►
From YouTube: KubeVirt Community Meeting 2020-05-20
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
So
we
we
have
a
use
case
where
we're
trying
to
not
just
create
a
VM
and
stop
and
start
it,
but
we're
trying.
We
actually
have
pairs
of
VMs
and
as
far
as
I
can
tell
it
looks
like
it
probably
doesn't
work
that
well
within
the
controller
for
Hubert
and
before
we
went
off
and
do
something
different,
I,
just
kind
of
wanted
to
present
it
as
an
idea
like
whether
there's
any
way
it
would
work
within
convert
or
we
could
expand.
Cuber.
B
A
No,
they
actually
have
to
exist
in
two
separate
nodes,
so
they're
paired
there
are
pairs
of
nodes
really
with
a
pod
in
each
and
and
that
is
what
allows
us
to
do
very
fast,
failover
from
one
to
the
other.
So
that's
why
they're
in
a
paired
configuration
but
they're
there
they're
failing
over
within
a
tenth
of
a
second
because
they're
directly
know
about
each
other.
So.
C
A
Listen
well.
We
actually
do
that.
We
we
take
care
of
the
networking
ourselves.
You
know
within
the
two
VMs,
by
the
way
that
we
configure
the
ARP
going
out
to
the
switches.
So
we
we
have
a
solution
for
that.
It's
more
we're
trying
to
control
them
from
kubernetes
the
kubernetes
api
and
that's
and
that's
where
we're
trying
to
use
Cubert
to
also
like
start
and
stop
them,
but
as
pairs
is
that
clear?
Okay,.
C
A
A
A
It's
exactly
exactly
right.
The
closest
thing
is
a
stateful
set
of
two,
but
it's
a
paired
stateful
set,
which
is
even
slightly
different
than
a
stateful
set.
Yes,
it's
very
close.
B
A
D
D
Now
it
was
not
a
controller
in
that
case,
if
I
don't
if
I
recall
so
it
may
it,
it
does
not
meet
your
needs,
but
me
I
think
the
trouble
there
was
that
if
the
we
had
a
template
virtual
machine,
but
if
it
was
deleted,
there
was
no
association
between
the
children,
the
the
instances
that
were
Ben's
on
so
you
could
get
in
weird
states
where
things
were
hanging
around
that
should
have
been
deleted,
and
that's
just
due
to
the
nature
of
the
fact
that
we
were
just
using
resources
to
coordinate
with
each
other.
So.
C
Yeah
yeah,
so
regarding
diversity,
david
has
a
more
advanced
experiment
going
on
regarding
to
that
where
you
could
use
a
stateful
set
or
deployment
natively
from
kubernetes,
and
it
would
spawn
the
games
for
you
still
I'm,
but
I'm
not
entirely
sure
is
when
I
and
I
think
what
then
it
just
says.
It's
not
exactly
like
a
stateful
set,
for
instance
right.
C
So
it's
when
I
understood,
you're,
correct
and
you
you
would
kind
of
expect
that
Princeton
a
stateful
set
you
have
to
behavior
if
you're
all
out
to
and
the
first
one
has
an
issue
or
whatever
or
you
scale
it
up
to
three
and
there's
an
issue.
It
just
doesn't:
do
it
anymore,
just
waits
until
this
is
resolved
and
that's
probably
not
what
you
want.
You
want
it
to
be
more
available,
not
less.
In
this
case,
yeah.
A
Yes,
yeah
yeah
I
mean
there's
a
lot,
it's
close
to
a
state,
it's
very
close
to
a
stateful
set
of
two,
it's
very
close
to
that,
and
we
could
probably
work
with
that
construct
of
a
stateful
set
of
two
and
we
would
be
willing
to
do
a
controller.
I
mean
we're
gonna
we're
gonna
go
in
that
direction,
probably
is
kind
of
where
we're
headed
and
so
I
just
kind
of
wanted
to
raise
it
and
just
see.
If
we
that
was
the
right
solution.
B
So
there's
two
purchase
that
I
see
with
this
there's
a
layered
controller
that
you
all
build
on
top
of
our
constructs
like
our
API.
It's
that
allowing
two
men
manage
a
single
object
that
underneath
it
man
just
two
virtual
machines.
However,
you
want
do
it
and
then
the
other
approach
is
to
something
we're
experimenting
with
right
now
and
that's
to
be
able
to
reuse.
The
pod
controllers
work
low
controllers
for
virtual
machines,
so
that
path
would
be
using
a
staple
set
directly
and
having
that
merged.
B
The
virtual
machines
for
you,
and
neither
of
those
things
exist
today,
but
they're
both
options
for
you.
If
you
were
interested
and
you
found
that
the
staple
set
of
two
idea
seemed
to
work
for
you,
that
would
kind
of
be
a
pilot
use
case
for
us
actually
implementing
that
logic.
So
part
of
the
reason
we
haven't
moved
forward
with
the
pod
work
load
controllers,
managing
virtual
machines
if
it
was
unclear
exactly
people
have
come
to
us
and
asked
for
things
like
that,
but
they
haven't
given
us
like
a
concrete
example
of
hey.
B
A
I
would
certainly
take
it
as
an
action
to
that's
what
I'm
in
the
process
of
really
doing
is
looking
at
what
we
should
do
at
this
point
and
and
I
have
to
respond
to
management,
pretty
quick
on
a
solution.
So,
yes,
I'm
actively.
We
are
actively
engaged
and
we're
adding
engineers
to
this
project.
So,
yes,
we
will
look
at
pod
work
controller
under
Qbert
as
one
half
and
the
other
one
would
be
really
adding
a
controller
to
guru
Nettie's.
A
B
B
E
Hi
yeah,
can
you
give
me
yeah
I
just
wanted
to
briefly
mention
that
the
117
Lane
is
now
mandatory
to
pass
for
PRS
to
get
merged.
That
Lane
has
a
certain
extent
forced.
So
if
you
have
a
test
that
keeps
failing
on
your
PR
on
that
specific
Lane,
it's
probably
Linux
so
I
would
recommend.
Looking
at
the
the
artifacts
appear
without
the
diet,
the
denials
and
try
to
fix
them.
E
D
E
E
D
F
G
F
G
F
G
F
G
So
I
think
of
maybe
we
can
create
a
filter
named
or
environmentally
owed
and
set
it
by
ours
ourselves
in
the
llamo
failure.
G
G
G
H
Yeah,
so
I
was
wondering
about
this
PCI
pass
through,
so
what
kind
of
devices
it
will
be
able
to
support?
Is
it
like,
when
we
think
about
the
storage
devices?
Will
it
be
like
when
we
can?
Can
we
use
the
existing
storage
devices
directly
into
the
beam
and
the
network
devices?
Is
it
like
that
I
think.
H
That's
why
I'm
pitching
in
here
so
the
storage
devices?
We
have
a
lot
of
block
devices,
so
we
would
like
to
have
a
beam
directly
accessing
our
recognizing
the
storage
disks
and
make
use
of
writing
the
file
systems
or
make
make
a
best
kind
of
thing.
So
is
that
directly
possible,
or
do
we
need
to
do
anything
in
that
case?.
F
H
We
just
imagine
this
VM
is
a
storage
solution
like,
and
we
have
number
of
disks
in
the
under
bare
metal
and
VM
has
to
recognize
day
of
SDA,
SDB,
SDC's
kind
of
thing
and
create
file
systems
on
that
mkfs
XFS
awesome
like
that,
so
that
that
is
one
of
the
use
cases
for
us
and
how
to
achieve
such
things
is.
Is
it
possible
directly
so.
I
I
Alternatively,
you
can
tag
it
and
then
on
the
API.
You
would
say
that
this
specific
volume
it
has
a
tag
and
we
would
provide
the
PCI
address.
Id
yeah,
the
PC
address
of
the
of
the
disk
and
the
tag
as
a
as
a
metadata,
and
then
you
could
read
this
metadata
and
recognize
recognize
the
storage.
According
to
this
PC,
the
coupling
of
the
piece
address
to
the
tag.
F
You
have
from
you
are
using
in
communities
as
well
right,
because
then
you
can
explore
the
like
that
local
device,
def,
sta
and
I
think
what
bloody
is
then
saying
is
that
in
DBM
definition
you
then
consume
that
local
PD
and
there
you
assign
a
serial
number
to
it
right
when
you
create
the
disk
or
you
assign
a
serial
number
and
then
nothing
in
blood.
Please
help
me
how's
that
new
mechanism,
you
introduced
all
with
an
export.
He
volume
information
skewed
against.
I
Yes
and
then
we
have
were
providing
metadata
through
third,
a
conflicting
score
means
like
that:
I
have
I,
can
post
a
link
to
the
documentation
and
how
exactly
does
it
work,
but
on
the
VM
definition,
we
can
provide
a
tag
for
specific
for
specific
discs
that
were
that
were
passing
to
the
VM
and
then
there's
a
known
location
on
the
VM
that
this
metadata
would
be
created.
You
could
read
this
metadata,
it's
a
device
metadata
and
it
will
provide
you
with
a
structure
of
a
tag
and
an
address
on
the
on
the
on
the
VM.