►
From YouTube: KubeVirt Community Meeting 2019-02-27
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
B
So
finding
the
unmute
and
hopefully
I
can
be
hurt
yeah,
so
I
finalized
the.
Finally,
the
discussion
in
the
mailing
list
went
to
some
conclusions,
so
I've
been
able
to
summarize
them
in
the
design
dog
which
is
linked
in
the
notes,
please
feel
free
to
comment
and
provide
a
feedback.
Basically,
what
we
agreed
is
that
there
will
have
to
be
a
high
level
controller.
There
is
still
an
open
question
how
to
implement
that
and
the
quick
choice
is
implement
custom
controller
or
have
it
as
a
kubernetes
jobs.
B
Job
discussion
is
in
the
document
and
still
open
question
is:
what
is
the
result
we
will
provide
where
we
will
store
the
result
of
snapshots.
There
are
three
options:
either:
create
new
virtual
machine
annotated
as
a
as
a
as
a
result
or
create
a
template,
annotated
as
a
result
or
create
custom
resource
which
will
be
a
result
of
virtual
machine
snapshot
again.
The
discussion
for
this
is
in
the
document,
so
feel
free
to
provide
feedback,
and
that's
it
from
me.
Thank
you
sure.
A
B
Yeah
there
are,
technically
speaking,
there
are
three
ways
to
store
the
result
of
virtual
machine
snapshot.
First,
one
is
to
create
a
new
virtual
machine
that
will
be
annotated
as
a
result,
it
has
its
downsides,
because
one
of
the
requirements
for
snapshots
is
that
it
cannot
be
run
able
you
there.
Basically,
there
has
to
be
no
way
to
run
by
option
the
snapshot.
B
Second,
one
is
a
template,
create
a
template
which
will
be
annotated
as
a
snapshot
again.
This
is
not
renewable,
but
it's
editable
and
it
has
to
be
heavily
annotated
to
prove
to
comprise
the
information
that
it
is
a
snapshot
and
to
point
to
the
overall
on
information,
and
the
third
option
is
a
custom
resource
for
astern
spiritual
machines,
nakshatras
all
which
will
store
everything
in
one
place
and
solves
all
the
issues.
It's
not
reliable.
It's
not
editable,
it
stores
the
result
it
links
to
the
volumes
nutshot
it
links
to
the
copy
of
virtual
machine
spec.
B
B
C
Wanted
to
ask
a
question
about
how
you
request
a
snapshot
to
virtual
machine
snapshot
to
be
created
in
the
first
place.
Maybe
it's
in
the
dock,
but
we
are
aware:
I
mean
a
little
bit
of
a
discussion
here.
So
I
assume
that,
in
order
to
be
declarative,
you
would
need
to
post
something
to
the
cluster
to
request
it,
which
I
would
suggest
might
be.
C
C
C
Update
okay,
yeah
I
just
wanted
to
jump
on
and
kind
of
talk
about
a
couple
of
things
that
we're
doing
over
in
really
in
the
containerized
data
importer,
land
and
but
with
some
touch
points
to
convert.
So
one
of
the
things
you
know
the
discussion
that
we
just
had
around
virtual
machines
snapshots.
Some
of
these
more
advanced
features,
start
to
require
storage
that
actually
has
more
capabilities
than
something
like
the
local
storage
provider
is
going
to
be
able
to
supply.
C
So
if
we
want
to
have
reliable
functional
tests
for
these
features
that
we're
adding
it's
going
to
be
necessary
to
pull
in
more
capable
storage.
Ceph
is
a
really
great
candidate
for
this,
because
it's
you
know
it's
freely
available.
There
are
containerized
deployments
of
it
that
run
on
top
of
kubernetes
clusters,
and
it
also
happens
to
have
all
the
features
that
we're
looking
for
to
enable
you
know
more
complex
virtualization
flows
like
snapshots
and
cloning.
C
So
one
of
the
things
that
our
team
is
looking
at
doing
is
coming
up
with
a
basically
a
storage
deployment
of
Ceph
that
can
be
incorporated
into
convert
CI
so
that,
basically,
when
you
do
and
make
cluster
up,
you
would
have
storage
class,
that's
served
by
Assefa,
CSI
provisioner
and
everything
ready
to
go
out
of
the
box.
So
we're
looking
to
do
that.
We
would
appreciate
you
know,
help
in
terms
of
best
practices
and
attention
to
any
PRS
that
come
out
at
the
point.
At
this
point,
we're
just
still
working
on
pulling
it
all
together.
C
C
Okay,
so
the
next
one
is
in
CDI
the
kubert's
already
supporting
block
TVs
for
running
virtual
machines
disks.
We
are
looking
to
enable
the
flows
and
CDI
to
import
clone
and
upload
to
those
devices
to
those
block
devices
to
really
complete
that
block.
Pv
story
across
you
know
creating
and
provisioning
virtual
machines,
so
that
work
is
ongoing
at
some
point.
We're
gonna,
be
you
know,
caught
having
wanting
Qbert
to
update
to
numeration
zuv
CD
eyes,
we
release
this
functionality,
so
that's
coming
up
and
then.
C
Finally,
the
last
update
that
I
would
share
is
around
the
upload
ingress.
So
the
way
that
CDI
upload
works
is
that
the
cluster
has
an
ingress
where
the
outside
users
uploading
data
to
in
the
past.
That
has
had
to
be
it's
basically
like
a
magic
string
that
you
kind
of
need
to
know
how
to
get
yourself.
We
are
adding
a
new
CDI
config
config
map
that
is
going
to
actually
supply
for
you,
the
the
proper
URL
that
you
need
to
use
based
on,
if
the.
C
A
D
D
D
D
To
some
external
thing
that
actually
creates
the
CI
pods
so
now
I
started
to
hit
some
of
the
issues
in
the
in
the
latest.
Ppr
runs
that
it
seems
to
be
like
what
we
already
saw
before
and
what
was
solved
by
my
fixes
in
this
external
thing.
I
don't
know.
First,
like
my
questions,
are
first
I,
don't
really
know
where
the
code
to
actually
create
these
iPods
leave
and
I
would
like
to
know
to
see
what
what
are
they
like.
D
The
actual
code
for
the
fixes
was,
and
not
just
description
of
it
and
second
I
would
like
to
know.
Is
it
deployed
automatically
to
all
the
notes
or
the
fixes
that
go
there,
or
there
is
a
chance
that
a
fix
was
applied
to
one
of
the
CI
knows,
but
not
to
another
one,
and
so
now
I
see
some
of
the
failures,
because,
because
that
that's
what
happened?
Perhaps
someone
knows
about
how
CI
boards
are
treated
or
where
to
look
for
the
code.
E
Don't
think
anyone
can
answer
that
so
you're
just
to
rephrase
what
you
said,
your
issue
is
that
you
do
something,
so
you
modify
something
which
requires
you
to
be
to
execute
the
job.
On
the
right
note,
and
you
now
don't
know
if,
if
your
scripts
are
executed
on
the
right
note
or
on
the
rock
and
the
scripts
just
fade
for
whatever
reason
or
if
you're,
on
the
wrong
notes
altogether,
I,
don't
I,
don't
think.
D
F
D
F
D
That's
good
and
another
thing
that
is
related
and
related,
so
we
Daniel
wrote
some
some
C
and
I,
which
is
really
just
the
bar
script.
That
is
called
once
per
SR
IVC
I
poured
by
CI
like
CI
cluster,
not
really
the
cluster
that
is
under
the
test,
and
it
is
triggered
for
CSR
avai
ports
to
move
nothing
devices
from
from
bare
metal
into
the
iPods.
D
E
E
E
G
D
D
D
E
We
deploy
on
ours,
yeah
yeah,
but
all
I
mean
is
we
don't
have
to
not
just
have
the
cube
root.
Ci
class
did
the
CI
classes
in
the
refere
other
stuff
they're
into,
for
instance,
the
test
images.
So
it's
I
don't
know
it's
in
in
communities.
They
are
going
even
further
at
their
their
development
near
their
whole
CI
system
in
one
repo
having
all
the
waste
containers
in
there
and
so
on.
Yeah
I
see
where
it
comes
from.
It's
monolithic,
culture,
I,.
A
F
The
last
part
that
we
need
to
integrate
if
they
masquerade
in
May
a
production,
ready
solution
for
that
and
I
will
like
it.
Someone
can
comment
there.
David
already
started,
do
and
also
yeah.
It's
a
very
small
PR.
The
only
thing
there
that
is,
let's
say
a
change
for
us,
is
when
you
use
East
EO.
It
will
attach
another
container
that
it's
called
the
East
your
proxy
and
this
your
body
will
not
be
in
ready
state
until
you
have
a
service
that
is
part
of
the
service
mesh.
F
F
Apr
22
the
dome,
XML
and
key
value,
and
not
only
use
it
on
the
DHCP
options
that
we
are
using
because
their
windows,
they
can't
read
it
so
now.
The
next
part
I
need,
is
to
update
our
Windows
driver
that
we
are
provided
to
the
latest
one
and
not
to
the
stable,
because
the
statement
doesn't
support
to
read
the
MTU
from
the
driver.
So
this
is
my
next
step:
I'm
going
to
open
a
PR
ability
to
change
from
stable
to
light.
G
G
So,
obviously,
when
you
create
the
VM
and
ask
for
the
feature
you
don't
know
whether
the
host
kernel
will
support
it
or
not,
and
if
you
ask
for
silic,
for
example,
and
the
host
kernel
is
not
good
enough,
the
VM
will
not
start,
which
is
not
an
outcome
we
want.
So
we
need
to
find
a
way
how
to
deal
with
this
on
the
execution.
G
You
know
at
the
execution
time,
because
you
can't
tell
before
you
know
before
you
schedule
the
part
and
the
other
thing
is.
You
could
depend
on
it
in
the
scheduler.
But
you
don't
really
know
if
this
feature
is
going
to
be
available
in
the
cluster
at
all,
and
it
might
not
be
important
enough
for
you,
so
it
might
be
case
of
use
it
if
it's
available
otherwise
ignore
it.
There's
no
mechanism
to.
G
B
A
G
G
H
G
The
default
CPU
model,
because
right
now
the
hard-coded
default
is
host
model.
Then
you
install
C&V,
meaning
all
the
layered
products
as
an
admin.
You
might
want
to
say
that
all
my
VMs
that
are
started
without
explicit
CPU
model
will
be
messed.
We're
compatible
things
like
that.
That's
important
for
heterogeneous
clusters,
so
you
know
if
you,
even
if
you
remove
node
ID
note,
you
will
still
keep
like
migrate,
live
migration
capabilities.
G
Nobody
had
to
think
about
it
till
now,
because
live
migration
was
never
fully
working
and
support
it.
But
now,
since
we
are
getting
close
to
having
live
migration,
actually
supportive,
we
might
believe
this.
The
couple
of
things
we
hit
where
that
first,
that
was
supposed
to
adding
it
to
the
config
map.
E
I
I
So
if
it
turns
out
this
as
a
pattern-
and
we
need
the
ability
to
express
lots
of
defaults
that
take
place
close
to
wide
across
lots
of
namespaces,
then
I
think
it
would
be
wise
for
us
to
go
ahead
and
adopt
something
like
a
global
preset
or
something
like
that
that
we
can
set
and
is
gets
applied
everywhere,
because
if
we
start
doing
these
one-off
things,
the
thing
I'm
concerned
about
is
as
a
our
API
evolves.
Then
what
happens
with
the
we're
mapping
things
in
a
config
map
to
a
version?
2
API
is
that
version.
I
G
So
I
think
I
have
an
answer
to
death,
so
I
have
some
experience
in
overt
I've
been
there
for
I,
don't
know
six
years.
Maybe
so
there
are
a
couple
more
options
that
are
cluster
wide,
but
none
of
them
would
be
fit
for
Cubert
because
they
were
mostly
resource
management
related
in
terms
of
memory,
ballooning
and
stuff,
and
that's
not
that's
not
something.
Qbert
cares
about
so
CPU
model
and
the
compatibility
of
those
of
the
of
the
VM
is
probably
the
only
thing
keyboard
cares
about.
Everything
else
is
probably
somebody
else's
job.
It.
E
G
G
G
You
will
just
add
a
device,
but
you
will
not
control
the
actual
resources,
because
that's
not
level,
you
need
to
take
a
look
at
all
the
VMS
and
but
that's
a
different
discussion.
Okay,
like
yeah
I,
think
Qbert
should
just
been
able
to
device
on
the
VM
side,
the
same
case
and
maybe
just
flag,
the
V
and
somehow
but
shouldn't
do
the
actual
resource
rebalancing
itself
by
itself
with
CPU
model.
That's
hard,
because
the
CPU
model
of
the
VM
XML
has
to
match
the
pot
specification.
G
G
I
Could
be
overkill
if
it's
not
you
had
a
clear
use
case
beyond
this
one
feature
and
that
that's
what
I'm
trying
to
tease
out,
if
we're
pretty
confident
that
it's
just
going
to
be
CPU
model
I,
think
I'd
feel
more
comfortable
with
it
being
in
that
config
map.
If
we
eventually
down
the
line
and
see
this
can
be
used
for
more
I
would
hate
for
us
to
keep
using
that
precedence
that
we
set
by
adding
into
the
config
map
and
instead
move
to
something
like
a
global
preset.
Forget.
A
A
G
I
was
actually
leave
our
discussants
at
least
I
was
liking
to
look
at
it.
It
was
one
of
the
options.
The
thing
is
the
preset
right
now.
If
you
define
something
in
there
and
then
you
change
it
in
the
VM,
it
will
throw
warnings
which
is
not
nice,
too
user
I
mean
it'll,
get
a
warning
every
time
it
tries
to
start
a
VM
with
explicit
CPU
model.
It.
G
A
G
H
So
sorry
for
jumping
in
this
Fabien,
actually
one
thing
I
wonder
it
is,
as
we've
just
discussed,
the
preset
mechanism,
if
migrations
today
are
bound
to
namespace,
so
I
understand
that
in
in
overt,
for
example,
the
cluster
we
have
a
cluster
default
for
for
the
CPU
model,
I
wonder.
But
oh
yeah
and
I
wonder
if
four
for
four
cubed
it
makes
sense
to
actually
have
that
I
mean
I.
Would
question
I
would
put
it
out
as
a
question.
It'll
make
sense
term.
H
They
have
a
namespace
default,
because
why
we
could
think
about
the
reason
I
mean
the
reason
why
I
thought
of
it
right
now
is
because
migrations
are
also
only
taking
place
within
a
namespace,
and
if
we
think
that
the
namespace
is
a
is
a
soft
boundary
for
a
tenant
right
or
for
project,
then
we
could
say
it
might
even
make
sense
to
have
that
to
have
this
default
for
the
CPU
mental
model
per
namespace.
Yeah.
G
Fabiana
I
have
a
couple
of
counter
examples.
First,
I
will
want
to
revisit
the
process
by
themselves.
Studio
thing:
the
issue
with
current
presets
that
was
never
solved
is
that
there
is
no
clear
ordering.
So
if
you
have
multiple,
you
don't
know
which
is
going
to
be
applied
first
and
now
to
do
fabulous
things
it
might
work,
except
notes
are
not
named
spaced,
so
you
get
the
same
notes
for
every
room
space.
Second,
if
you
it
would
require
it.
G
Somebody
creates
this
default
for
every
new
namespace
created,
so
the
operator
somebody
would
have
to
watch,
watch
namespaces
and
create
some
kind
of
a
default
which
might
be
tricky.
I
mean
it
would
require
everybody
to
set
their
default.
They
would
have
to
know
about
it
and
it
wouldn't
be
default
anymore.
It
would
be
configuration
/,
namespace
and
customers.
I
mean
the
the
users
actually
might
not
know
about
it.
This
is
actually
something
we
want
to
abstract.
Anyway.
I
mean
the
user
shouldn't.
G
You
know
in
openshift
world
it's
in
in
ref
you
care,
but
in
open
ship
world
you
don't
really
care
about
notes
too
much,
and
if
we
want
to
keep
that,
then
basically
the
the
cluster,
the
OpenShift
cluster
sysadmin,
would
know
what
the
baseline
CPU
model
is.
So
he
would
configure
the
whole
openshift.
That's
why,
with
namespaces
you
wouldn't
get
that.
H
H
G
G
I
E
Yeah,
so
let
me
just
repeat
that
I
gave
yesterday
a
short
presentation
about
fro
and
what
it
can
do
for
us,
and
also
edit.
The
link
to
the
video
into
the
community.
Talk
feel
free
to
have
another
look
reach
out
to
me.
If
you
have
further
questions
and
if
there
are
no
blocking
concerns
within
the
next
days,
I
will
just
continue
to
migrate
everything
over
and
we'll,
hopefully
soon
improved
the
overall
experience
regarding
to
running
the
tests
and
having
success
with
this.
A
All
right
on
the
open
floor,
one
thing
to
note
is
daylight:
savings.
Confusion
week
is
coming
up.
The
United
States
moves
to
daylight
savings
on
March
10th,
which
is
I
think
three
weeks
before
most
European
countries.
This
meeting
follows
the
CET
or
CST
time
zone,
which
is
around
central
Germany,
basically
so
for
the
dates
of
the
13th,
20th
and
27th.
This
meeting
will
actually
be
at
11
a.m.