►
From YouTube: KubeVirt Community Meeting 2021-07-07
Description
Meeting notes: https://docs.google.com/document/d/1kyhpWlEPzZtQJSjJlAqhPcn3t0Mt_o0amhpuNPGs1Ls/edit#heading=h.xiju5cm9vxe9
A
And
hello,
everyone
welcome
to
another
week
of
cooper
community
meeting,
I'm
your
host
chris
caligari.
A
A
Okay,
we
do
let's
get
right
into
alex's
hole,
117.
B
So
it's
actually
not
my
pull
request
but
looks
like
rick
might
have
issues
connecting
so.
C
A
Yeah,
I
got
it
it's
very
early
in
the
morning
up
here
in
the
pacific
northwest
okay
proposal
for
network
volume.
This
is
a
design
request
into
the
community
repo.
B
Let's
see
if
we
can,
I
guess
I
can
sort
of
go
over
it
and
sort
of
explain
what
the
thing
is:
we're
proposing
to
add
a
new
volume
source,
in
this
case,
a
network
volume
source
and
in
particular,
we're
interested
in
using
this
for
cd-rom
iso
files.
B
So
right
now,
if
you
want
to
add
a
iso
to
your
vm,
you
have
to
get
a
container
disk
and
do
all
kinds
of
weird
stuff
to
to
get
that
to
work,
and
instead
what
we
propose
is
that
if
you
create
an
mbd
service,
for
instance,
that's
that's
our
our
example.
Here
you
created
the
service
and
then
you
use
a
network
volume
to
connect
with
this
service.
The
services
is
doing
read-only
images
and
you
can
have
multiple
vms
use
the
same
service
and
get
the
images
served
in
your
cd-rom.
B
We
have
another
repo
that
actually
has
an
example
of
a
container
that
starts
the
md
surface
and
right
now
we're
actually,
when
we're
building
the
service,
we're
actually
importing
a
bunch
of
cube
vert
registry
images
and
copy
them
into
a
different
container.
We
probably
want
to
do
this
slightly
differently,
but
just
so,
we
had
an
example
service
to
sort
of
show
what
we
were
trying
to
do
here.
B
Yeah,
that
is,
that
is
a
good
point.
It
is
very
similar
to
a
read-only
ppc.
Yes,
the
the
the
this
service
can
actually
serve
multiple
images,
and
you
know
the
next
stage.
After
this,
we
want
to
be
able
to
switch
cd-roms
on
the
fly,
which
is
something
you
can
do
with
a
pvc.
D
Yeah
describe
so
you'd
be
able
to
switch
images
on
the
fly
like
so
you'd,
be
able
to
the
equivalent
of
eject
and
like
insert
a
desk.
Yes,.
B
That's
that's.
The
this
proposal
does
not
include
that
part,
but
that
would
be
the
next
step.
Therefore,.
D
D
Hold
up
you're,
distorted
yeah,
I
immediately
muted
them.
Okay,
sorry,
stu
something's,
going
on
with
your
audio.
E
Apologies
through
the
magic
of
zoom
on
it,
but
the
gain
all
the
way
to
100
so
my
fault
so
gosh.
What
was
I
going
to
ask
okay?
So
the
the
the
the
benefit
of
this
is
that
I
could
say:
take
an
iso
image
and
change
out
what
the
what
the
image
is.
So
you
can
basically
update
what
it
points
to
on
the
fly.
What
happens
if
the
destination
url
is
not
there.
B
B
B
But
the
the
main
benefit
of
this
is
is
that
we
can
change
the
volume
source
on
the
fly
which
we
can't
do
with
a
pvc.
So
that's
why
we
started
out
with
a
network
volume
source,
and
then
you
know,
our
our
main
use
case
is
having
a
cd-rom
where
we
want
to
be
able
to
eject
and
switch
images
by
changing
the
volume
source.
D
So
I've
pulled
up
the
community
document
myself
now,
okay,
so
we
have
two
user
stories
as
a
vm
owner.
I
want
to
boot
multiple
vms
from
iso
image,
and
I
want
to
import
the
iso
image
content
into
my
cluster
only
once
boot,
multiple
vms,
so
that
can
be
solved
with
a
pvc
read
only
pvc
and
the
second
one
is
a
keeper
administrator.
I
want
to
easily
package
iso
images,
make
them
available
to
clusters.
My
end
users
can
access
them
easily
and
again
that
right,
but.
B
Pvcs
are
our
name
space
and
you
know,
since
it's
a
network
service,
it's
not
namespace.
That's
one
thing
that
we
can
give
you.
B
That
is
a
good
question.
We
have
not
solved
the
particular
thing
we
this.
This
is
purely
introducing
the
network
volume
source
and
then
discovering.
This
would
be
an
interesting.
You
know
sort
of
depends
on
what
kind
of
network
source
you
have.
You
know
in
our
existing
similar
problem
with
the
golden.
D
Yeah,
you
have
to
add
some
sort
of
our
back
to
everyone
to
to
view
read-only,
but
that
is
a
discoverable
resource,
though
once
we
add
the
correct
permissions.
B
Well,
I
think
similar
making
the
service
discoverable
is
sort
of
outside
of
the
scope
of
this
this.
This
is
purely
introducing
the
network
volume
source
so
that
I
connect
to
a
network
endpoint
as
a
volume
source.
That's
basically
the
purpose
of
this
particular
proposal,
but
it's
the
foundation
for
things
like
you
know,
being
able
to
switch
cd-roms,
discovery.
B
If
you
go
down
to
the
xml
in
the
design
document,
there's
actually
an
example
of
us
using
it
right
there.
This
type
is
network,
and
you
know
the
source
has
done
the
protocol
and
you
know
basically
that's
the
url
of
where
you
connect
to
the
network
or
resources.
F
B
B
D
Yeah,
that
makes
sense,
I
think,
the
things
so
I'm
not,
I
know
I'm
pushing.
It
sounds
like
I'm
pushing
back,
I'm
just
trying
to
get
at
the
trying
to
like
drill
down
into
like.
What's
the
ultimate
reasoning
for
why
we
would
want
this,
it
sounds
like
it's
eventually.
D
Okay,
can
we
have
some
sort
of
motivate
like
put
that
on
the?
Maybe
there
isn't
the
motivation
sure
just
a
little
bit
more
details
on
where
this
is
headed,
so
we
have
a
stronger
collective
understanding
of
why
we
would
want
to
do
this
right.
B
B
D
I
think
that's
exactly
what
we're
looking
for
in
the
design
as
far
as
like
user
stories,
we're
able
to
describe
something
very
practical
like
that,
like
that,
that's
really
helpful.
B
The
main
one
of
those
vlogs
on
it
that
I'm
aware
of
that
you
know
I
can
just
eject
and
insert
my
my
windows,
installer
and
my
video
drivers
without
having
to
add
you
know
two
cd-roms
and
then
once
I've
installed
it
and
modified
the
vms
pack
to
take
the
cd-rom
off,
because
I
don't
need
it
anymore.
B
D
A
C
A
A
Okay
on
to
the
next
item,.
A
Bug
scrub
filter.
Why
do
we
want
issues
without
labels?
Only
are
you
referring
to
this
piece
right
here.
G
A
Yeah,
usually
peter
or
roman
or
david
run,
the
the
bug
scrub,
and
this
was
just
to
try
and
like
figure
out
what
their,
what
they
did
to
run,
that
bug
scrub,
so
yeah,
correct.
A
Okay,
you
also
have
the
next
item
remove
broken.
G
Dhcp
server
yeah
this
one
is
a
little
bit
more
serious,
so
I
was
just
I
was
looking
into
the
dhcp
code.
We
have
right
now,
and
I've
noticed
that
we
have
some
sort
of
on
item
potency
mechanism
in
place.
Basically,
when
we
start
the
http
server
afterwards,
we
create
a
file
and
store
it
on
the
keyword,
private
folder,
on
the
vert
launcher
namespace.
So
if,
for
whatever
reason
we
try
to
start
the
server
again,
we
first
check
to
see
if
the
file
is
there.
G
If
the
file
is
there,
we
do
not
start
again
now.
I
think
that
we
have
merged
like
months
ago,
a
code
that
prevents
the
plug
phase
two
code
from
being
executed
twice.
So
I
just
don't
think
this
makes
any
sense
and
we
can
just
get
rid
of
this
entire
thing
altogether.
I'm
not
sure
if
someone
here
knows
more
than
I
do,
or
any
reason
why
we
would
end
up
executing
the
plug
phase,
two
network
code
or
twice.
F
I
mean
in
general
code
I
mean
we
just
discussed
if
all
networking
parts
are
it
important
at
the
moment,
but
I
mean
I
expected
anything:
any
stage
will
be
rerun
if
it
fails
like.
If
phase
one
of
networking
setup
fails
from
the
retention
side,
I
would
expect
that
it
would
retry
and
if
phase
two
fails,
I
would
also
expect
that
it
gets
retried.
G
Okay,
so
I'll
take
a
better
look
at
this
and
another
thing
about
this
like
this,
this
marker
file
it
like
it
only
works.
One
way:
it's
not
that
if
the
server
dies
we
do
not
go
and
remove
the
file
so
that
we
know
that
we
have
to
start
the
server
again
so
one
way
or
the
other.
This
thing
is
a
little
bit
broken,
but.
C
Basically,
what
happens
is
that
when
we
use
virtue
to
verify
a
con
cucao
image
before
mounting
it
to
start
the
via
the
bmi
or
migrate
to
vmi,
what
happens
is
really
interesting,
because
when
the
go
run,
time
allocate
some
memory
to
do
the
to
do
the
exact
system
call
and
basically
what
we
do
in
virtue
to
protect
ourselves
from
what
I
understood
a
memory
bomb
that
could
happen
if
you
craft
a
cucao
image
in
some
way
is
that
we
set
an
art
limit
to
the
to
the
qmo
image
invocation
and
to
virtue
itself
and
right
now
the
limit
is
to
1.2
gig.
C
C
So
I
mean
this
is
the
problem
and
I
was
trying
to
think
of
solutions,
and
I
wanted
to
gather
some
feedback
suggestions.
If
you
have
any
one
thing
I
was
thinking
to
kind
of
like
work
around.
This
was
that
to
create
another
c
binary
which
would
do
the
execute
part
of
the
virtual
route
to
which
we
would
offload
the
capabilities,
dropping
and
the
our
limit
setting
before
before
calling
qmo
image
so
that
we
keep
all
the
space
it
needs
to
to
the
core
runtime,
and
we
just
limit
qmo
image.
C
We
don't
limit
the
go
around
time
to
do
whatever
it
has
to
do,
because,
as
a
you
know,
some
background
is
that
we
already
had
some
comments:
we'd
increa,
which
increased
this
limit,
because
we
hit
this
in
the
past
between
different
goal
versions.
So
I
don't
know
if
I
was
clear
enough,
but
this
is
the
story
and
I'd
like
to
go
to
some
suggestions
or
feedback.
If
you
guys
have
any.
F
B
F
And
yeah
we
hit
it
and
then
to
end
tests
at
some
point.
When
we
updated
from
go
130
into
116,
then
we
increased
it
from
one
gigabyte
to
1.2
gigabyte
and
since
then
it
didn't
occur
a
single
time
anymore.
So
it's
a
little
bit
hard
to
tell
what's
causing
this
because,
as
far
as
I
know,
you're
building
it
you're
building
your
own
version
of
word
launcher.
Not
the
upstream
releases.
C
Yeah
yeah,
it's
it's
built
with
the
different
image
based
image
and
co-compiler,
but
basically
it
does
the
same
thing.
I
guess
there
might
be
subtle
differences
though.
Yes,
I've
been
trying
to
reproduce
this
as
well,
because
we've
been
seeing
it
as
a
you
know,
well-running
test,
and
I
couldn't
reproduce
it
and
also
to
be
honest,
it
kind
of
sounds
silly
that
to
go
around
time,
allocate
so
much
memory
to
overcome
a
1.2
gig.
F
C
F
C
F
By
the
way
I
tried
to
like
it
in
the
past,
I
tried
on
the
latest
master
or
the
latest
main,
to
reproduce
the
issue
also
by
setting
at
a
lower
limit
with
just
compiling
the
binary,
like
I
did
in
the
past
and
there
in
the
past.
I
could
reproduce
the
issue
pretty
easily
right
now.
I
can't
reproduce
it
at
all.
I
really
have
to
go
down
with
the
limit
to
to
a
few
megabyte
only
it's
very
interesting
also.
C
Yeah,
that's
that's
exactly
the
same
thing.
C
Yeah
yeah,
true
yeah,
the
the
thing
that
makes
me
uneasy
is
just
what
you
just
said
that
it's
kind
of
unpredictable
I
mean.
Maybe
we
just
pump
up
the
limit
and
it's
fine,
but
we
don't
really
know
what's
going
on.
F
C
F
Also,
it
may
make
sense
to
reiterate
that
this
limit
is
there
to
protect
against
malicious
qco2
files,
and
this
is
just
an
estimate
that
any
qcar2
file,
which
would
for
validation
where
more
memory
is
likely
bad.
But
I
mean
it's
still
just
an
estimation
which
openstack
uses
right
now
and
to
start
it
also
with
a
lower
one,
and
it
gets
increased
over
time
with
more
reddit
use
cases
coming
so
having
it.
Configurable
in
general
may
not
be
that
bad.
C
A
Okay,
thank
you.
We
don't
have
any
pull
requests
that
need
specific
attention.
Let's
check
our
mailing
list
really
quick
to
make
sure
there's
no
hanging
threads.
A
Peace
pass
through
that
might
be
worth
mentioning.
We
had
a
meeting
yesterday
morning
to
discuss
serious
bug
regarding
virtual
machines
crashing
when
pci
pass-through
devices
are
assigned
to
them.
A
A
D
A
C
A
A
F
F
A
F
Great
is
there
and
then
the
next
issue
fill
the
necessary
e2e
test.
That's
supposed
to
be
run
on
arm
yeah
yeah
howard
is
working
on
that
issue.
I'm
working
with
him
on
that.
So
we're
fine
on
that.
D
Yeah-
and
I
looked
into
that
briefly-
the
thing
that's
odd
about
it
is
just
looking
at
the
code
paths.
The
vm
starts.
We
actually
get
to
a
running
state,
which
is
odd.
F
Yeah-
and
I
just
tried
to
to
play
a
little
bit
with
the
code
to
force
it
to
not
create
the
file,
but
I
really
don't
see
a
way
in
the
code
how
we
could
get
that
far.
If
the
file
it's
not
there
and
did
the
vm
start
no
yeah,
I
don't
even
get
that
far
oh
yeah,
so
so
so
I
I
found
the
bug
which
I
told
you
in
the
code,
but
there
is
no
way
that
it
does
not
fail
in
the
next
step.
Then.
F
F
And
what's
interesting
is
is
that
we
have
another
issue
cloud
in
its
conflict.
Drive
easel
cannot
be
generated
immediately
in
windows,
vm,
and
this
person
says
that
when
it
stresses
the
server
a
little
bit,
he
ends
up
with
some
vms,
where
he
sees
in
the
lock
that
the
the
file
should
be
generated.
But
it
is
in
fact
not
there,
but
this
is
already
on
release
0
35,
but
it
could
be
the
same
issue
because
I
don't.
H
In
that
case,
the.
F
D
F
Michael
sorensen
added
some
additional
checks
to
the
tests
and
before
these
checks
they
would
just
succeed.
We
would
I
mean
we
were
just
assuming
that
the
files
were
generated
and
everything
was
fine.
D
I
don't
understand
how
anything
would
have
passed
on
azure
previously,
though,
because
we
depend
on
cloudinet
to
do
things
for
us
as
well.
So.
F
D
F
F
F
D
I
guess
my
point
is:
we
do
have
tests
that
depend
on
it
to
pass
yeah.
So
since
we
do
have
tests
that
depend
on
it,
those
should
have
failed
assume.
I
assume
those
should
have
failed
as
well
as
long
as
they
were
being
executed
by
the
smoke
test
in
a
previous
release.
So
if
it's
not
a
regression,
then
it's
curious.
Why
we're
just
discovering
this
now.
A
A
F
F
F
F
F
A
A
Okay,
okay,
well,
we're
that
takes
us
to
7
45
and
the
end
of
the
meeting,
so
I'll
return,
15
minutes
to
your
morning.
Thank
you,
everybody
and
we'll
see
you
next
week.
Thank
you,
chris.