►
From YouTube: Multi-clusters made easy with Liqo: getting rid of your clusters by keeping them all | A. Palesandro
Description
KCD Italy 2021* - The video of the talks are now available!
Many companies are experiencing a dramatic increase in the number of their Kubernetes clusters, for reasons such as geographical/legislative constraints, data/service replication, etc. However, when the number of clusters increases, the complexity of deploying apps, managing the entire multi-cluster infrastructure, and keeping its state under control, becomes rapidly an unmanageable problem.
A possible solution is Liqo, an open-source project that simplifies the creation of multi-cluster topologies by replicating the Kubernetes “cattle” model also to clusters.
Liqo creates a virtual cluster that spans multiple real clusters, either on-prem or managed (AKS, EKS, GKE), and instantiates the desired applications seamlessly in the appropriate cluster. This talk will discuss the potentials and roadblocks of this vision and highlight how Liqo brings multi- cluster transparency to the users.
🤓 SPEAKER: Alex Palesandro, Research Assistant @ Politecnico di Torino
👉🏼 Here the slide deck: https://www.slideshare.net/KCDItaly/multiclusters-made-easy-with-liqogetting-rid-of-your-clusters-alex-palesandro-keeping-them-all
______________________
* KCD Italy 2021 is the first edition of a community-organized non-profit event that will take place on November 17th and 18th and will gather adopters and technologists from open source and cloud native communities for education, collaboration, and networking.
📯 Follow us to be notified about the upcoming MeetUps: https://community.cncf.io/kubernetes-cloud-native-milano/
A
B
Well,
we
are
ready
to
officially
start
with
today's
line
up,
which
is
truly
amazing.
Just
two
important
information
for
today
you
can
write
your
questions
for
the
speakers
in
the
session
q
and
in
the
that
you
have
on
your
right
and
I'll.
Show
it
to
you.
Also
in
here
is
a
simpler
slide.
So
please
ask
your
questions
in
which
we
collect
them
and
ask
our
speakers
and
then,
more
importantly,.
B
How
do
you
see
the
t-shirt
made
the
t-shirt
with
the
logo
of
the
eataly
cases
we
have
made
available
a
number
limited
number
of
t-shirts
during
the
day,
be
very
careful
in
the
chat
for
a
forum
will
appear
with
a
limited
number
to
be
able
to
order
your
t-shirt.
You
can
order
one
t-shirt
per
person,
I
invite
you
to
be
very
careful
in
the
chat,
so
you
do
not
miss
this
opportunity.
We
will
then
send
them
to
end
of
the
event.
The
very
last
thing,
all
the
videos
you
saw
yesterday.
B
B
C
They
can
be
a
joke
hi.
There
are
three
mcclures.
No,
I
don't
do
it
like
the
simpsons,
but
hello,
everyone,
I'm
dario,
tarchi
tel.
Maybe
you
know
me
because
I'm
also
the
administrator,
the
kuber
rete
group
on
telecom,
but
that's
another
story,
and
today
I'm
here
to
represent
classics,
so
I'll
try
to
be
a
little
more
polight.
What
we
do
in
classic.
We
are
the
startup
we
are
distributed.
We
are
between
turin,
milan,
catania,
ps,
rome,
now
even
worse
than
new
york,
and
we
are
very
verticalized
super
in
testing.
C
We
are
the
creators
of
capsules,
which
is
an
open
source
project
flex
on
products
which
deals
with
managing
the
multi-
tenancy
and
in
the
meantime
we
do
many
interesting
things
with
very
particular
and
very
complicated
coopers.
So,
yes,
we
are
absolutely
a
nerd
company.
We
are
absolutely
delighted
to
be
so
many.
C
Absolutely
then,
we
have
alex
fa
alessandro
with
us
and
formerly
worked
as
arendi
engineer
at
orange
l'ubs
specializing
in
multi
cloud
technologies.
He
then
dedicated
several
years
as
a
consultant
to
drive
the
adoption
of
container
automation
and
gubernatis,
and
his
research
includes
multi
systems
cloud
expo
computing
and
networking
function,
gus
position,
many
buzz
wars,
but
certainly
very
interesting.
No
more
talking
good
talk,
alex.
A
A
D
Will
tell
you
about
di
nico,
it's
a
project
that
we
started
in
turin
two
years
ago,
so
here
we
are
so
there
are
some
photos.
So
it's
a
project
born
so
the
university
professor
riso,
with
many
students
today,
a
team
of
five
people,
two
doctoral
students
and
three
engineers
who
work
on
the
project
more
obviously
than
external
contributions.
D
The
lico
project
stems
from
a
reflection
that
I
made
two
years
ago
when
we
found
ourselves
observing
that,
in
fact,
the
whole
infrastructure
was
more
or
less
migrating
to
aq
bernès,
not
only
the
cloud
but
also
the
law,
and
so
the
idea
was
to
be
able
to
build
something
that
would
allow
multiclassing
and
therefore
the
interaction
between
these
clusters.
The
first
question
we
asked
ourselves
was:
why
do
people
and
companies
and
organizations
have
multiple
clusters?
And
there
are
we
did
a
bit
of
surveying.
Some
reasons
appeared
to
us
more
than
others.
D
Acquisitions
or
other
different
departments
can
give
rise
to
different
clusters,
which
then
an
organization
must
maintain,
and
therefore
this
is
a
first
reason,
which
is
absolutely
interesting,
absolutely
important.
The
second
reason
was
the
fact
that
cooper
in
the
tests
represent
open
technology
and
therefore
people
naturally
try
not
to
bind
to
a
single
supplier
but
to
have
more
suppliers
and
therefore
tend
to
make
more
clusters.
D
The
third
reason
was,
more
than
one
reason,
other
geographical.
There
are
legislative
constraints.
There
are
technical
constraints
such
as
latency,
which
impose
that
the
clusters
are
located
in
different
places
and
therefore
another
reason
to
make
different
clusters
to
have
even
more
clusters.
The
last
reason
is
clearly
a
question
of
scalability
when
they
begin
to
fill
my
cluster
with
many
pubs,
it
becomes
more
and
more
difficult
to
maintain
a
certain
point.
I
can't
fill
them
anymore,
so
I'm
going
to
do
another
one.
D
D
against
glamping,
how
to
be
able
to
move
applications
from
one
cluster
to
another
in
the
simplest
and
most
transparent
way
possible
and
from
another.
How
then,
to
make
sure
that
these
applications
that
can
be
on
different
clusters
can
communicate
transparently,
so
these
were
a
little
bit
the
challenges
that
we
found
ourselves
facing
and
having
to
solve
the
ico.
We
started,
as
I
said
at
the
beginning
from
this
one
vision,
absolutely
like
saying
cooper
in
the
tests
everywhere,
where
in
fact,
one
could
also
imagine
that
the
clusters
in
proximity
to
each
other
could
share
resources.
D
D
However,
we
soon
realized
that
what
we
had
developed
could
also
be
used
to
do
real
workload,
orchestration,
which
is
now
a
little
bit
the
focus
of
balance
and
therefore
being
able
to
deal
with
the
speech
of
cards.
There
then
to
place
the
pods
where
these
are
our
user
interests.
To
put,
then
we
clearly
have
the
interconnection
is
last
and
it
is
something
we
are
working
on.
B
A
D
If
we
are
going
to
see
a
where
it
is
located,
we
count
them.
Therefore,
the
projects
that
exist
today
to
solve
the
problem
of
the
other
cluster.
We
have
approaches
that
are
based
more
on
solving
the
first
problem,
I.e.
the
control
plane,
making
sure
that
the
applications
can
therefore
be
scheduled
on
multiple
classes
way
as
transparently
as
possible,
and
we
have
with
a
series
of
projects
that
instead
deal
with
interconnecting
clusters
in
a
transparent,
way,.
D
Riccò's
idea
is
that
of
galaxies
in
the
middle,,
solving
both
problems
so
that
the
cluster
and
therefore
the
bonds
of
are
on
a
other
cluster
are
perceived
as
on
the
south
cluster
gender,
I.e.
there
is
total
transparency
from
the
point
of
view
of
the
use
of
resources
by
applications.
This
is
done
as
we
will
see
in
the
demo,
starting
with
and
from
a
primitive
the
primitive
of
the
ing
which
allows,
let's
say
a
complete
interconnection
between
two
clusters
and
therefore
incorporates
let's
say
both
of
these
problems
so
going
forward.
D
D
Virtual
are
handlers
of
the
way
to
address
the
resources
that
are
on
the
earthquake
class.
They
have
a
size
that
is
decided
by
the
earthquake
trash
and
in
fact
they
are
absolutely
interchangeable
from
the
point
of
view
of
the
governance
control
plane
and
therefore
can
receive
collins
that
pop
from
there
and
then
they
will
be
scheduled
on
the
actual
cluster.
Therefore,
from
the
point
of
view
of
abstractions
and
there,
it
uses
the
virtual
north
as
the
main
abstraction.
Therefore,
a
virtual
node.
A
D
Need
to
use
lico
di
make
a
patch
meeting
take
with
alex,
which
would
have
been
something
very
invasive
and,
above
all,
it
can
be
used
league
with
the
same
piai,
which
are
the
traditional
ones
used
on
Cuban
plasters.
So
our
basic
idea
was
to
avoid
having
to
migrate
to
the
link
for
being
able
to
use
multiplast
and
then
what
is
it
for
I
say
so.
This
is
a
want
to
show
what
the
first
more
traditional
new
skies
chris
can
be
useful
for
and
that
of
the
ed
cloud.
D
D
Perhaps
a
private
cluster
on
which
I
want
to
carry
out
a
maintenance
rose
empty
and
I
am
going
to
do
the
maintenance
on
the
minors
by
temporarily
applying
on
another
plaster
other
you
scales
that
have
been
there
as
a
cap
communicated
by
the
community
that
has
been
created
around
alico.
They
are
more
linked
to
jokes.
D
Let's
say
peer
to
peer,
therefore:
a
provider
of
a
sas
service
who
can
use
lico
to
deliver
services,
not
in
his
plaster
hands
cluster
of
a
customer
and
therefore,
for
example,
can
go
and
create
this
peering,
which
goes
to
put
dei
dei
workloads
in
a
class
earthquake.
This,
bearing
clearly,
is
something
that
does
not
require
administrator
access
to
the
cluster.
It
is
going
to
be
open
connect.
So
the
customer's
point
of
view
the
castro
in
contains
control
over
their
workloads
and
whether
they
have,
for
example,
data
guide
to
access
applications
would
be
accessible.
D
D
Possibility
within
dico
to
create
these
tenants
where
the
pods
are
created.
This
is
provided
to
us
thanks
to
the
cup
sul
technology,
which
uses
them
precisely
as
a
technology
to
segment
workloads
that
are
created
within
a
peeling.
So
there
is
also
an
isolation
speech
which
I
say
obviously
allows
you
to
obtain
the
provider.
Then
it
will
be
able
to
have
more
customers,.
Therefore,
it
will
have
unique
visibility
on
what
are
the
workloads..
D
This
can
be
imagined,
for,
let's
say,
external
visibility
towards
the
customers
of
third
party
entities,
but
it
can
also
be
imagined
within
the
same
organization
and
therefore
making
a
control
plane
available,
which
in
fact
takes
care
of
making
bearing
using
the
ricoh
lexicon
with
all
the
clusters
that
are
in
the
company.
This
peeling
thing
we
have
done
tests
scale
very
well,
so
we
manage
to
have
hundreds
of
connected
clusters
and
therefore,
this
super
cluster
that,
as
we
have
called
him
to
visibility
on
all
the
resources
that
exist
and
he
can
decide
and
make
choices.
D
Let's
say
globally,
optimizing
the
location,
let's
take,
for
example,
the
former
project
of
this
team.
Therefore,
the
application
of
the
directives
in
Europe
using
company
policies,
therefore
he
will
have
to
finish-
or
here
I-
have
the
choices
here
so
that
two
crimes
that
end
up
in
germany,
because
the
cost
of
the
cluster
is
lower.
So
the
super
cluster
can
have
this
type
of
logic
and
choice.
D
Ok,
so
this
is
a
little
bit
the
architectures
that
we
thought
so
with
which
can
be
built
using
them
co.
Let's
take
a
look
at
how
the
peering
we
are
talking
about
works
so
far,
then
the
bearing
is
in
fact
based
on
a
first
step
after
step
of
discovery.
Therefore,
the
clusters
that
exist
and
are
available
must
be
able
to
be
discovered
either
manually
or
automatically
with
technologies
such
as
dns
and
are.
D
At
this
point,
the
real
peering
phase
takes
place,,
I.e.
the
exchange
of
information
that
will
be
used
both
for
resource
management
and
therefore
for
the
creation
of
the
virtual
and
not
for
the
creation
of
the
interconnection
of
network.
Therefore,
when
the
phase
of
peeling
is
completed,
the
network
fabric
is
created.
D
Therefore,
the
interconnection
between
two
clusters
secure,
as
we
will
see
in
the
next
slide,
and
then
we
go
to
manage
the
resources,
therefore
create
the
virtual
node
and
ensure
that
this
node
allows
itself
to
those
resources
to
be
used,
and
at
that
point
they
can
be
used.
All
of
this
is
clearly
managed
by
lico
without
him,
the
user,
the
administrator
becoming
aware
of
it
a
few
words
about
the
network
fabric,
because
clearly
it
is
the
part
of
vico
that
is
below,
and
you
see
a
a
little
less.
D
Therefore,
in
our
opinion,
it
deserves
to
be
seen
a
little
more
than
2
in
detail.
Then
in
fact,
there
with
a
model,
let's
say
extended,
diener
space,
therefore
rain
and
space
between
clusters.
Therefore,
we
imagine
we
are
here
on
a
gelding
here
on
a
cluster
2,
the
nec
space
on
this
cluster.
Then
a
ssx
on
this
cluster
becomes
the
first
or
even
hicks
on
the
second
cluster,
then
a
twin
is
created.
They
are
angioplasty
at
this
point.
Within
these
two
years,
space
pods
are
created.
D
D
This
vpn
that
we
are
going
to
create
is
able
to
pass
the
traffic
and
thanks
to
the
parameters
that
have
been
exchanged
during
the
peering
lico
able
to
create
and
manage
two
nats
that
allow
these
clusters
to
communicate,
even
in
the
case
in
which
the
address
space
overlaps,.
Therefore,
in
fact,
the
two
clusters
that
interconnect
through
the
peering
can
also
have
the
same
address
space.
There
is
no
problem
there,
so
I
will
handle
this
case
to
make
sure
that
the
pods
to
the
services,
the
communication-
can
be
poorer
with
the
services
and
it
works
correctly.
D
The
basic
idea
I
tell
you
is
that
not
to
disturb
the
operation
of
the
siena
and
therefore
everything
that
is
done
at
the
traffic
management
level
is
done
only
for
the
traffic
that
is
destined
for
the
other
clusters,.
Therefore,
the
internal
traffic
remains
under
the
control
of
itself,
splash
of
the
mai
that
manages
it
as
designed
to
make
it
ok,
so
I've
already.
D
A
D
To
the
demo,
which
maybe
is
a
little
more
interesting,
then
we
have
here
two
clusters
that
I
previously
installed
in
these
two
clusters
for
a
matter
of
time
already
installed.
Lico,
let's
go
see
how
a
feeling
works
between
these
two
clusters.
So
I
in
this
first
screen
or
cluster
1
cluster
2
and
here
or
the
terminal
for
commands
for
one,
is
the
terminal
for
commands.
This,
then,
first
of
all
to
do
a
peeling
there.
As
of
the
exposure,
a
command
in
the
his
layman
is
called
so
tl.
D
D
This
pod
I
hope
it
is
readable,
then
a
virtual
cue
black
that
takes
care
of
managing
and
administering
the
peering.
So
it's
one
and
a
version
of
the
virtual
cubelite
modified
by
the
queen
to
work
in
this
context.
This
is
what
he
said
to
the
qube
letta
responsible
for
creating
the
famous
record.
We
didn't
talk
before.
D
We
can
look
and
see
that
there
is
a
virtual
nobili
type.
Milk
club
can
be
recognized
and
if
we
go
to
see
its
size,
we
see
that
it
actually
corresponds
to
a
certain
percentage
of
the
earthquake
klass
approximately
60
70
percent.
So
the
basic
idea
is
that
this
virtual
cannot
have
a
size
that
is
decided
by
cluster
2
and
that,
for
example,
in
a
case
of
software
service,
I
can
decide.
Ok
for
my
supplier
of
do
only
ten
cda
because
of
he
no
longer
needs
them
and
he
will.
D
A
D
A
D
D
They
use
the
image
directly
from
the
bulls
things.
The
only
thing
we
have
changed
is
to
change
the
definitions
and
make
sure
that
the
pods
are
scheduled,
half
and
half
then
half
on
the
virtual
node
and
half
on
the
remaining
nodes.
Let's
say
physicists
at
this
point:
let's
go
flai
this
of
this
manifesto.
D
Here
so
our
applications
are
created
and
let's
see
how
half
of
our
pods
are
not
in
a
maximum
local
on
the
virtual
node.
If
we
go
to
this
other
screen
and
then
go
to
see
on
the
other
cluster,
we
see
that
it
is
actually
in
question
space
that
looks
like
nicodemo,
but
it
is
a
slightly
different
live
space,
so
it
is
the
famous
gianello
expenses
went
to
these.
Other
pods
have
been
created,
so
we
wait.
D
D
D
A
D
D
So
we
go
to
see
the
spirits
slice.
As
you
surely
know,
the
end
points
are
precisely
the
connection
between
the
profs
who
provide
a
service
and
the
service.
So
they
say
that
such
a
prof
can
serve
a
certain
service,
since
some
of
these
bonds
are
on
cluster
1
and
therefore
would
not
be
visible
in
cluster
2,
where
there
are
only
these
pods
licor
takes
care
of
creating
it
then
slices
and
making
the
whole
application
work.
D
D
D
D
D
One,
the
only
difference
we
can
see
here
is
change
the
finite
to
say,
put
a
constraint
that
says
preferentially
use
resources
that
are
not
on
the
virtual
mode,
otherwise
go
where
where
where
succeeds-
and
this
remains
an
object,
deployment
bernet
600
percent
compatible
with
cooper,
nate
vanilla.
So,
let's.
B
D
D
D
When
the
bobs
that
are
on
the
local
cluster
here
we
are
again
on
cluster
1,
so
the
pods
that
are
here
are
restarting
sled
from
cluster
1
and
total.
So
we
see
what
there
it's
kinda
on
the
other
side,
it's
being
terminated
and
we
have
some
brutal
that
will
do
likewise.
In
this
case,
the
remote
pod
is
not
terminated
yet
thing
because
the
local
pod
and
still
isar
on
ok.
D
So
this
allows
us
to
see
that
if
we
now,
let's
go,
do
a
sky
mi,
for
example,
four
replicas
of
our
application-
let's
see
how
they
are
no
longer
there,
so,
let's
scale
all
the
services
of
four
networks.
We
see
that
most
of
the
gods
of
the
pods
of
what
is
going
to
be
scheduled.
We
can
see
it
here,
therefore,
going
to
see
the
pods
in
cluster
1.
Most
of
them
are
local,
only
those
that
can
no
longer
enter
the
local
cluster
will
be
scheduled
on
the
born
cluster.
B
A
D
Sky,
but
if
imagining
it
as
a
cras,
a
public
cluster
and
instead
all
most
of
the
workloads
are
on
the
local
cluster.
So
we
tried
to
conclude
this
demo
to
use
lico
in
a
context
a
little
more.
Let's
say
extreme
here
we
have
created
a
great
disk
because
if
I
don't
have
50
clusters
using
k3
to
build
a
job
that
he
did
precisely
done
within
the
job,
let's
say
on
the
scalability
of
lico
that
on
this
working
on
our
doctoral
student,
marco
is
therefore
we
have
50
clusters.
It
is
a
cluster.
D
D
We
know
that
there
are
50
lico
nodes
that
work
and
if,
let's
use,
for
example,
an
application
that
is
this
50
cluster,
so
we
have
another
nicodemo,
which
we
have
told,
don't
show
local
white
to
go
and
put
itself
on
each
node.
We
note
that
in
fact
we
can
use
the
super
cluster
on
50
different
classes,
so
have
a
replica
of
this
application
on
each
on
each
cluster.
D
To
use
a
service
that
is
based
on
these
three
meters
and
to
see
it
better
on
these
on
these
pops
that
are
anjou
enix
who
respond
with
their
haim
stands.
We
see
that
the
service
always
responds
with
a
different
stream,
precisely
because
this
service
is
built
not
only
on
one
cluster
but
on
all
50.
So
lico
is
also
able
to
manage
a
number
of
clusters,
which
is
not
only
the
single
cluster
but
also
a
greater
number.
D
Ok,
then,
I
conclude
by
saying,
precisely
by
doing
see
a
little
where
ligo
has
arrived,
so
we
met
two
years
ago.
We
now
have
traianos
activated
with
topix,
so
internet
exchange
I
turin,
which
is
testing
lico
and
we
will
use
coins
from
the
production
share
for
computer
science
exams
at
the
polytechnic.
Therefore,
thousands
of
students
should
extend
a
bit
the
capacity
of
the
primary
cluster
that
doesn't
make
it
manage
all
clusters
at
the
same
time
and
we
will
use
dico
to
do
this.
D
D
A
D
B
Alex
really
interesting,
we
have
started
to
bomb
this
day
of
grace,
also
to
be
testimonial
with
the
pci
t-shirt
of
that
time.
I
arrived
really
many
questions,
I'll,
ask
you
some
I,
don't
know
if
we
can
do
them
all.
If
anything,
then
there
is
the
chat
you
can
join
the
chat
on
live
to
continue
the
conversation.
Then
the
first
is
william
rizzo.
He
asks
how
he
behaves
lico
against
starmi
smash,
then.
D
On
this,
we
have
a
work
that
was
carried
out
by
a
thesis
student
on
the,
inter
of
who
we
made
a
prototype
with
the,
inter
of
we
should
publish
an
article
shortly
that
details.
It
is
in
fact,
we
managed
to
extend
lico
su
on
multiple
clusters,
so
the
management
of
the
multi
cluster
becomes
much
easier,
but
we
need
to
find
a
way
to
then
go
to
counterbalance
the
fact
of
having,
for
example,
only
one
control
plane
of
the
inter.
D
A
D
Then
we
use
we
started
the
project
relying
heavily
on
the
blackest
beat
so
using
the
project
and
forking.
We
currently
have
changed
some
approaches
a
little
bit
because
they
allow
us
to
scale
a
little
better
than
sportube
the
town,
let's
say
a
way
to
interconnect
to
a
source.
So
it's
a
mechanism
where,
in
fact,
at
one
and
the
floors,
let's
say
more
black
towards
the
cluster
and
then
behind
an
interface
that
can
be
implemented
to
communicate
with
any
type
of
service.
B
D
B
D
Also
on
this
we
have
prepared,
we
are
preparing
a
scientific
paper
on
which
marco
iorio
is
working.
The
PhD
student,
who
clearly
worked
the
most
on
lee
co's
scalability.
Our
results
are
that
it
has
no
more
impact
than
having
a
cluster
plus.
No,
he
actually
says
about
virtual
north.
They
are
like
grandparents,
15
c.
B
B
B
D
Have
measures
on
this
too?
The
basic
answer
is
no
our
measures,
I.e.
there
is
not
much
traffic
that
that
it
comes
that,
let's
say
the
control
traffic
is
generated
which
would
pass
the
return
of
the
bpm
and
crisis
is
visible
with
respect
to
the
data
traffic
of
the
applications.
That
yes,
clearly
depends
on
the
application.
However,
we
have
not
voted
and
we
will
show
it
in
the
paper.
Therefore,
with
the
measures
mounted
a
very
high
consumption,
also
with
a
very
large
number
of
objects,.
D
B
B
Terms
then,
if
anything,,
if
you
want
to
integrate
in
chat
during
the
day,,
there
is
a
problem,
then
we
will
also
be
able
to
perhaps
share
some
additional
material
with
the
public.
However,
we
will
put
the
links
of
the
speeches
on
youtube
and
then
maybe
lips
to
put
something
more
if
needed
willingly.
These
are
two
questions.
I
ask
them
in
one:
how
permissions
are
managed
for
scaling
workloads
on
class,
earthquakes
also
works
with
stateful
workloads.
B
D
The
second
easiest
for
now,
no
so
a
version
that
we
don't
have
now,
but
it's
the
next
feature,
so
it
should
be
available
starting
from
december.
We
have
this
trials
in
january,
which
requires
us
to
have
the
feature
big
stacks
on
set.
So
we
think
recently
in
the
master
version
of
the
market.
The
pr
project
is
still
missing
something.
But
if
you
want
to
do
a
test
willing,
you
can
try
with
clearly.
D
In
fact,
the
presence
of
that
those
rules
that
we
put
a
certain
point
on
myspace
is
a
way
to
warn
vico's
web
book.
That
says,
ok
I,
put
such
a
reason
that
causes
that
to
end
up
on
lico's
way.
So
from
that
point
of
view,
we
try
to
protect
workloads
that
are
a
sleeve
that
don't
want
to
benefit
from
vico,
but
that
could
end
up.
Who
knows
where
that
there
is
this
node
that
has
so
many
resources.
D
It
will
be
scheduled
if
there
are
resources
from
the
point
of
view
of
the
shakes,
and
there
is
therefore
usually
it
will
be
subjected
to
all
the
constraints
of
of
that
namespace
us
on
that
there
we
support
cap
sul
to
ensure
that
all
the
spaces
created
in
a
tenant
have
the
right
rules,
for
example,
to
prevent
cross
earthquakes
in
saturated.
So
with
too
many
goals.
I,
don't
know
if
this
can
answer
the
question
or.
B
D
This
is
a
great
question
and
we
did
n't
see
the
installation
we
have
mechanisms,
let's
play
how
we
still
did
this
vip
fight
thing,
the
nodes
physicists
for
now
we
have
concentrated
on
the
provider.
So
if
anything,
for
example,,
we
create
his
k
s
ii
in
clusters,
so
that
the
install
able
to
attach
to
that
that
node,
that
will
be
created
all
the
rules
needed
to
say
that
he
is
in
europe
on
the
tablets
where
he
had
to.
D
However,
wherever
the
thing
is
to
be
able
to
attack
them,
I
read
physical
nodes
and
the
complexity
is
clearly
interesting
and
that
some
balls,,
let's
say,,
have
a
value
if
they
are
on
all
the
physical
nodes,,
some
others
don't,.
So
we
have
to
figure
out
for
a
moment
whether
to
put
it
in
case
there
is
above
all
if
there
is
only
one,
it
is
a
little
bit
complicated
for
now
it
hasn't
been
requested.
So
we
didn't
do
it.