►
From YouTube: KubeVirt Community Meeting 2023-07-19
Description
Meeting Notes: https://docs.google.com/document/d/1nE09vQWcCTW-9Ohe9oCldWrE0he-T_YFJ5D1xNzMtg4/
A
Alrighty
we
three
minutes
past
the
hour,
so
g'day
everybody
Welcome
to
today's
Cuba
community
meeting
I,
don't
know
what
day
it
is.
It
is
the
19th
of
July
welcome
I,
have
posted
the
link
to
the
meeting
minutes
and
to
chat,
and
if
you
please
add
your
name
under
the
attendees
here,
something
I
often
forget
to
do.
That
would
be
great,
it's
nice
to
know
who
attends
these
meetings.
A
We
will
start
today
with
customer.
Do
we
have
anyone
new
here
or
anyone
that
hasn't
introduced
themselves
before
in
this
meeting?
That
would
like
to
say,
g'day
and
and
tell
us
about
why
you're
here.
A
A
We
will
check
in
with
the
schedule-
and
you
might
notice
Keen
as
might
notice,
that
we've
got
a
new
release
schedule.
Thank
you
to
Luber
for
posting
this
earlier
in
the
week,
keep
the
version
1.1.
A
A
Oh
I
thought
I'd
solve
this
problem.
Well,
then,
is
anyone
else
able
to
share
their
screen
today,
because
I
am
once.
A
C
A
Well,
that
so
I
don't
think
we
need
to
click
on
the
1.1
schedule
thing
people
can
can
see
that
in
their
own
time,
early
in
the
pace,
so
there's
nothing
too
important
in
there
I
think
if
you
scroll
down
just
a
tiny
bit
more
Mark
you'll
get
to
today
to.
C
A
At
excellent
all
right,
we
have
a
relatively
short
agenda
and
no
open
floor,
so
if
anyone
does
have
anything
that
they
would
like
to
talk
about
or
question
or
just
Gap
about
by
all
means,
please
add
things
to
the
agenda
or
the
open
floor.
That's
why
we're.
D
That's
great,
thank
you,
so
just
a
quick
mention
that
it's
not
too.
D
It
should
be
noteworthy
somehow
because
in
previous
times
everything
was
scattered
throughout
a
couple
of
links,
I
think
and
I.
We
just
created
some
reports
overview
page
inside
the
cube
word
project
info
repository.
D
That
holds
a
couple
of
things
that
could
be
beneficial
for
looking
at
the
stats
of
how
the
qubit
project
itself
is
doing
like,
for
example,
flake,
finder,
flake,
stats
and
so
on.
So
if
you
would
mind,
maybe
Mark
you
might
click
on
the
cubic
reports,
overview
page
and
open
the
link,
probably
I'm
a
little
bit
higher
below
the
agenda
nodes,
yeah
exactly
that
one
and
then
opened
that
one
yeah,
exactly
that's
just
a
quick
page
that
holds
the
links
for
for
now
for
keeper
Cube
word
and
the
cube.org.
D
You
have
the
test
grid
link.
You
have
the
CI
Health
Link.
These
are
several
things
that
help
understand
how
the
overall
project
is
doing
like
what
the
virtue
currently
entails.
For
example,
what
the
flake
finder
says
about
the
test:
flakiness,
what
that
we
have
the
flag
stat
that
generate
a
condensed
view
of
where
you
can
see
which
tests
are
the
most
flaky
and
where
they
hit
the
most
and
also
which
pre-sublets
we
currently
have
with
their
settings
and
which
periodics
so
yeah.
D
That's
that's
that
and
then,
if
you
want,
you
could
also
click
on
the
flex.
That's
link
right
on
this
reports
yeah.
So
that
actually
is
I.
Think
I
showed
it
the
previous
times.
Probably
this
isn't
a
condensed
overview
of
the
flakes
of
the
flake
data
that
has
been
condensed
for
the
last
14
days,
where
you
can
clearly
see
where
we
have
the
hot
spots.
D
So
this
color
coding
is
like
by
the
way
the
the
explanation
is
of
the
upper
right
side,
where
you
can
see
what
each
what
each
color
means
so
each
color,
the
more
reddish
it
gets,
the
higher
the
the
percentage
of
total
flakes
total
flake
failures
is
so,
for
example,
what
you
can
see
here.
If
you
look
at
the
overall
statistics
per
Lane,
then
you
can,
for
example,
see
that
the
six
storage
126
line
and
the
1276
Dodge
Lanes
have
both
been
very
flaky
over
the
last
14
days.
D
Most
of
you
might
be
aware
that
we
are
currently
having
issues
on
those,
so
they
this
overview
gives
you
a
general
impression
of
where
we
are
hit
the
most
by
flaky
tests.
If
you
scroll
down
a
little
bit,
then
you
go
to
the
yeah
that
that's
already
enough.
E
D
You
so
that's
a
list
of
the
flaky
tests
that
we
encountered
over
the
last
14
days
and
those
are
ordered
by
the
overall
failures
descending.
So,
for
example,
for
now
you
see
that
there
is
a
six
search
test
that
has
been
very
flaky
in
recent
days,
but
it's
progressing
to
be
less
flaky.
So
when
you,
for
example,
look
at
the
per
day,
that's
exactly
the
impression
that
you
can
get
from
that.
So,
okay,
I
think
I
talked
enough
about
that.
I
think
I
already
introduced
it
a
couple
of
weeks
ago.
D
F
It's
probably
a
stupid
question,
but
I
I
did
look
through
this
the
other
day
for
any
kind
of
instance,
I
put
preference
favors,
and
there
was
one
preference
test
that
I
think
is
in
one
of
the
snapshot
lines
that
was
coming
up.
It
looked
like
we
were
continuing
to
run
tests
after
the
first
failure.
Are
all
the
lanes
set
up
so
that
even
if
one
of
the
tests
fail,
we
continue
attempting
to
run
additional
tests.
F
D
It's
not
good.
Sorry
I
was
just
wanting
to
chime
in
already
because
normally
in
general,
we
run
the
tests
in
two
stages,
if
it
so
at.
First,
the
parallel
tests
are
being
run
and
then
the
serial
tests
are
run
afterwards.
So
if
the
parallel
tests
fail,
then
for
pull
requests,
the
entire
test
run
is
stopped,
so
the
serial
test
run
is
not
being
run
after
that,
but
for
the
periodic
runs.
D
This
will
be
different
because
we
want
to
see
all
test
runs
there
and
therefore
we
run
in
the
C
in
the
periodic
test
runs.
We
run
both
parts,
so
we
are
on
the
parallel
and
the
serial
tests.
D
But
to
get
back
to
your
question
in
general,
we
don't
stop
after
the
first
test
failure,
because
we
want
to
see
for
each
part
at
least
all
of
the
tests
in
their
entirety,
so
that
we
don't
Shadow
other
tests.
So
imagine,
for
example,
that
you
might
have
one
test
that
has
a
couple
of
other
or
one
Buck
introduced.
D
That
introduces
a
couple
of
things
that
are
breaking
and,
of
course,
you
want
to
see
all
the
areas
where
it's
breaking
so
it
should
make
sense
to
still
run
all
the
tests.
Somehow
does
that
answer
your
question
or
do.
F
I
yeah
yeah.
Now
it
does
yeah,
it
does
I
think
in
my
specific
case,
it
was
just
leading
to
that
test
showing
up
in
this
Matrix.
Even
though
the
test
itself
wasn't
the
problem,
there
were
failures
in
the
environment
prior
to
that
I
think
that
were
leading
to
the
failure,
so
it
kind
of
muddies
the
water
a
little
bit.
I
I
appreciate
the
logic
for
trying
to
run
everything
so
that
a
Dev
might
get
an
overview
of.
You
know
the
impact
of
something
but
yeah.
D
So
can
I
ask
again
about
that:
how
came
how
did
it
come
to
that
test?
Failing
were,
was
there
some
cleanup
missing
in
another
test,
or
how
did
it
get
to
that.
F
I,
don't
have
everything
to
hand,
but
the
failure
of
the
test
was.
It
was
something
basic
about
just
reaching
out
to
the
API
like
the
API
had
fallen
over
so
actually
posting
something
to
it.
F
I
will
go
back
and
we
can
chat
offline
about
it
because
yeah
it
was
like
it
was
an
environmental
problem
that
was
leading
to
it.
That
happened
previously,
so
it
was
just
gonna
lead
to
noise.
Essentially
in
this
Matrix
I've.
G
Been
looking
into
this
quite
a
bit,
if
you
want
to
talk
about
it
Daniel
but
yeah,
basically,
what
was
happening
in
a
lot
of
these
storage
Lanes
is
that
a
TV
was
in
memory
and
the
Etsy.
It
was
only
512
Meg
for
its
persistent
storage
and
that
was
filling
up
and
then
it's
so
ET
CD
would
fail,
and
then
that
would
cause
you
know.
Api
requests
to
fail
and
some
of
those
API
requests
were
for
like
renewing
leadership
leases.
So
then
the
controllers
and
stuff
would
exit
and
restart
and
yeah.
G
So
basically
and
then
switching
back
to
persistent
NCD
fixed
a
lot
of
this,
but
it
was
interesting
in
that
it
did
expose
I,
think
some
issues
and
failover
of
like
the
snapshot,
controller,
which
I
think
is
the
test
that
Lee
was
talking
about
so
sort.
F
G
Some
unintentional
chaos
testing
was
happening
and
led
to
some
potential
issues.
D
Yeah,
that's
always
the
problem
with
e2b
tests
right
there
are
lots
of
components
that
kind
of
side
effects
somehow
that
might
lead
to
concluding
failures
that
actually
no
one
attended,
somehow
right,
so
yeah,
okay,
so
yeah
thanks
yeah.
We
of
course
we
can
discuss
this
offline.
If
you,
if
you
want
okay.
B
A
Oh,
we
have
nothing
else
on
the
agenda
or
the
open
floor.
The
pull
requests
only
has
kindly
included
say
list
of
PR's.
F
Yeah
I've
just
been
thinking
about
this
Upstream
for
a
little
while,
so
we've
got
a
couple
of
instance,
type
PRS
that
have
been
reviewed
but
just
need
a
bit
a
further
review
and
potential
approval,
so
I've
pinged
a
few
folks
privately
for
that
yeah
I
thought
I'd
bring
up
in
the
meeting
as
well
they're,
pretty
straightforward.
It's
nothing!
Nothing
too
big
in
the
queue
at
the
moment,
but
yeah.
If
any
approvers
on
a
call
have
time
any
help
would
be
appreciated.
A
And
I
think
I've
got
one
of
yours
coming
up
as
well.
Well,
I
could
please
move
down
to
186
to
1086,
allow
live
updating,
VM
mode
affinity.
A
A
Yep,
so
that
one
is
there
sitting
and
waiting
for
some
people
to
have
a
look
at
it
and
says
that
people
are
excited
about
it,
but
haven't
had
a
chance
to
review
it
yet
so
if
that
falls
into
your
wheelhouse
by
all
means,
please
have
a
look
at
that,
and
my
memory
certainly
is
the
next
one
yours
host
disc
or
am
I
thinking
about
something
totally
separate.
F
No,
that
one's
mine
as
well-
it's
not
important,
though
it's
just
a
follow-up
that
we
would
want
to
miss
a
look
at
so
yeah.
If
people
are
interested
feel
free
to
look,
it's
a
very
obscore
part
of
the
code
base,
so
I
doubt
anyone
will
be
but
yeah
it's
not
very
important,
but.
A
I
just
try
and
hit
things
that
yeah
don't
have
any
otherwise
attention
on
them.
A
All
right,
the
mailing
list
review
we've
got
two
new
design
proposals
that
I
think
are
worth
having
to
dip
into
again
this.
This
is
one
of
yoursly
I,
like
how
active
you've
been
this
this
past
week,
yeah
just.
F
Spamming
everything
yeah.
This
is
an
interesting
one.
F
So
the
background
to
this
is
the
the
current
implementation
of
instance,
types
in
Cuba
have
you
relies
on
taking
snapshots
of
the
instance,
type
and
preference
resources
when
we
first
see
a
VM
using
them,
and
so
we
always
have
a
consistent
view
of
the
resource
and
we
can
always
apply
it
and
generate
the
same
VMI,
essentially
at
the
end,
and
we
obviously
stash
them
in
controller
revisions
and
because
that
point
in
time,
copies,
they're,
obviously
pinned
to
a
specific
version
of
the
instance
type
API
or
the
crds
that
we
use.
F
That
means
we've
got
various
versions
currently
in
controller
revisions
and
we're
looking
to
move
to
V1
for
the
overall
API
and
crds
with
Cube
1.1,
and
so,
if
that's
introducing
yet
another,
so
the
the
whole
design
proposal
is
just
about
moving
the
stashed
objects
in
those
controller
revisions
forward.
So
we
can
RM
RF
the
code,
the
order
code
as
soon
as
possible
and
just
load
the
overhead
of
that
feature
on
the
code
base
and
yeah.
As
the
email
says,
the
design
for
this
is
strangestly
copied
from
Cube
store
storage
version
migrator.
F
F
The
design
proposal
has
a
load
of
to-do's
left
in
it
because
it's
whip
in
its
draft,
mostly
asking
kind
of
my
existential
questions
but
questions
about
where
they
should
go.
Should
it
be
control
additional
controllers
within
the
controller
or
do
we
need
to
start
I,
think
and
start
thinking
about
splitting
that
out
the
the
overall
design?
F
Do
we
need
to
really
have
two
controllers
I've
tried
to
address
some
of
the
kind
of
points
I
think
I
could
have
come
up
about
this,
potentially
being
better
as
a
job
and
given
the
recent
machine
type
migration
work,
that's
going
on
in
the
design
proposal.
That's
still
being
worked
on
for
that
and
as
part
of
the
design
proposal
for
this
so
yeah,
that's
still
something
to
discuss
as
well,
but
yeah
if
anyone's
got
any
comments
now
on
the
call
I'd
love
to
hear
them
otherwise,
yeah.
F
Please
take
a
look
at
the
design
proposal
and
throw
comments
that
way,
positive
or
negative.
Anything
would
be
good
at
this
point.
Yeah,
that's
pretty
much
all
I've
got
to
say.
A
I'm,
jumping
out
of
that
now
so
I'll
leave
people
to
respond
to
that
on
the
mailing
list.
The
second
design
proposal
is
raised
by
Simone
and
I
must
admit:
I
didn't
actually
read
it,
but
it's
something
about
logs
for
the
serial
console.
A
A
The
third
one
here
is:
there
isn't
an
issue
related
to
this.
It's
a
CPU
pending
issue,
I,
don't
know
if
anyone
is
able
to
help
with
this
right
now,
but
the
thing
that
really
struck
me
about
this
one,
which
actually
is
related
to
what
I
was
saying
to
Victor
before
we
started
the
call
the
versions
of
these
same
really
really
old,
Cuba
zero.
A
Four
three
with
the
kubernetes
version
of
1.16
I,
think
1.16
is
from
three
four
years
ago
and
now
four
or
three
is
two
years
old
like
as
a
community.
We
say
you
know,
we
support
the
previous
last
versions,
but
we
try
and
be
helpful
to
people
out
there.
Who
are,
you
know,
potentially
start
running
old
versions
of
things?
At
what
point
do
we
draw
the
line
and
say
this?
A
Is
this
is
too
old
or
or
that
it's
more
just
like
as
a
philosophical
question,
if
anyone
has
any
thoughts
or
opinions
on
this.
B
B
A
The
world
began
the
training
that
makes
sense.
Sport
is
supported
and
we
have
bandwidth
to
cover.
Then
people
can
be
helpful,
but
otherwise
you'll
need
to
upgrade
at
some
point.
A
That
seems:
does
everyone
else
have
any
thoughts
or
opinions?
Does
that
seem
like
a
reasonable
take
to
have.
C
It's
a
reasonable
ask
to
ask
people
to
get
to
current,
to
reproduce
it
in
a
current
environment,
because
that's
that's
where
you'll
get
the
most
attention
and
Community
Support.
Everything
else
is
voluntary.
I!
Imagine
for
outside
of
those
bounds.
A
And
when
it
comes
to
something
like
the
the
user
guide,
which,
unlike
other
projects
or
products,
we
have
a
single
single
version
user
guide.
How
far
like
do
we
want
to
remain
relatively
stringent
with
with
that
level
of
supportability,
where
it's
like?
If
the
the
thing
in
question
is
regarding
the
live
migration,
what
do
you
call
it?
The
feature
label
the
future
game,
we're
now
past
that
stage
where
we
no
longer
have
to
tell
people
that
we
need
to
enable
the
future
gate
explicitly.
A
Seeing
as
how
that's
what
I
think
five
versions
ago,
you
know
it
do
we
get
to
the
point
where
our
user
guide
really
is
just
a
three
version:
Hardline,
no
more
where's,
the
gray
area
here.
B
F
D
D
Yeah
I
I
looked
at
the
issue
and
I
saw
that
there
was
someone
else
chiming
in
on
that,
and
that
seemed
to
be
helpful.
If
you
could
click
on
the
issue,
then
you
might
see
that
there
was
some
response
so
inside
the
inside
the
mail
mail
thread
that
he
mentioned
an
issue
which
he
opened
and
yeah
at
the
bottom
I
think
it's
a
it's
below
that
giant
screenshot.
There
should
be
something
like
yeah
there.
It
is,
and
someone
else
answered
and
tried
to
help
him.
D
I,
don't
know
that
person
to
be
honest,
but
they
seem
to
resolve
it,
but
anyhow
yeah
I
think
it
might
be
valuable
to
just
add
some
command
probably
about
you
would,
of
course
get
more
support
if
you
had
used.
D
If
you
would
use
the
latest
version,
something
like
that,
but
yeah
it
seemed
to
be
like,
like
it
was
CPU
manager
related,
so
I
hope
they
they
sort
out
that
and
yeah
another
take
for
me
is
as
long
as
the
the
the
behavioral
feature
doesn't
change
I
think
we
could
still
probably
try
to
provide
support
as
we
needed,
but
yeah
all.
That
said,
of
course,
I
agree
with
with
everyone
here
in
the
calls
that
this
version
is
very
old
and
that
you
rather
should
upgrade
somehow.
D
By
the
way,
there
is
actually,
of
course,
there
is
a
release,
support
paragraph,
so
I'm
just
dropping
the
link
in
the
chat,
if
you
don't
have
it
handy
so
that
we
can
just
probably
attach
that
to
this
one.
D
C
A
First,
one
KVM
process
and
host
consumed
hybrid
CPU,
much
higher
than
observed
inside
guests.
Now,
I,
don't
know
if
this
is
the
same
person
reporting
this,
but
I
do
remember.
Someone
else
had
reported
this.
Maybe
it's
not
this
I
think
I'm
gonna
slack
channel
two
or
three
weeks
ago
and
I
don't
know
if
I
would
help
them.
E
Yeah,
this
is
normal.
Kimu
just
allocates
all
of
the
memory
when
it
starts
so
yeah.
They
expect
only
the
used
memory
to
be
used
by
the
gaming
process,
but
that's
not
the
case.
B
But
I
think
we,
we
saw
some
some
issues
like
this
even
before,
but
I
do
not
believe
I
believe
that
that
was
a
bug.
B
A
B
You
go
is
there?
Is
there
a
way
from
mentioned
in
this
issue?.
A
A
A
F
Yeah,
you
can
do
device
tagging
on
other
platforms
so
that
I
can
still
tag
the
disks.
This
is
this
a
missing
feature
in
Cube
for
the
for
the
disks
to
be
targets,
and
it
says
the
interface
is
attacked
correctly.
I,
don't
know
if
we
support
it
for
the
disks.
Thank
you,
but.
A
Michael,
do
you
happen
enough
to
stop
your
head.
G
A
A
A
And
nothing
else
has
been
added
to
the
agenda
or
the
Open
Floor,
which
brings
us
to
the
end
of
there's
Cuba
community
meeting
and
just
before
we
all
sign
off.
Does
anyone
have
anything
that
I'd
like
to
add
off
Peace
by
any
chance.
A
A
good
question
I
have
not
really
heard
and
in
fact
that
bug
that
we
looked
at
I
think
is
the
first
time
I've
seen
it
deployed
and
having
a
bug
raised
from
it.
I'm,
probably
not
the
best
person
to
ask
on
that.
The
the
Tweet
which
I
we
do
get
to
look
at
the
analytics
of
quote,
unquote
blew
up
so
I
I.
Guess
that's
something.
B
A
B
I
was
maybe
maybe
mention
or
bring
to
attention
one
PR
I
I
post
in
the
chat.
It's
a
design
purpose
of
our
vtpa
using
it
just
for
networking
right
now.
I
guess
something
interesting
already,
but
hey
male
or
you
know.
H
Yes,
it's
it's
pretty
similar
to
SRV.
So
in
this
sense,
if
we
solve
one
I
think
we
solve
the
second,
this
one
as
well.