager,
pr
which
I
know
folks
in
the
community
we're
looking
for,
and
hopefully
we
can
unblock
that.
I
think
I
got
all
of
tim
eau
claire's
stuff
unblocked.
C
I
know
renault,
you
have
a
item
here
on
cri,
but
if
folks
can
open
a
section
in
the
docs
or
on
the
agenda
here,
folks
can
just
link
to
prs
that
we
might
be
missing
that
we
wanted
to
land
this
week.
I
think
that
would
be
helpful
for
ensuring
we
can
get
priority
focus.
Otherwise,
I'm
just
going
by
what
kubernetes
shows.
D
So
one
pr
to
call
out
derek
is
that
the
pr
to
support
exact
plugins
for
fetching
registry
credentials,
I
think,
is
close
to
merging
jordan-
has
been
reviewing
it.
He
thinks
the
api
is
in
a
good
spot,
but
I
do
think
either
you
or
don
should
review
the
you
know
the
cube
level
changes
for
that.
One.
A
D
C
And
then
the
other
question
I
don't
know
if
anyone
from
sig
windows
is
here
is
there's
a
change
open
right
now
for
windows,
support
to
update
container
resources,
which
I
thought
also
touched
cri.
I
didn't
know
if
mernal
had
looked
at
that
at
all,
and
maybe
we.
E
I'm
I'm
here
from
from
windows.
I
think
that's
that's
a
forward-looking
change
that
we
need
in
the
cri
to
be
able
to
resize.
You
know
container
pods
and
everything,
so
that
was
just
cri
changes
that
there's
no
implementation
behind
it
at
this
point
in
kubernetes,
but
there
will
be
work
being
done
in
container
d
to
use
those
changes
after
the
cri
changes
go
in.
C
E
C
Okay
and
then
maybe
dems
we
should
figure
out.
I
I
wanted
to
make
sure
the
cri
beta
transition
happened
before
we
deprecate
your
do
your
pr,
so
maybe
we
can
just
make
sure
we
cover.
G
C
Okay,
all
right!
Well,
then,
we're
good
on
that
we
can
go
through
the
giant
items.
C
So
the
first
item
I
had
thrown
on
here
last
week-
we've
had
a
lot
of
discussion
internally
here
at
red
hat,
about
various
use
cases
that
might
potentially
benefit
from
kubernetes
having
more
intelligent
usage
around
swap
and
that
that
caused
me
to
look
back
at
the
issue,
which
is
kind
of
reinforcing
that
there
were
a
number
of
participants
in
the
community
who
seem
surprised
when
cube
does
not
recommend
supporting
swap,
and
so
what
I
was
curious
about
is
if,
if
there
are
a
group
of
folks
that
wanted
to
potentially
get
together
and
kind
of
put
together
a
cap
on
how
we
could
handle
this-
and
maybe
we
could
get
like
a
a
small
group
and
a
head
start
on
this
for
for
121..
C
So
some
of
the
workloads
that
we
were
debating
around
this
was
things
like
the
kubert
project
as
an
example,
and
then
some
databases.
So
I'm
kind
of
curious.
If
there's
any
on
the
call
that
maybe
represents
communities
that
might
be
beneficial
from
better
supporting
for
swap.
H
Well,
I
am
interested
in
the
last
topic
specifically
in
like
not
looking
conventional
but
more
in
terms
of
that
sort
of
implementation
and
second
part.
There
is
also
with
different
types
of
memory
like
on
the
same
system
when
they
have
a
fast
and
slower
memory.
It
also
have
the
same
scenarios
like
with
the
swap
and
this
promises
to
reports
on
what
kind
of
memory
and
how
much
it
should
be
using.
C
So
maybe
what
I'll
do
is,
if
there's
interest,
I
could
start
a
google
doc
together
that
we
could
kind
of
capture
individual
use
cases.
One
of
the
things
I
know
marinol
and
I
were
debating
was,
if
swap,
should
actually
be
a
schedulable
resource
or
some
some
fixed
amount
of
overhead
that's
available
on
the
node,
but
maybe
in
general.
The
ask
here
will
be
if
I
throw
the
stock
here,
if
folks
are
actually
interested
in
closing
on
this,
we
can
get
use
cases
settled
in
events
of
121.
I
Yeah,
I'm
here
dave
gilbert
from
the
qmu
side
of
things,
and
you
know
I
raised
this
with
derek
and
one
of
the
reasons
is
from
qmu.
We've
got
a
very
loose
definition
of
how
much
ram
that
we'll
use.
I
While
previously
people
have
tried
to
get
heuristics
of
the
amount
of
ram,
we'll
need
everybody's
previously
run
away
and
given
up
and
added
some
swap
and
certainly
on
open
stack.
C
Yeah,
so
that's
coming
from
the
use
cases
around
the
cuber
project,
but
I
know
that's
just
one
of
many
projects
that
seem
to
desire
some
amount
of
swap.
So
maybe
then
it's
a
next
step
here.
It
sounds
like
there's
reasonable
injuries,
they'll
throw
a
dock
together
and
place
it
in
here
and
if
we
can
get
use
cases
settled
on
that
will
help
them
form
a
proper
cup.
C
J
Next
hi
everyone,
so
this
is
just
a
follow-up
from
the
discussion
last
week
about
pod
level
metrics
in
the
resource,
metrics
endpoint
and
specifically
this
is
just
a
question
about
naming.
We
have
two
competing
naming
schemes.
J
J
J
C
Yeah,
let's
try
to
follow
up
with
him
on
slack.
I
know
I
was
talking
about
this
a
little
bit,
but
I
think
we
just
need
to
get
a
group
of
us
together.
K
If
you
can
add
me
or
or
just
I
just
want
to
watch
so
I
can
update
the
pr.
My
main
thing
is:
I
just
wanted
to
get
this
in.
I
was
trying
to
be
consistent,
so
yeah
solve
this
in
the
next
day
or
so
that's
my
main
concern
I'm
using
this
downstream
now,
and
I
want
to
make
sure
that
this
gets
in
eventually.
C
Yeah
so
erica-
I
think
I
know
clayton
is
around
today,
so
hopefully
we
can
get
all
four
of
us
on
the
same
page.
I
know
when
I'm
here
pr,
I
was
thinking
he's
working
in
a
similar
spot,
so
I
just
want
to
make
sure
we
get
all
aligned.
C
F
F
So
what
we
need
to
do
is
we
need
to
add
a
parallel
v1
beta
1
in
addition
to
the
v1
alpha,
2
proto
then
get
the
get
container
d,
updated,
cryo,
updated,
bring
them
back
into
the
ci
and
then
switch
over
the
cubelet
and
the
rest
of
the
code
to
v1
beta1,
so
the
so
the
pr
I
have
in
the
dock
right
now
is
adding
that
photo.
So
we
need
to
get
that
in.
F
But
now
my
worry
is
that
what
do
we
do
for
the
runtime
changes
like
mike
brown
was
working
on
it,
but
he's
out
this
week.
So
I
was
wondering
if
there's
anyone
on
call
from
container
d
that
can
follow
on
the
containery
changes
to
bring
them
back
into
this
year.
C
F
C
G
C
F
F
I
would
so
we
can
add
the
proto.
Maybe
then
update
it
in
the
next
version,
but
is
it
okay
like
if
you
think
about
the
changes
we
added
to
the
beta?
It's
just
incremental
two
three
changes
to
v1
alpha
2..
So
is
it
okay?
If
we
get
in
this
change
and
go
beyond
the
code
free
state
to
update
the
runtimes,
or
are
we
okay
waiting,
one
more
cycle
to
get
the
runtimes.
F
A
It
is
just
so
we
just
think
about
it.
It's
just
too
hard
to
go
through
to
reni's.
A
So
manu
it
is
basically
you
just
call
out
to
someone
to
partner
with
you
and
to
add
this
corresponding
support
and
release
on
continuity.
F
F
C
Yeah
right,
if
we
do
the
gi
step
two,
because
basically
I'm
also
assuming
I
mean
you're,
going
to
make
these
changes
and
then
we
need
someone
to
vendor
them,
and
people
would
only
want
to
vendor
them
from
stable
releases.
F
F
We
didn't
have
container
d
or
cryo,
which
is
not
blocking
yet
in
ci
back
then
so
we
didn't
hit
it
all.
The
changes
were
in
docker
shim,
so
docker
shim
was
updated
along
with
the
alpha
two
changes.
C
F
Yeah
on
the
runtime
side,
I
think,
like
container
d
may
prefer
to
support
both,
since
they
are
not
as
strongly
tied
to
the
cumulative
version
as
cryo
is
but
again
like
that's
the
impression
I
got
from
talking
to
mike,
but
he's
not
around
this
week
so
but
on
the
on
his
side,
we
should
just
stick
to
one
version.
I
think.
A
A
A
F
C
F
C
A
And
also
student,
like
the
csi
cni,
have
the
same
problem
even
worse
problem,
because
we
only
have
the
two
implementation
so
far.
We
we
start
from
three,
but
four,
then
later
long
only
have
two.
So
so
student
like
the
csi,
have
much
more
vendor
to
support
that
they
should
have
the
even
worse
problem
than
us.
F
Yeah,
so
I
think,
like
some
of
this
probably
plays
into
how
container
d
wants
to
manage
the
releases,
so
maybe
next
week
when
mike
is
back
or
we
have
representation
from
them,
we
can
have
a
deeper
conversation.
Meanwhile,
we
can
merge
this
and
I'll
bring
folks
there
to
see
if
there's
interest
in
rendering
this
pr
and
making
the
changes,
if
possible,.
F
A
You
I
believe,
by
design
shouldn't
affect,
because
that's
how
we
design
initial
design
continuity
and
especially
when
they
say
I
continue,
because
we
try
to
treat
that
as
the
plug-in
and
under
support
multiple
version
even
different
have
to
give
the
possibility
even
different
api
can
plug
in
so,
but
still
who
knows
things,
the
change
might
change
over
time,
because
everything's
changing
so
so
yeah.
So
I
want
to
make
sure
it's
not
the
hard
production
yeah.
B
B
C
A
B
Do
we
need
to
release
new
container
manager
that
basically
copy
paste
of
the
existing
one?
So
no
like
a
runtime
selection,
but
by
feature
flags
we
can
at
least
make
cooperate,
uses
new
beta1
interfaces,
so
people
can
test
it
at
least.
F
A
Let's
carry
on
this
on
your
issue:
it's
a
menu,
maybe-
and
then
we
can
discuss
through
the
issue
together
with
the
container
d
community-
is
that
okay,
okay
yeah
so
circuit?
Do
you
want
to
talk
about
the
runtime
class
here.
B
Oh
yeah,
runtime
class
is
just
basically
pr
that
was
already
looked
at
couple
times,
so
minor,
tweaks
from
the
big
original
pr,
but
it's
generally
was
lgtm
couple
times
already.
So,
if
you
have
time,
take
a
look
and
I
was
wanted,
I
want
to
ask
whether
we
need
to
keep
alpha
api
around
technically.
We
can
remove
it
right
away.
There
is
no
support
window
like
we
need
to
keep
api
around
four,
so
beta
needs
to
be
kept
around
for
like
two
releases,
I
think,
but
alpha
can
be
removed
right
away.
E
On
on
windows,
we're
doing
a
lot
of
we're
using
runtime
classes
to
pass
extra
annotations
along
yeah,
and
I
think
it
would
be
helpful
to
have
one
release
of
an
overlap
for
some
of
the
work
that
we're
doing
with
that.
B
We're
going
to
ga
so
there
are
three
versions
right
now
I
mean.
E
B
Okay,
once
this
pr
is
nourished,
hopefully
this
week,
I
will
stop.
E
A
So
so
so
sounds
like
the:
where
consensus
on
this
one
remove
the
we
want
alpha
one
and
then
it's
multi-text
tag
next
topic
and
the
statewide
the
matrix.
M
Yeah,
that's
me
so
I
know
that
dashpol
has
had
an
issue
open
for
a
few
years
to
reduce
the
metrics
that
are
exposed
by
the
kublet
and
now
there's
a
cap
open
to
fully
deprecate
the
sea
advisor
endpoints,
which
includes
stats
container,
which
is
a
source
of
detailed
resource,
container
resource
utilization
metrics.
M
M
A
A
So
this
is
why
that's
kind
of
when
we
design
long
time
now
long
time
back
go.
You
have
the
some
huge
debate
that
signaled
and-
and
so
that's
how
we
decided
back.
Then
we
over
time
we
didn't
really
deprecate.
There's
the
corporation.
One
reason
is
the
performance
like
the
even
today
people
still
see
the
observer
of
the
continuity
resource,
have
some
problem
and
performance
and
consent.
So,
but
another
way
is
just
priority,
so
I
just
give
show
you
some
context.
J
J
J
If
there's
information
that
isn't
available
in
the
prometheus
endpoints
that
was
available
in
the
previous
json
endpoints,
especially
if
you're
relying
on
it,
that
would
be
useful
to
know
there
is
the
possibility
that
we
can
add
small
amounts
in
order
to
make
sure
that
people
have
migration
paths.
So
that
would
be
helpful
information.
If
you
can
add
that
to
the
cap
or
comment
on
it,.
M
Yeah,
that's
fantastic.
I
wasn't
aware
that
it
was
a
consolidation.
A
J
If,
if
you
do
prefer
a
json
api
one,
I
would
be
very
interested
to
know
why
you
prefer
it,
but
two
there's
also
the
cubelet
summary
api,
which
is
also
available
at
slash
stats.
Summary.
M
A
So
much
I
also
want
to
know
what
kind
of
the
details
you're
looking
for
and
the
summary
api
is
missing
and
that's
so
I
just
want
to
say
that
we
actually
do
want
to
initially
when
designs
there
are.
We
do
think
about
the
even
like
the
first
one
prometheus
matrix,
remove.
We
do
concern
about
the
backward
compatibility
and
also
some
customers
rely
on.
We
didn't
expect
people
will
have
concerns
on
the
than
api
so,
but
I
also
wondering
what
kind
of
the
special
metrics
you
are
looking
for:
network
io.
M
I
yeah
we
need
a
cpu
memory
file
system
and
network
three
of
those
aren't
too
hard
to
get,
but
network
particularly
has
been
tricky
to
get
through
other
sources.
So.
J
I'll
say
that
that
is,
I
think,
or
at
least
that
was
my
goal
when
I
was
working
on
that,
I'm
not
actively
working
on
that
right
now.
Perhaps
someone
else
will
pick
that
work
up,
but
this
cap,
in
particular,
is
not
aimed
at
removing
content.
It's
just
aimed
at
reducing
the
number
of
endpoints
in
a
number
of
different
formats.
We
support.
M
Okay,
I
would
point
out
that
I
think
that
clarity
that
that
these
metrics
are
available
elsewhere
isn't
clear
to
everyone
on
that
issue.
So
maybe
a
little
more
sort
of
clear
discussion
would
be
helpful
to
the
community.
Okay,.
A
L
Yeah,
I
think
yeah
I
just
wanted
to
call
out.
This
is
related
to
the
first
thing
we
started
with.
So
you
know
nice
wrap
there,
but
yeah
all
npd
development
is
blocked
because
of
a
failing,
end-to-end
test
for
like
three
weeks
now
or
two
weeks
now,
so
we
have
fixes,
but
unfortunately,
the
approvers
of
npt
are
not
approvers
of
the
end-to-end
tests.
L
So
there's
a
little
bit
of
latency
there,
but
you
know
if
there's
a
process
for
getting
approval
other
than
just
designing
people
and
hitting
them
up
on
slack.
You
know
we
can
go
through
that,
but
yeah.
Basically
after
these
two
pr's,
the
test
will
be
green
and
we
can
resume
development
on
on
the
nvidia
repo.
L
I
I
yeah,
I
know
it
makes
sense
to
me.
I
I
just
started
looking
at
npd
last
week,
and
you
know
these
this.
This
particular
test
doesn't
is
a
pre-submit
on
the
npd
repo
and
not
the
main
kubernetes
repos.
It
doesn't
quite
make
sense,
but
maybe
don
you
have
some
background
for
that.
A
A
So
and
then
we
also
have
the
node
test,
especially
for
for
for
sale
right
there
mpdy
is,
and
but
but
this
one,
no,
the
problem
detector.
Actually
it
is
not
a
level
test,
it's
just
by
happening.
If
you
look
at
that's
the
e2e
node,
and
so
basically
it
is
inside
of
the
not
the
interest
of
the
test-
and
I
guess
just
over
time,
engine
people
just
kind
of
like
that,
because
today's
gk
not
gk
today's
kubernetes,
a
lot
of
tests
generated
by
node
team,
actually
signal
team
or
k,
not
so
they're.
A
A
That's
all
so
so
the
I.
I
don't
have
the
detail
why
it's
become
a
class
level
of
end-to-end
test,
but
I
think
about
the
people,
if
want
to
use
an
mpd
in
the
production
they
do
expected.
The
community,
just
like
the
expected
community,
have
some
of
the
runtime
right
container
runtime
test.
So
then
they
can
at
least
share
the
tester
without
so
they
can
using
that
out
of
box.
That's
all.
C
A
C
I
think
the
history
is
just
convenience
that
it
was
done
here
and
seems
like
it's
biting
us,
but
I'm
happy
to
approve
this.
We
can
approve
it.
It
just
seems
like
the
test
better
fit
with
the
add-on
right
like
what
we
want.
Node
feature
discovery
tests
to
get
integrated
into
this
suite,
probably
not
right.
So
if
we
can
fix
the
mpd
pre-submits
to
not
depend
on
this
testing
cube
cube,
that
seems
like
a
good
piece
of
debt.
We
could
clean
up.
J
Yeah
just
added
something
quickly.
I
just
wanted
to
do
quick,
shout
out
since
we
didn't
have
so
you
know.
Last
week
I
sent
out
a
pr
for
the
graceful
shutdown
cap.
The
implementation.
A
There,
sergey
and
ronald
took.
J
A
quick
look
at
it,
so
I'd
appreciate
anyone
else
review
comments
to
take
a
quick
look,
I'm
trying
to
get
it
in
for
120,
but
still
has
some
api
review
and
a
couple
things.
So
if
anyone.
D
A
C
F
Sasha
may
not
be
on
the
call
it's
late
for
them.
C
B
C
F
B
F
Actually,
it
doesn't
have
to
be
that
way.
Right,
I
mean
so
container
d
can
continue,
supporting
both
v1
alpha
1
and
v1
at
the
same
time,
depending
on
how
they
want
to
support
different
kubernetes
versions
with
the
same
version
of
container
d
or
not
on
cryo
side
will
mostly
just
end
up
switching
over
to
v1.
After
a
brief
transition
period,.
C
Yeah,
basically,
the
cubelet
would
speak
one
client
version,
but
the
runtimes
could
be
bilingual,
but
I
I
think
trying
to
make
the
cubelet
bilingual
is
tough
and
the
package
version
here
it
just
seems
semantics
at
best
and
felt
like
we
could
save
six
months
of
of
of
overhead.
If
we
were
good
with
that.
F
A
So
so
so
that's
why
early?
I
I
personally
think
about
by
design
continuity.
It
is
design
activities
I
think
I
mentioned,
and
but
I
cannot
go
out
because
mac
brown
is
not
here.
Mac
cosby
also
is
not
here.
I
just
mentioned
that
over
time,
so
the
continuity
implementation,
maybe
is
not
as
flexible
as
what
we
design
treat
those
api
as
the
plugin.
A
So
because,
initially
we
designed
and
continually
have
their
internal
api
and
the
cis
external
api
to
talk
about
the
kubernetes,
then
they
could
plug
in
some
different
version
or
plug
in
different
of
the
api.
So
that's
kind
of
the
original
design,
just
like
the.
We
also
have
the
container
share
design
right
so
like
the
plugin
too.
So
that's
why
I
support
different
use
cases,
but
over
time,
since
I,
after
the
original
first
year
design,
I
connect
detach.
So
I
don't
know
kind
of
the
product
and
development
it
is
so
I'm
glad
the
circuit
co-op.
A
Otherwise
I
will
poke
you
circuit,
so
circuit
actually
is
looking
into
those
ci
and
continuity
integration
right
now.
So
so
we
need
to
make
decisions
as
soon
as
possible,
but
I
by
design
I,
if
we
could
do
continuity,
can
do
that.
Well,
I
don't
have
any
question.
I
don't
have
any
concern
but
shouldn't
the
neck.
The
shouldn't
make
the
crowd
have
the
same
problem.
F
Yeah
yeah,
both
of
them
should
have
a
similar
problem,
but
cryo
is
not
blocking
ci
right
now.
That's
why
yeah
yeah
yeah
herschel
is
working
on
making
that
happen.
H
E
N
Yeah,
so
it
is
the
same
plug-in
model,
I'm
not
all
that
familiar
with
what
would
be
involved
with
changing
that
out
or
it
should
be
trivial
to
add
a
second
plug-in
for
for
supporting
v1.
I
can't
think
yeah,
okay
yeah.
I
think
it
would
just
have
to
be.
You
have
to
have
one
or
the
other
involved
because
of
because
of
listening
forwards
and
all
that
kind
of
stuff,
but
yeah
it
should
be,
should
be
not
difficult.
A
Yeah
yeah,
I
even
remember
there
are
some
optimization
made
by
michael
cosby.
Even
support
multiple
and
the
resource
overhead
is
minimal
and
performance
actually
also
the
and
also
is
very
minimal.
So
I
remember
that
that's
a
long
time
ago
when
we
had
this
requirement,
so
I
have
the
huge
debate
so
but
okay,
thank
you.
Bran.
C
I
think
the
macro
question
don
is
like
do
we
expect
on
a
going
forward
basis
that
the
cubelet
speaks
one
version
of
cri
or
speaks
many
versions
of
cri
and
I
feel
like
the
keyboard
speaking
one
language
was
like
the
design
goal
of
cri
to
begin
with,
so
this,
if
we
could
skip
to
v1,
then
in
the
package
version
we
can,
you
know
we
can
externally
message
it
as
still
beta
once
that
step
in
121
is
done,
but
I
think
it
saves
a
lot
of
grief.
A
Sure,
okay,
if
you
have
more
content,
can
you
raise
that
one
I
believe,
but
the
bad
design?
That's
definitely,
but
I
understand
they
may
have
some
like
the
actual
work.
A
A
A
A
Thanks
thanks
everyone,
that's
all
for
today
and
see
you
through
the
this
week.
I
will
stay
on
the
slide
more
often
and
see
you
if
you
need
me
and
the
derek
and
talk
to
us
through
the
slack
thanks.