►
Description
VM Import Operator (VMIO) allows Kubernetes administrators to easily import their oVirt- and VMware- managed virtual machines to KubeVirt.
Konveyor's Forklift is a project that leverages VMIO to propose a user interface for large scale migrations, introducing the concept of migration plan and implementing inventory and validation services.
In this talk, the speakers will explain the design of the Virtual Machine Import Operator and how it can be used to import virtual machines to KubeVirt. Afterwards the speakers will show how Forklift uses VMIO to deliver better user experience while importing virtual machines to KubeVirt.
The attendees will:
- learn how VM Import Operator works
- see an oVirt virtual machine import to Kubernetes (KubeVirt) with VMIO
- see VMware virtual machines import to Kubernetes (KubeVirt) with Forklift
- know how to import their VMware or oVirt workloads to Kubernetes with VMIO or Forklift
Presenters:
- Jakub Dzon, Senior Software Engineer, Red Hat
- Fabien Dupont, Senior Principal Engineer & Engineering Manager, Red Hat
References:
- VM Import Operator (VMIO): https://github.com/kubevirt/vm-import-operator
- Konveyor's Forklift: https://konveyor.io/forklift-documentation
B
B
B
Okay
and
the
agenda
for
the
presentation
is
as
follows:
with
others,
I
will
introduce
the
beautiful
machine
import
operator
and
afterwards
I
will
demo
an
actual
virtual
machine
import
from
audio
to
kubert.
I
will
explain
what
the
virtual
machine
import
operator
is
and
how
it
can
be
used.
The
virtual
machine
import
operator
part
will
end
with
a
demo
of
an
alternative,
a
manual
way
of
transferring
your
audio
vm
to
keyboard.
B
The
virtual
machine
operator
is
a
kubernetes
operator
that
is
able
to
import
compatible,
kvm
and
vmware,
based
virtual
machines
into
that
cluster,
where
they
would
be
run
by
qb,
saying
compatible
is
important,
not
all
options
that
are
available
in
vanilla,
kvm
hypervisor
are
available
in
compute.
The
operator
takes
care
of
checking
the
source,
vm
configuration
and
reports
its
findings
to
the
user.
B
B
B
B
B
B
B
B
B
B
B
B
B
B
The
last
system
that
takes
part
in
the
import
process
is
ovird
that
manages
the
imported
machine.
Vm
import
operator
reads
metadata
the
imported
vm
requests,
shutdown
for
the
import
process
and
startup
in
case
of
an
import
failure
cdi
in
turn
connects
to
auger
to
download
the
disk
images
and
stores
them
in
kubernetes
storage.
B
After
the
overview
of
the
infrastructural
part
of
the
operator,
let's
have
a
look
at
the
domain
side
of
it.
The
virtual
machine
import
is
a
namespace
custom
resource
that
defines
the
source
of
the
vm,
the
identifier
of
the
vm
on
the
source
and
the
mapping
to
be
used
for
the
import.
The
mapping
of
resources,
network
and
storage
from
the
external
vm
provider
to
kubwear
is
defined
in
the
resource
mapping.
Custom
resource
kubernetes
secret
is
needed
to
define
the
endpoint
and
credentials
to
the
source
provider.
In
this
case,
let's
go
over
the
vm
import
flow.
B
Next,
the
virtual
machine
import
controller
retrieves,
the
configuration
of
the
source
vm
from
the
source
provider
validates
the
cr
and
the
vm
and
stops
the
vm
on
the
source
provider
in
the
next
step.
The
virtual
machine
import
controller
creates
a
disk
class
virtual
machine
afterwards,
the
virtual
machine
import
controller
creates
a
data
volume
cr
for
each
vm
disk
in
the
source.
Vm.
B
B
B
B
And
if
we
have
a
virtual
machine
with
many
disks,
we
need
to
create
that
many
data
volumes
for
manually
to
import
those
those
contents,
we've
created
the
virtual
machine
and
it's
running.
It's
got
an
instance.
It's
connected
to
a
network
and
again
we
can
connect
it
with
gnc
and
confirm
that
it's
exactly
the
same
virtual
machine
as
previously
and
as
in
the
object
cluster.
C
C
The
idea
behind
that
is.
It
should
be
easy
to
do
a
migration,
since
we
know
exactly
the
technical
aspects,
so
we
are
creating
many
tools
and
one
of
those
tools
is
forklift.
Its
goal
is
to
pro
to
help
do
mass
migration
of
vms,
so
not
just
one.
At
a
time
and
in
terms
you
see
some
concepts
as
the
plan,
the
mappings,
so
the
mappings
were
already
mentioned
by
yaku,
but
this
time
we
are
doing
large
mapping
of
all
possible
networks
and
storages.
C
So
the
idea
is
we
leverage
that
we
do
it
once
and
for
all,
and
then
we
do
migration
and
we
also
have
migration
analytics.
I
will
come
back
to
that
aspect
a
bit
later,
so
conveyor
forklift
is
actually,
as
you
can
see,
a
bunch
of
new
custom
resources.
So
we've
got
the
network
mapping,
storage
mapping,
which
are
basically
telling
if,
if
my
vm
is
connected
to
this
network
and
the
destination
should
be
connected
to
that
other
network,
so
nick
bionic,
the
conversion
will
be
done.
C
The
we
also
have
the
notion
of
provider,
we
have
a
source
provider
mainly
and
also
the
target,
but
the
target
by
default
is
created
when
you
install
the
operator
and
it's
basically
the
credentials
and
the
url
clinic
to
the
to
the
provider.
And
of
course,
we've
got
the
list
of
virtual
machines.
We
want
to
migrate.
So
from
all
those
components
we
create
a
plan.
C
So
the
plan
is
basically
the
aggregations
of
well
source
and
destination
providers.
The
mappings
we're
seeing
where
everything
is
going,
and
one
and
two
and
the
tuition
machines,
so
the
machines
we
want
to
migrate
together
with
the
same
parameters
for
migration.
So,
for
example,
when
you
create
a
plan
you
can
actually
create.
When
you
create
a
second
plan,
you
can
have
different
mappings
as
the
first
one.
It's
really
up
to
you,
you
can
reuse
existing
mappings
or
creating
one
on
the
flight.
C
Then,
when
we
are
ready,
we
create
a
migration.
The
migration
is
the
instantiation
of
the
plan
for
execution,
so
it
will
have
all
the
status
of
the
different
vms
migrating
and
if
some
of
them
are
failing
well,
the
vm
is
act.
You
can
actually
restart
the
plan
by
creating
another
migration,
and
only
the
vms
that
have
not
been
migrating
successfully
will
be
migrated
again.
C
So
did
this
provide
an
easy
way
for
retrieves
and
from
there
we're
leveraging
vmio
each
vm,
for
so
for
each
vm
we
are
going
to
create
a
vm
import,
cr
based
on
the
mapping
and
the
informations.
We
have
on
the
source
vm.
So
from
there
we
delegate
the
work
to
vmware,
it's
actually
a
layer
of
orchestration
we're
trying
to
mainly
focus
on
orchestration.
C
The
other
part
I
mentioned
is
analytics
so
as
jacob
mentioned
in
in
one
of
his
first
slides.
Not
all
vms
are
fit
for
converts
now,
so
we
know
that
some
migration
will
fail
and
we
know
it
right
away.
We
don't
have
to
wait
for
the
micromachine
to
fail.
We
know
it
from
the
parameters
of
the
vm,
so
the
first,
our
first
paradigm
is
one
don't
block.
C
The
idea
is
to
provide
insights
to
the
users
and
they
make
the
decision
to
run
the
migration
anyway,
so
it
may
fail,
but
we
want
them,
and
the
second
idea
is
fail
fast.
So
sometimes
we
know
that
there
is
absolutely
no
way
for
the
vm
to
migrate.
So,
even
if
the
user
has
clicked
on
the
migrate
button,
we
can
still
fail
really
fast
in
the
validation
stage.
C
C
So
we
want
to
be
able
to
analyze
many
components
of
the
vm
of
the
of
the
vm,
including
its
infra,
its
infrastructure,
so
that
what
we
also
want
to
do
is
to
use
the
concerns.
So
what
we
call
concerns
is
a
result
of
the
analysis
in
from
in
many
places.
The
first
one
is
to
explore
that
in
the
creation
in
the
plant
creation
wizard
so
graphically,
you've
got
a
ui
for
conveyor
and
in
the
ui
you
will
have
the
concerns
for
each
vm
of
your
plan.
C
You
can
expand
the
list
you
we
can
show.
It
show
you
that
in
the
demo
later-
and
you
will
know
exactly
why
we
are
warning
you,
the
second
one
is,
the
validation
is
reflected
in
the
planned
condition.
So,
if
you're
familiar
with
custom
resources,
when
you
reconcile,
you
can
set
condition
on
the
vm
and
actually
when
we,
when
we
in
the
plan
controller,
what
we
do
is
we
also
ask
the
validation
service
for
the
concerns
of
all
the
vms
of
the
plan,
and
we
will
reflect
that.
C
So
if
something
is
red
or
or
will
fail,
absolutely
will
reflect
that
in
the
planned
condition,
just
just
a
way
for
the
user
to
have
a
warning.
Also
after
the
creation
of
the
plan-
and
we
also
do
pre-migration
analysis
at
scale.
One
one
difference
with
vmio
is
that
we
have
an
inventory
component.
That
is
when
you
create
a
provider,
we
create
an
inventory
pod
and
that
inventory
will
actually
ask
the
validation
service
for
the
concerns
before
you
even
create
a
plan.
C
C
It
may
become.
It
may
be
considered
as
a
value
and
machine
web.
Sorry,
if
you
want
to
know
more,
we
have
a
repo
where
we
store
all
our
rules.
If
you
want
to
contribute
to
the
rules
also
feel
free,
it's
working
great
working,
a
part
inside
our
operator
and
now
let
me
do
a
quick
demo
and
see
well,
let
me
pause
slow
it
down.
C
C
You
can
see
that
you
have
an
interface
to
create
a
provider.
Here
is
a
vmware
provider.
I
enter
all
the
credentials
and
now
that
my
vmware
provider
has
been
created,
I
I
have
a
tab
with
all
the
providers
and
you
can
see.
My
inventory
is
already
running,
giving
me
information
about
the
vms,
the
host
and
everything
I
can
switch
to
creating
a
migration
plan.
C
My
source
provider
is
the
vmware
provider
I
created
and
I'm
targeting
a
cluster
name
host.
It
is
the
the
cluster
where
the
it
is,
the
kubernetes
cluster,
where
my
operator
is
running
so
directly
because
we
are
already
connected
to
it.
We
can
get
access
to
the
cluster.
Here
is
a
way
to
select
vm.
I
have
many
multiple
ways:
folders.
We
also
have
a
view
with
a
host
and
clusters,
which
is
interesting
here.
C
I
just
want
to
show
a
few
vms
and
you
can
see
they
all
have
warnings
with
a
validation
from
the
list
of
vms,
I'm
creating
a
new
mapping.
You
only
the
networks
used
by
the
vmi
selected
are
shown
here
and
that's
the
same
for
the
net
storage.
So
we
are
not
asking
you
to
map
all
the
storages
and
networks
at
once.
You
can
also
do
it
plan
by
plan
and
from
there
get
the
information
about
everything,
a
quick
review
and
we
have
our
plan.
C
C
C
C
C
No,
don't
we
start
so
we
have
a
long
roadmap
for
conveyor.
I'm
using
this
quick
summary.
One
of
our
main
features
is
the
one
migration
where
we're
going
to
migrate.
The
data
before
shutting
down
the
vm,
it's
based
on
change,
block
tracking
people
used
to
backup
would
be
familiar
with
that
and
it
will
allow
us
to
reduce
the
downtime
of
the
migration
to
two
three
maximum
five
minutes.
Well,
I
say
maximum.
It
really
depends
on
how
much
storage
you
have
jakub.
Do
you
want
to
to
add
another
word
about
it?.
B
C
A
So
there
are
no
questions
here
in
the
chat,
but
there
were
a
couple
of
questions
in
in
so
oh
wait.
I
see,
I
know
so,
there's
a
thread
by
the
way
in
slack
where
and
hash
virtualization,
where
we
can
follow
up
on
on
this
session
or
any
other.
So
sashim
there
was
asking
if
there
any
experience
or
best
practices
on
importing
a
windows,
machine
and
common
is,
I
remember,
trying
the
import
a
while
ago,
especially
for
the
windows
vm,
and
it
took
longer
than
expected
so
basically
looking
for
best
practices
on
that.
C
C
C
A
C
Yes,
so
con
conveyor
is
the
community
we
have
and
forklift
is
the
the
tool.
Forklift
is
actually
leveraging
vmios,
so
the
tool
that
jacob
explained
the
operator
that
the
the
the
heavy
lifting
it's
vmio
and
cdi.
I
also
want
to
mention
that,
because
yeah
talked
about
it
with
data
volumes,
the
heavy
lifting
is
done
there
and
because
it's
working
great,
we
are
reusing
it
in
in
forklift.
So
forklift
is
mainly
an
orchestration
and
analyze.
This.