►
From YouTube: KubeVirt community meeting 2020-12-02
Description
Recording of the KubeVirt Community meeting held on 2020-12-02 at 16:00 CET
A
Okay,
welcome
everyone.
This
is
the
keep
birth
community
meeting
wednesday
2nd
of
december
2020,
so
we
have
a
short
agenda
so
far
feel
free
to
add
any
additional
items,
but
so
far
we
have
one
item
from
brian
about
threat
and
discussion
on,
save
and
restore
ryan.
All
yours,
hey
thanks.
Pat.
B
All
right,
I'm
gonna,
share
my
screen
here.
C
B
All
right
the
link
for
the
the
docs
in
the
the
community
meetings,
so
we
talked
a
little
bit
about
this.
Last
week
we
had
a
number
of
discussion
items,
so
I
just
got
three
three
topics
in
here,
but
we
don't
need
to
adhere
to
those
specifically.
Those
are
just
three
that
I
know
that
some
changes
that
I
made,
but
there's
been
some
discussion
on
the
mailing
list
as
well,
and
we
can
cover
that
too.
B
All
right
so
I'll
go
to
the
doc
here
and
just
kind
of
go
over
one
of
the
first
items
to
kind
of
like
introduce
this
again
like
talk
a
little
bit
about
like
use
cases,
goals
and
things
like
that,
and
so
the
first
thing
is
one
of
the
things
like
that
I
want
to
accomplish.
With
this
proposal
I
kind
of
titled
it
I've
kind
of
tried
to
sum
it
up
here
like
offline
virtual
machine
migration.
B
So
essentially
the
idea
that
I
can
take
a
virtual
machine.
I
can
sort
of
like
put
it
aside.
I
can
save
it.
I
can
save
its
ram.
I
can
save
everything
associated
with
that
virtual
machine.
Take
its
execution,
take
it
down
and
then
at
some
point
later
on,
I
can
then
bring
it
back
up
and
start
where
exactly
where
I
was
and
continue
off
with
whatever
I
was
doing,
I
think
offline
virtual
machine
migration
is
kind
of
the
term
that
I've
seen
use.
So
I
figured
I'd
use
it
here.
B
It
seems
appropriate
essentially
taking
that
machine
down
and
moving
it
somewhere
else
and
using
it.
So
now
the
use
cases
kind
of
generally
go
along
with
this.
The
idea
that
an
administrator
can
take
a
misbehaving
virtual
machine
and
save
it
and
investigate
it
at
a
later
date.
Maybe
in
an
environment,
that's
more
secure,
doesn't
have
internet
access
and
do
some
sort
of
debugging
there
and
making
sure
or
figure
out
what's
wrong.
B
Another
example
in
like
kind
of
an
appliance
type
situation
where
a
customer
who
is
using
an
appliance
from
a
vendor,
they
don't
quite
understand
entirely
what
is
in
this
appliance,
it's
maybe
broke
it's
broken
perhaps,
and
they
need
to.
B
They
don't
know
how
to
fix
it,
because
they
don't
understand
entirely
what
the
appliance
is
doing,
so
they
can
save
it,
send
it
off
the
vendor.
The
vendor
can
look
at
it,
restore
it
fix
it
and
return
it.
So
those
are
kind
of
general
use
cases
of
this
and
what
some
of
what
I'm
looking
for.
B
I
have
a
few
definitions
terms
here,
since
I
think
there's
a
lot
of
there's
a
ton
of
terms
associated
with
this
and
from
what
I
could
tell
there's
a
lot
of
these
terms
in
different
definitions
based
on
platforms
and
hypervisors
and
all
sorts
of
things.
So
I
kind
of
define
that
here
just
so
that
we
have
a
consistent
understanding
and
kind
of
what
I'm
talking
about
was
when
I
say
use
some
of
these
terms.
Pause,
suspend
keyword,
has
pause
so
essentially
temporarily
putting
machine
on
hold
no
cpu
network
or
disk
activity.
B
D
Here
I
have
a
question
already
yes,
so
it
looked
to
me
that
you
actually
would
want
to
pause
the
vm
and
then
save
it
right
from
the
use
cases
above.
But
here
it
looks
like
you
see,
safe
as
if
you
first
hibernated
normally
and
then
saves
your
m5.
B
B
Yeah
so
I'd
say
I
even
kind
of
yeah,
so
I
guess
you
could
say
like
for
both.
That
would
be
the
case.
That's
fine!
I
actually
think
this
does
that,
like
I,
I
was
looking
at
this
like
when
I've
done
some
testing
in
this.
I'm
I'm
pretty
sure
that
this
command
delivered
actually
does
a
pause
and
then
does
a
save
just
whether
or
not
what
happens
afterwards
and
that's
kind
of.
B
D
D
So
you
just
meant
that
when
you
go
down
to
the
acronyms
again
yeah
a
little
bit
you're
saying
is
non-gold
create
an
api
for
vitamin
safe
flex.
This
is
just
you
mean
you
don't
want
to
say.
Is
you
want
it?
Don't
want
to
implement,
live
migration,
a
live
snapshot,
I
guess
just
to
save
operations.
Degree,
I'm
just
a
little
bit.
Maybe
I'm
asking
the
questions
too
early
and
you
want
to
go
more
into,
but
maybe
I'll
come
back
to
that
later.
Just
go
on
and
explain
your
stuff.
Sorry
for
drafting
too
early.
Okay.
B
That's
okay!
I
I
do
want
to
talk
about
that.
B
So
the
the
yeah,
because
so
the
next
thing
non
goals
and
the
reason
why
I
had
to
find
this
here,
because
I
I
keep
hearing
a
lot
like
live
virtual
machine
snapchat
and
so
I've
kind
of
explicitly
been
trying
to
use
the
word
offline
sort
of
like
an
offline
snapshot
and
and
what
I'm
trying
to
say
is
that
the
result
of
the
save
up
operation
is
that
the
virtual
machine
is
no
longer
running
and
so
then
the
inverse
meaning
that,
if
I'm
doing
a
live
sort
of
a
live
snapshot,
would
mean
that
the
virtual
machine
stays
running
after
I
do
the
save
okay,
I
don't
know
where
I'm
at
yeah.
E
It's
the
same.
Mechanics,
though
we're
just
saying
that
for
a
live
snapshot,
we
go
through
the
same
stuff,
we're
gonna,
pause
or
suspend,
and
then
just
save
the
memory
and
then
unpause
it.
But.
D
I
guess
it
makes
sense
from
the
well,
I
it
depends
the
live
snapshot.
I
guess
this
only
even
goes
as
far
as
you
know.
If
the
guest
agent
is
installed,
you
can
really
do
live
snapshot
with
pretty
much
no
visible
down
time
right,
so
live
snapshot.
Snapshot
is
much
more
complicated,
so
I
guess
in
general
it
makes
sense
to
say
that
this
does
not
want
to
implement
it.
I
guess.
B
Yeah
I
like
I
get
like
the
the
way.
The
way
I'm
approaching
is
that,
like
a
way
I
look
at,
is
that
the
sort
of
the
end
result
like
is
the
virtual
machine
running
yes
or
no,
and
I
don't
want
to
end
up
like
what
I
don't
want
to
implement
here
is
where,
after
I
do
this
save
or
whatever
snapshot,
we
end
up
with
a
live
virtual
machine,
a
running
virtual
machine.
That's
not
quite
what
I'm
after.
I
think.
B
E
Think
from
our
standpoint,
if
we're
looking
at
actually
getting
this
thing
implemented,
we
actually
have
to
understand
both
because
it's
something
where
we're
going
to
approach
live
snapshots
at
some
point,
and
this
is
unfortunate-
it's
overlapping.
I
understand
that
you
don't
really
want
to
to
implement
necessarily
the
use
case
where
it's
running
afterwards,
but
we
at
least
have
to
have
understanding
of
how
that's
done
and
how
this
interact
like
how
these
things
are
related,
at
least.
Does
that
make
sense
like.
B
Yeah,
okay,
I
I
understand.
Okay,
I
understand
you
guys
are
saying
okay,
so
I
the
reason.
So
let
me
tell
you
my
motivation.
I
reason
I
wanted
to
focus
just
on.
One
was
because
I
felt
like
we're.
I
felt
like
the
discussion
was
getting
lost
in
terms
of
like
what
we
were
talking
about.
So
that's
why
I
wanted
to
focus
on
one.
I
understand
how
to
relate
it,
so
I'm
fine.
If
we
want
to
do
talk
about
both
together
and
yeah,
I
mean
that's
fine
with
me
and
even
as
part
of
the
implementation.
E
B
Okay,
that's
fine!
Okay!
Then
let
me
I'll
just
revisit
this
then,
because
this
part
of
me
specifically
because
yeah
I
I
got
what
you're
saying.
D
D
B
Okay,
cool,
okay,
all
right,
so
then
I
think
that
so
that's
like
the
first
topic.
I
guess
is
that
kind
of
use
cases
that
and
those
those
three
things
so
I
mean
is
that
do
we
have
any
more
discussion
on
that
because
I
know
like
we
had
some
comments
last
time
about
some
of
these
cases
and
is
that
clear
kind
of
does
everyone
understand
what
I'm
trying
to
accomplish?
E
I
guess
the
use
I
hate
to
bring
this
up
going.
We've
been
trying
to
cover
it
on
the
mailing
list
a
little
bit.
It's.
F
E
What
we're
talking
about
we're
talking
about
a
virtual
machine
instance
or
a
virtual
machine?
It
says
virtual
machine
instance
in
this
use
cases,
but
what's
the
admin,
I
don't
completely
understand
the
end-to-end
use
case.
If
we're
looking
at
virtual
machine
instances
river
instead
of
virtual
machines,.
B
Okay,
so
this
could
be
my
understanding,
so
the
way
the
wave
looked
at
is
that
I
have
looked
at
the
virtual
machine
instance,
and
this
is
what
I've
considered
my
running
vm,
that's
like
how
I'm
using
it
in
the
language
here
and
virtual
machine
being
not
running.
That's
like
how
I'm
looking
at
it,
and
so
I'm
not
looking
at
it.
Maybe
the
way
you
are
with
like
the
objects,
so
I
that's
kind
of
my
perspective
on
this
because
running
virtual
machine
being
that
how
that's
how
you
get
the
ram.
E
I
guess
I'm
thinking
about
okay,
so
somebody's
using
a
virtual
machine
object,
so
the
higher
level
object
where
you
can
start
and
stop
it
and
all
that
stuff.
I'm
seeing
this
idea
of
a
of
a
suspended
or
whatever
virtual
machine
is
a
as
another
state,
perhaps
to
that
higher
level
object.
E
So
you
have
a
suspended
virtual
machine.
That's
saved
off
completely
the
ram
and
everything
is
the
state
where
you
can
restore
it,
but
it's
like
suspended
in
memory
and
it's
not
actually
running
anywhere.
It's
not
taking
any
resources
in
the
cluster
other
than
the
storage
resources
like
I
can
represent
that
in
that
same
high
level
object.
The
vm
object
that
when
we
talk
about
virtual
machine
instance,
I'm
trying
to
represent
just
as
a
virtual
machine
instance
with
other
like
save
state
and
things
like
that,
it
becomes
more
disjoint
and
I'm
just.
D
Yeah
I
mean
it's,
it's
the
same
like
at
with
it
like
with
the
virtual
machine
instance.
As
such,
there
is
no
strong
binding
in
any
way
or
just
in
virtual
machines.
It
just
references
some
stuff,
and
you
have
to
keep
these
references
right,
restoring
exactly
it's
another
object
like
like
you
have
with
the
virtual
machine
snapshot
where
also
there
is
no
virtual
machine
anymore
afterwards.
B
So
what
I
guess,
what
what
do
you
see?
Sorry,
the
relationship
between
the
virtual
machine
snapshot,
api
and
the
virtual
machine
api,
because
it's
on
the
way
I
view
it
as
sort
of
like
very
similar
and
and
like
the
virtual
machine
api
it
has.
There
are
sort
of
things
associated
that
you
can
do
with
it,
but
it's
it's
generally,
a
representation
of
a
virtual
machine
that
you
can
run.
D
What
we
should
probably
explore
in
that
direction
when
we
also
said
that
we
want
to
have
a
high
level
plan,
also
for
for
a
live
snapshot
later
on,
it
would
can
make
a
sense
to
explore
if,
for
instance,
let's
suppose,
you
have
a
vm
and
we
want
to
do
a
live
snapshot,
if
you
would
also
have
to
save
the
vmi
definition
to
a
degree
to
keep
some
information
for
it,
and
I
don't
know,
explore
this
relationship
a
little
bit.
H
D
We
have
more
on
it
like
it's.
A
vmi
is
also
kind
of
the
equivalent
of
a
part
like
you
can
have
different
controllers.
One
of
them
is
a
vm
which
adds
some
features
and
the
other
one
which
you
have
is
very
nice
virtual
machine
instance
replica
set
which
can
create
multiple
running
games
for
you
out
of
a
single
entry
point
and
manage
it
for
you
and
scale
it
for
you
and
yeah.
That's
also
there.
So
it's
not
just
that.
There's
only
the
virtual
machine.
B
Understand
so
this
is
so
you
have
an
api
that
can
sort
of
isn't
necessarily
a
representation
of
a
running
machine.
A
running
virtual
machine
with
a
virtual
machine
replica
set,
is
kind
of
what
I'm
understanding
and
then
you
have
a
virtual
machine
api
which
is
kind
of
similarly
holds
some
of
that
ground
as
well,
and
just
the
virtual
machine
instance
itself
is
a
single
representation,
and
so
when
we
shut
it
down,
it's
technically
no
longer
a
bmi
yeah.
I
think
I
understand
that.
I
guess
the
thing
that
I'm
missing,
though,
is
like.
B
So
I
guess.
Maybe
this
is
a
discussion.
The
direction
here,
like
the
one
that
I
describe
in
in
this
document,
is
I'm
talking
about
like
it's
really
it's
an
api
that
I
know
michael
worked
on
for
the
for
the
snapshot
I
sort
of
viewed
as
sort
of
a
way
that
holds
all
the
information
of
a
formerly
running
bmi.
But
technically
I
could
see
how
a
virtual
machine
is
also
can
be
considered
a
formally
run,
a
snapshot
of
a
formally
run
api
or
sorry,
a
formerly
run
virtual
machine.
B
E
The
virtual
machine
snapshot
is
a
point
in
time
of
that
vm
object
and
it
can
be
restored
to
that
previous
point
in
time.
We're
talking
about
restoring
the
vm
on
just
the
higher
level
object.
It
only
works
in
the
higher
level
object
right
now.
So
let's
say
it's
like
think
of
it
like
a
restore
point
for
like
your
laptop
or
something
where
you
can
restore
it
to
the
previous
state.
E
A
D
B
Do
do
you
mind
if
we
cover
that
in
like
a
few
minutes?
Sorry,
I
don't
know
who's
talking.
I
do
you
mind
if
we
cover
that
in
a
few
minutes-
and
we
just
just
finished
like
the
kind
of
the
api
discussion
here.
First,
let
me
come
back
to
that.
Is
that
okay,
because
I
think
that's
a
I
do
want
to
answer
that
too.
B
D
When
so,
I
think
we
can
explore
this
direction
of
a
kind
of
forensics
debug
flow,
where
you
would
then
just
transform
any
vmin
to
vm
as
part
of
a
snapshot.
I
wouldn't
see
anything
speaking
against
it.
Okay,
I
guess
the
yeah
david
also
brought
up
this
vm
replica
set
instead
of
a
vmi
replica
set.
E
That's
strange:
when
we
look
at
the
virtual
machine
replica
virtual
machine
instance
replica
set.
The
thing
that's
strange
is
how
do
we
represent
individual
state
of
vmis?
So
so
right
now,
it's
all
ephemeral.
If
somebody
has
a
pvc
that
they
have
in
their
virtual
machine
instance,
replica
set
it's
read-only
and
the
femoral
state
is
stored
on
disk
with
the
vmi
in
the
in
that
container.
E
So
when
we
talk
about
taking
a
live
snapshot
or
a
say,
restore
point
or
something
like
that,
where
there's
nothing
to
we'd
have
to
somehow
get
that
local
storage
onto
something
persistent
like
it
doesn't.
D
I
mean
that's
what
you
have
to
do
anyway.
I
mean:
what
do
you
have?
What
do
you
have
with
this
shared
storage?
Is
that
you,
anyway
right
into
overlays
right
in
your
vms,
so
you
have
a
read,
only
base
image,
which
is
in
the
pcs,
so
you
can
just
snapshot
it
at
any
time
with
including
the
overlay
where's.
D
E
Okay,
yeah
okay,
see
so
you
would
create
a
vm.
For
that
I
mean
pvc
for
the
overlay.
D
E
D
D
D
E
Not
a
per
vmi,
oh
there's!
No!
When
we
talk
about
the
virtual
machine
instance,
replica
set
sorry
we're
getting
a
tangent
here.
The
point
I'm
trying
to
make
is
ryan.
If
you
end
up
moving
to
something
like
that
virtual
machine
instance
replica
set,
then
we're
going
to
have
issues
trying
to
even
save
and
restore
a
vmi,
because
a
portion
of
that
vmi's
data
at
the
disk
is
on
an
overlay.
E
Have
options
with
vms,
so
you
could
create
a
per
vm
like
that.
The
data
prime
templates,
whatever
could
create
a
per
vm
persistent
volume
claim.
So
then
you'd
actually
have
you.
B
Is
this
like?
Would
you
say
I
mean?
B
Is
this
something
that,
like
with
the
virtual
machine
instance
replica
set,
is
it
something
that
sort
of
I
mean?
Is
there
a
way
to
sort
of
that
we
can
make
it
like
incorporate
this
as
part
of
like,
like
I
don't
like?
How
would
I
how
would
I
get
like?
Is
it
was
kind
of
what
you're
telling
me
that,
like
there
isn't
really
a
way
we
can
make
this
work,
or
is
it
that
we
started
to
change
some,
do
some
major
changes
to
make
it
work.
D
I
guess
it's
hard
to
tell
what
you
actually
try.
It
really
depends
on
what
you
try
to
achieve,
so
the
virtual
machine
instance
replicas.
D
That
has
some
advantages
when
you
think
about
fmr
vms
regarding
to
overhead
and
complexity,
because
you
can
use
you
can
use,
I
don't
know
a
container
disk
and
just
write
in
overlays
or
whatever
easily,
but
but
on
the
other
hand,
not
everything
may
be
there
yet
like
getting
a
pvc
pair
vmi
from
the
replica
set
right
you
you
can
only
upfront
define
what
you
want
accessible
for
vmis,
basically
except
for
local
storage,
stuff,
okay,
but
it's
how
it
is
today.
D
B
Okay,
I
guess
so
let
me
before
I
don't
want
to
take
too
much
time.
I
guess
I'm
going
through
this,
I
mean,
would
would
one
of
you
guys
be
able
to
like,
maybe
summarize
this
in
the
doc?
I
don't
know.
Can
you
do
me
a
favor?
Maybe
I
don't
know
like
is.
Is
it
on
the
thread
like
kind
of
because
I
do
want
to
capture
this?
I
guess
maybe
it's
part
of
this
or
something
like
a
caveat
and
then,
but
I
think
like
and
then
kind
of
move
on
with.
E
That's
fine.
I
think
that
I
think
we
can
get
together
offline
or,
let's
just
have
a
meeting
to
discuss
this
with
whoever.
E
B
I'll
make
a
note,
okay,
all
right,
so
I
think
to
kind
of
conclude
here
what
so
I
I
do.
I
do
like
the
idea
of
okay,
we
go
from
vmi
and
we
follow
the
chain
of
going
to
vm
and
then
vm
snapshot.
I
don't
know,
is
there,
so
does
anyone
object
to
that?
Like?
I
don't
think.
That's
like
the
wrong
path.
I
think
that
seems
seems
reasonable
to
me
because.
E
B
So
maybe
some
details
I
need
to
explore.
There
then.
E
Yeah,
look
let's
talk
about
when
we,
when
we
have
the
focus
discussion
about
perhaps
why
bmis
are
only
in
use
for
your
use
case
and
maybe
understand
kind
of
I'd
like
to
understand
better.
Why
you've
approached
this?
The
way
you
have.
B
E
B
Okay,
all
right,
I
think
we
can
yeah.
We
can
talk
about
that
further
than
separately.
Okay.
Now,
who
I
don't
know
who
was
talking
earlier,
but
look
let's
dive
into
the
use
case?
B
Yeah
do
you
want
to
say
we
used
that
earlier.
F
Yeah,
I'm
just
trying
to
understand
what
solver,
what
problem
are
we
solving?
I
mean,
even
if
we'll
create
the
vm
or
anything
like
this
it,
it
will
still
stay
on
this
system
on
the
system
that
we
actually
took.
The
snapshot
and
what
you're
describing
in
the
use
case
is
that
you
will
want
to
export
this
and
then
someone
on
another
system
with
another
cluster
when
another
set
of
nodes
will
have
to
restore
this,
and
I'm
not
entirely
sure.
How
realistic
is
this
and.
B
So
the
the
general
sort
of
principle,
I'm
kind
of
looking
to
apply
is
that
is
sort
of
that.
I
could
have
that
once
I
take
this
virtual
machine,
it's
it's
now
mobile.
Now
I
guess
you
could
say
like
how
could
I
take
this
virtual
machine?
That's
sort
of
snapshot
in
a
ppc,
that's
in
a
crd
in
a
kubernetes
cluster
and
then
offload
it
put
it
on
a
usb,
stick
and
move
it
to
like.
You
know
I
don't
know
another
another
state
and
then
plug
it
into
my
home
lab
or
something
yeah.
B
I
understand
that,
like.
I
guess,
that's
not
necessarily
like
what
I'm
after
it's
more
like
the
idea
that
that
this,
the
mobility
of
this
of
this
virtual
machine
can
go
beyond
just
where
it
currently
sits
on
the
current
host
that
it's
in
it
can
somehow
go.
I
don't
know
slightly
to
a
different
name
space.
B
At
the
very
least
I
mean
possibly,
I
don't
know
like
the
other
side
of
that.
I
think
sort
of
is
not
really
like
what
like
I
want
to
cover
like,
because
I
think
you
know
maybe
there
can
always
some
system
and
stuff.
I
can
always
work
out,
but
mostly
the
idea
that
I
can
take
it
offline
and
I
can
move
it
like.
It
is
snapchatted
in
this
place
that
I
can
move
it
around
in
an
object.
F
I'm
just
saying
that
when
we
are
when
we
are
saving-
and
we
are
when
we
are
saving
memory,
we
need
to
know
what
kind
of
cpus
were
used.
You
know
the
specifics
of
a
hardware.
So
basically,
when
you
restore
this,
it
has
to
be
restored
in
the
same
kind
of
a
node
and
then
yeah.
What
I'm
trying
to
get
it
is
that
maybe
what
we?
What
you're
after,
is
really
some
kind
of
a
import,
export
or
export
import,
and
maybe
that's
what
we
need
to
really
implement.
B
Well,
I
I
I
mean
to
me,
maybe
maybe
it
might
to
me
it
just
seems-
might
be
like
if
we,
if
we
go
down
the
road,
if,
like
maybe
someone
correct
me
if
I'm
wrong
like
like,
if
you
take
like
a
the
virtual
machine
and
you
have
it
stored
in
an
object
and
then
you're
literally
offloading
it
out
of
the
cluster,
like
maybe
that
sounds
like
kind
of
what
you're
implying
like
export.
B
H
H
F
So
rom,
maybe
we
can
save
it
as
a
dump
and
then
no
I
mean
the
the
the
use
case
here
is
that
something
happened
with
the
virtual
machine
right
and
we
want
to
debug
it.
So
there
are
two
situations
here:
we
want
to
the
same
image,
the
same
kind
of
setting.
F
We
want
to
restore
somewhere
and
then,
on
the
other
hand,
we
have
a
ram
so
round
we
can
save
as
a
dump
and
provide
it
together
with
the
import,
expert
and
dump
can
be
analyzed
separately,
but
then
for
for
the
administrator
or
for
somebody
who
wants
to
debug
a
specific
card
on
on
a
specific
image
that
he
will
be
able
to
restore
in
his
lab.
F
He
needs
the
way
to
import
an
image
and
then
restore
and
use
it
with
with
the
with
a
similar
card.
H
H
B
I
don't
yeah,
I
mean
I'm
not
sure
I
have
the
answer.
I
I
yeah
for
yeah.
I
don't
I'm
not
sure
how
I've
answered
that
I
I
think
like
yeah
go
ahead.
B
Oh,
so,
oh
sorry,
I
I'm
maybe
missing
okay,
so
I
yeah
so
there
there
is
yes
like
this
is
sort
of
this
is
the
case
in
an
existing
system.
Yeah,
it's
it's
sort
of
not
like.
I
I
don't
know.
I
guess
my
perspective
on
this.
Is
that,
like
I
could
think
of
ways
to
get
like
the
files
off
like
for
me.
What
I'm
after
is
like
getting
this
state
stored.
H
But
you
are,
you
are
using,
so
this
is
if,
if,
for
example,
there
is
a
existing
system
or
existing
solution,
rev
openstack,
I
don't
know
anything
else-
busy
box
something.
Oh.
H
The
customer
service
of
that
appliance
once
wants
me
to
d
wants
to
debug
it.
So
I
can
just
take
a
snapshot
or
something
similar
and
send
it
to
him
and
he
can
evaluate
it.
Obviously
it
cannot
work
on
everything.
For
example,
if
you
have
hardware
dependencies
or
like
your
cpu
is
not
emulated
or
something
uses,
some
kind
of
is
based
on
your
hardware.
That
will
be
impossible
to
restore
what,
but
maybe
we
can
add
these
limitations.
H
B
Okay,
I
made
a
note
for
myself.
I
I'll
take
a
look
at
openstack
and
maybe
they
can
get
a
few
links
to
like
export
or
something
see
what
they
have.
Okay,
thanks,
vladik,
okay!
Is
there
any
any
other
questions
about
so
we're
still
at
use
cases
and
and
the
section
do
we
any
more
comments
about
that.
D
I
think
it's,
the
initial
idea
is
pretty
clear.
Now:
okay,
we
can.
We
can
probably
refine
more
on
what
to
do
with
it
than
like.
Edward
and
larry
hinted
a
little
bit
like
once.
You
have
snapshotted
it.
What
would
be
the
kind
of
variations
you
would
want
to
do
with
it
like,
like
mentioned
cordon
brand,
you
may
want
to
add
starting
it
in
the
same
name,
space
or
another
namespace.
D
Another
namespace
is
tricky,
but
anyway
to
live
debug
it
maybe
with
similar
devices-
or
I
don't
know
stuff
like
this-
that
we
have
you
know,
because
it
kind
of
fits
an
impact.
I
guess
some
implementation.
Okay,.
B
Yeah,
those
are
things
yeah
the
concern
we
can
add
on
over
time.
Okay,
all
right.
What
was
the
second
thing
now?
I
forgot
device
management
okay,
so
I
added
another
section
in
here
about
oh
actually,
there's
so
two
things
in
here
so
last
time,
there's
a
question
about
whether
based
on
the
flags
that
we
could
provide
what
will
happen
to
the
virtual
machine.
B
So
I
just
added
a
section
with
the
two
functions
that
I
found
at
lever
that
do
the
save
and
here
are
some
examples
of
what
they
do
so
I'll
leave
those
there
whenever
you
want
to
read
them,
but
the
main
thing
I'm
looking
for
is
this
and
then
at
live
doing,
live
at
a
later
point.
We
can
do
this
and
then
there's
obviously
a
bunch
of
other
things
that
we
could
do
so.
That's
cool
all
right
devices.
B
So
one
of
the
things
that
I
have
had
brought
up
or
mentioned
a
few
times
here
is
devices.
And
what
happens
you
know
when
you
save
restore
and
you
have
devices
attached,
so
one
of
the
things
is
and
dan
baron
j
was
the
one
who
commented
on
this.
It's
somewhere
in
this,
I
think
it's
below
and
sort
of
the
expectations
of
what
happens
when
you
do
a
save.
B
So
the
expectation
is
that
you'll
have
to
detach
the
the
devices
before
you
can
do
the
save
and
it's
basically
the
same
as
in
live
migration.
B
That's
what
I
was
reading
and
essentially
what
we
need
to
do
is
do
some
sort
of
hot
unplug
of
the
devices,
and
I
can
tell
kubernetes-
does
not
support
this,
but
I
think
that's
okay.
This
is
essentially
hot
unplugged
with
kubernetes,
since
kubernetes
makes
those
devices
available
to
the
pods
hot
unplugging
would
be
removing
it
from
the
pods.
So
one
way
around
this
would
actually,
since
we
have
the
vert
launcher
there,
maybe
we
can
have
the
launcher
sort
of
call
it
delivered
and
say:
hey
just
unplug
these
devices.
B
I
call
this
the
software
and
plug-
I
don't
even
call
whatever,
but
I
think
this
makes
sense
we
can
make
it
so
that
save
operation
is
triggered
and
sort
of
the
contract
would
be
that
save
is
triggered
for
launcher
is
going
to
go
unplug
those
devices
with
the
expectation
that
this
pod
is
not
long
for
the
world
and
it's
going
to
be
terminated,
and
this
is
probably
actually
going
to
wear
a
good
place
like
we
can
even
discuss,
live
as
well,
because
this
also
has
a
different
path,
but
not
something
I
have
here
right
now,
but
essentially
that
pod's
not
long
for
the
world
it
gets
removed.
B
The
devices
are
freed
up
on
the
pod
and
the
save
can
execute
without
a
problem.
F
Brian,
just
a
question
yeah:
what
kind
of
benefit
is
it
for
you?
Then?
For
example,
you
were
debugging
some
kind
of
a
system
with
the
with
the
gpu
attached
and
it
was
misbehaving
yep
and
then
and
then
in
order
to
save
this
the
memory
and
restore
it
later,
for
example,
you
will
need
to
to
detach
this
gpu
and
then
so
I
mean.
Is
there
still
a
benefit,
because
I
mean
the
system
doesn't
run.
I
mean
the
the
memory
snapshot
that
you
are
taking.
It
will
be
without
gpu.
B
Yeah
so
like
well
minor
sydney
is
that
yeah
you're
right
it
is
without
gpu.
I
don't
know
if
I
don't
think
it
affects
the
use
case.
I
mean,
I
believe,
the
operations
or
whatever
it
basically
like
the
idea
is
that
the
the
devices
need
to
be
shut
down
so
that
nothing,
I
guess
I
don't
know
liver,
maybe
doesn't
understand
what
the
internal
operations
of
it
are.
So
they
need
to
be
removed.
B
So
I
guess
the
I
don't
know
I
mean
I
don't
think
it
effect
would
affect
the
running
operations
of
the
vm
or
like
of
the
of
the
ram,
is
part
of
the.
I
don't
know
I
mean
that's
a
good
question.
I
think
that's
one
I
can
maybe
take.
H
B
Well,
the
that
side
of
things,
I
think
I
think,
would
be
okay,
I'm
gonna
talk
about
it
down
here,
but
and
the
the
store
side,
but
I
think,
like
we,
we
heard
what
flat
what
was
saying
is
that
what
is
this
virtual
machine
exactly
like?
Is
it?
Is
it
in
that
sort
of
problematic
state
if
we
detach
the
device
right?
Is
that
that
what
you're
saying
like
is
it?
Is
it
actually?
E
That
would
be
my
expectation,
but
I
I
don't
know
I
would
expect
that
it
gets
paused.
We
remove
devices,
I'm
not
entirely
sure
why
that
needs
to
happen
still,
but
that's
okay,
and
then
we
snapshot
and
then,
when
we
restore
you're
kind
of
putting
everything
back
the
way
it
was
and
then
kind
of
hitting
the
button
to
unpause
it
so
yeah.
I
guess
I'm
unclear.
H
F
David,
the
only
thing
that
I
mean
the
the
reason
for
this
is
that
some
of
the
memory
that
the
vm
is
using
is
being
stored
inside
the
device,
so
the
device
state
and
some
of
the
caching
and
some
of
the
things
that
are
going
on
at
that
particular
time.
It's
it's
being
stored
in
the
device.
If
you
don't
have
that
specific
device
right
with
that
cache,
I
mean
it
will
not
yeah.
So.
F
The
benefit
of
is
there
still
a
benefit
in
doing
this,
save
and
restore.
If
we
don't
have
that
that
information,
if
we
don't
have
this
device
saved,
yeah.
E
E
That's
something
really
important
to
understand.
I
would
go
back
and
try
to
gather
what,
with
the
gpus,
if
you're,
even
able
to
save
the
state
of
the
gpu
in
the
way
that
you
need
to
restore
whatever
app
you're
running.
B
H
F
E
B
E
Okay,
so
for
your
use
case,
what
would
would
you
still
have
the
same
benefits
if
you
shut
down
the
vm
entirely
and
did
an
offline
snapshot
like
I'm
trying
to
understand
how
important
is
it
that
you
capture
it
right
at
that
moment
like
the
live
state
everything?
Is
it
possible
to
reboot
it
and
get
back
into
that
state
or
I
I
don't
fully
understand
the
applications
that
you're
running.
B
No,
so
it
it's
specifically
yeah
and
once
you
shut
it
down,
you
sort
of
the
idea
would
be
that
we
lose
where
the
execution
is
like.
Like
a
good
example,
be,
like
I
mean
yeah,
I
mean
I
mean
this,
this
kind
of
sums
it
up.
I
mean
like
what
I
have
here.
It's
it's
really
that
yeah,
you
have
your
virtual
machine
running,
I
mean
you
can
say
like
it's
a
game
or
something
and
and
it's
behaving
wrong
and
we
need
to
diagnose
it.
B
B
D
B
Okay,
I
think
it'll
probably
offer
me
I
think,
yeah
we
don't
say
what
was
the
last
pvcs
yeah
we
can.
We
can
kind
of
talk
about
that
in
the
time.
Then.
Let
me
take
this
this
item
and
kind
of
do
some
more
investigation,
and
then
we
can
talk
about
some
of
the
other
items.
I
think
there's
probably
even
more
topics
we
can
cover
too,
as
well
as
long
as
I
get
the.
Let
me
get
some
more
information
on
this
and
get
back.
A
A
So
thank
you
for
the
discussion.
I
learned
a
few
things
anything
else
on
that
topic.
To
summarize,
there
are
transaction
items
to
take
forward
between
now
and
next
meeting
anything
else.
A
Let's
just
wait
a
few
seconds
but
fabian
you.
We
we
heard
you
and
then
you
appear
to
be
on
mute.
A
A
Okay,
okay,
okay!
So
that's
okay!
Then,
if
no
other
topics
we'll
close
it
here.
So
thank
you.
Everyone
for
attending
and.