►
From YouTube: Success of CRI: Bringing Hypervisor Based Container to Kubernetes [I] - Lei Zhang, HyperHQ
Description
Success of CRI: Bringing Hypervisor Based Container to Kubernetes [I] - Lei Zhang, HyperHQ
CRI, aka Container Runtime Interface, is the new mechanism designed to plug any container runtime in Kubernetes. In this presentation, I will introduce how we introduce the hypervisor based container into Kubernetes as native container runtime by using CRI. And enables users to serve their customers directly with virtualized containers, instead of wrapping them inside of full blown VMs. CRI design principles and implementation details will be explained. As well as the essential differences between Kubernetes and other projects like SwarmKit and how hyper.sh made the choice. Today, many developers are not comfortable with Linux containers as an effective boundary, and requires for a stronger degree of isolation, particularly for those running in a multi-tenant environment. We believe HyperContainer with Kubernetes is one of the best answers.
About Lei Zhang
Phd candidate, and also a feature maintainer of Kubernetes project. I once worked for Cloud Foundry team in VMware and Baidu. Now as HyperCrew, the author team of world's leading open-source hypervisor based container. I mainly focus on Kubernetes upstream about scheduler and CRI stuff, responsible for leveraging HyperContainer as runtime for Kubernetes , which is also known as Hypernetes project. As a active community member, I was speaker of KubeCon 2016, LinuxCon 2016, and InfoQ Conference. As the Tech Lead of SEL lab of Zhejiang University, I have proposed my lab as the academic member of CNCF in 2015.
Join us for KubeCon + CloudNativeCon in Barcelona May 20 - 23, Shanghai June 24 - 26, and San Diego November 18 - 21! Learn more at https://kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy and all of the other CNCF-hosted projects.
A
So
I
start
from
a
survey
when
I
were
reading
PhD
so
and
it's
about
the
container
boundary
and
the
question
is
quite
simple:
are
you
comfortable
with
Linux
container
boundary?
So
there
are
two
kinds
of
answer:
actually
half
of
them
is
like
and
the
use
containers
in
there
provide
or
safe
environment.
So
another
half
of
the
answer
is
no
because
they
are
using
container,
for
example,
trying
to
reserve
in
a
public
cloud
or
in
a
multi-tenant
environment.
A
So
as
long
as
we
care
about
these
kind
of
security
so
lot
of
people,
they
have
to
wrap
their
containers
being
stifled
local
machines.
This
is
actually
what
most
of
the
container
cloud
provider
is
doing
today.
But
in
that
case
we
will
actually
lose
the
circle
clone
native
deployment.
For
example,
we
have
to
tolerate
into
the
slow
start
time
of
the
virtual
machines
right,
and
we
also
have
to
tolerate
the
huge
resource
wasting
caused
by
your
virtual
machine
cluster
in
your
data
center
and,
on
the
other
hand,
every
container
in
your
class.
A
They
actually
have
to
pay
a
extra
memory
tax
because
there's
extra
hot
water
layer
running
in
your
data
center.
So
this
is
problem
and
that's
why
we
have
hyper
container
project
because
we
are
trying
to
first
being
secure
and
keep
the
cloud
native
features.
So,
let's
begin
to
revisit
the
container
of
example
for
docker,
and
we
can
actually
divide
the
container
into
two
parts.
A
The
first
part
I've
enjoyed
the
contra
runtime
part
is
actually
the
dynamic
view
and
the
boundary
we're
running
four
stages,
for
example
in
docker,
it
is
implemented
by
a
group
of
namespaces
and
restricted
by
using
state
groups.
So
in
this
example,
your
your
education
per
state-
we
should
echo
hello-
is
running
inside
a
standard
box
which
is
isolated
by
namespaces
and
the
rest
part
of
the
container
which
we
can
describe
as
container
image
is
actually
the
static
view
of
your
program
of
data
or
your
dependencies
files
and
directories.
A
Anything
you
need
to
run
application
actually,
so
in
darker,
it
is,
you
will
have
a
first
of
all
the
redraw
layer
and
then
an
inter
layer
which
studied
so
that
you
can
modify
the
contents
in
your
hosts
files
and
then
a
group
of
read-only
layers
which
are
uni
multi
together
to
form
the
so
called
talk
image.
So
that
is
a
container,
so
this
picture
actually
murdered.
You
might
ask
what
happens
if
we
replace
the
container
runtime
part
to
hypervisor
best.
So
that's
the
five
container.
The
hyper
container
is
actually
composed
by
three
small
projects.
A
The
first
one
is
the
OCI
compatible
hypervisor
implementation,
which
named
in
the
wrong
way,
and
another
project
is
a
hyper
demon
in
his
own
control
daemon,
and
we
also
have
the
very
important
project,
which
name
is
hyper
start,
and
it
is
the
init
service
inside
this
small
lightweight.
He
works
machine
and
the
reason
we
need
is
that
we
cannot
start
a
virtual
machine
just
with
your
application
process
at
the
first
process.
Right.
A
So
that's
why
we
have
hyper
start
and
we
need
to
deal
with
things
like,
for
example,
the
execute
command
the
attach
port
of
all
these
things
like
that
is
a
streaming
API.
We
need
to
have
the
engine
inside
and
another
example-
is
that
hyper
start
will
be
responsible
to
manager
the
user
containers
being
started.
Is
this
type
container?
Is
post
sandbox?
So
that's
why
this
project
is
very
popular
and
we
also
reached
agreement
with
Intel
clear
container.
A
They
are
also
using
over
and
this
small
project
underneath
other
unity
service
and
for
the
docker
image
of
pearl.
Harbor
very
follows
the
standard
darker
image
implementation
and
in
a
future,
where
soon
we
go.
We
also
posted
our
image.
We
had
a
maintainer
OCI,
so
yeah
that's
hot
container
and
you
can
see.
Actually
the
idea
is
about
how
to
combine
in
the
how
to
combine
the
best
parts
of
work,
machine
and
container.
So,
first
of
all,
it
will
be
portable
and
behaves
like
earliness
containers
that
you
can
see.
A
Hasta,
for
example,
start
application,
a
union
hyper,
sticky
or
wrong
blah
blah
blah
and
that
container
actually
works
machine
will
be
started.
Reading
second
and
it
only
cost
Elmo
MB
memory
extra
for
hacker
at
the
park
and
on
the
other
hand,
it
is
actually
using
a
hardware
level
virtualization
this
independent
kernel.
So
if
you
execute
a
command
inside
that
container,
you
will
see
there's
independent.
It
gets
the
kind
of
running
totally
different
from
your
host
machine.
So
that's
hide
container
and
other
information
is
that
hyper
container
actually
naturally
match
to
designer
pod.
A
Just
check
this
example.
Normally
when,
for
example,
when
you
are
you
needing
darker,
so
there's
no
poet
anymore.
Actually,
pod
is
a
largely
concept
being
any
kind
of
container
runtime
implementation,
but
in
hyper
it's
not
that
case
you
will
have
a
so-called
post
and
box
running
a
physical
stand,
bugs
running,
which
is
implemented
by
light
rated
virtual
machine,
and
your
applications
will
be
running
inside
these
sandbox
like
a
normal
containers.
A
A
but
please
note
we
don't
have
talking
one
inside
with
just
use
namespaces
to
set
up
these
are
containers
inside,
so
you
can
see
here,
lap
hold
and
that's
why
they
began
to
thinking
about
bringing
the
hype
container
to
kubernetes
from
the
very
beginning.
But
at
that
time
we
do
not
have
control
on
time
interface,
so
we
have
to
use
a
wallet
hell:
internal
interface,
/
crew
Brad,
which
is
exactly
what
Rockhead
is
worth
using
in
the
past
few
years.
A
So
in
that
case
we
have
to
add
a
hyper
package
instead
inside
of
the
red
code
base,
and
we
have
to
implement
everything,
the
cupola
need
to
run
container
to
run
polled
everything
else,
including
lifecycle
and
opposed
lifecycle.
The
containers,
the
network
plug
in
the
wall
you
plug
in
mount
volumes.
Everything
we
need
to
do
inside
of
this
small
package,
so
you
can
see
it's
very
complicated
and
on
the
right
hand,
there
are
some
limitations
of
a
hypervisor.
For
example.
We
cannot
support
a
host
network
right,
it's
totally
different
thing.
A
So
that's
why
we
cannot
merge
this
hyper
package
into
an
upstream
kubernetes
that
we
have
to
maintain
our
own
branch
and
every
time
we
need
to
do
the
reverse.
It
will
become
an
animator
EE
like
that.
So
that's
why,
since
2015
I
think
we
began
to
propose
a
so-called
chrysler
mode
runtime
to
the
crib
native
upstream,
and
if,
thanks
to
the
work
by
Brennen
worse,
we
finally
reach
our
goal.
A
That
is
that's
that's
why
we
have
the
console
called
container
runtime
interface
and
since
that
time
we
don't
need
to
have
to
focus
anymore,
so
the
hyper
Nettie's
project
we
had
before
will
finally
transfer
to
a
repo
which
only
contains
unnecessary
plugins
like
network
plugins,
volley,
plugins
or
end
these
third-party
resources.
That's
all-
and
we
now
have
our
own
implementation
of
a
country
welcome
interface,
which
is
frankly,
project.
I
will
talk
about
that
later.
A
So,
let's
begin
from
a
controller
one
hand,
so
the
CRI
is
actually
a
interface,
a
group
of
interface
to
describe
what
crew
blade
expect
from
a
container
runtime.
For
example,
I
am
crew
blade,
and
maybe
you
are
talker,
so
I
can
say:
hi
docker,
please
prepare
API,
which
name
is
start
container.
So
I
can
pause
the
video
group
up
configurations
and
you
use
that
to
start
a
container.
So
that's
it,
but
as
a
dollar
you
cannot
say
to
me:
hey
Kublai's
I
have
API,
which
name
is
swarm
so
much.
A
We
do
like
that
at
the
cupola,
I
can
say
no
I
don't
need
that.
So
that
is
the
responsibility
of
can
I
random
interface.
It
will
help
you
to
narrow
down
what
is
needed
to
implement
a
controller
run
time
for
kubernetes
hopeful
kubernetes
project,
and
the
second
thing
is
that
a
CRI
is
imperative
container
stage
week
interface.
So
maybe
you
are
curious.
Why
not
polled?
Well?
There
are
two
kinds
of
problems
here.
So
first
one
is
in
that
case
every
continent
run
time.
A
Interval
in
implementation
have
to
know
how
to
understand
the
concept
of
a
bode
well
for
rocket.
Maybe
the
works
before
docker
is
impossible
so
that
we
all
live
on
a
burden
to
implement
everything,
but
a
pod
in
the
ship
in
a
random
machine
for
the
ramp
in
Tokushima.
You
have
to
care
about
that.
That's
not
a
very
good,
and
the
second
is
being
is
in
that
case
you
have
to
change
the
country
random
interface.
A
Every
time
when
there
is
a,
for
example,
a
new
feature
area
to
poor
level,
for
example,
the
pot
level
we
added
a
new
field
through
the
port
hole
specification
there.
Maybe
we
have
to
change
the
country,
random
interface,
and
we
all
know
that
that
specification
actually
changes
a
lot
on
companies
upstream.
So
that's
also
not
good.
That's
why
we
use
concern
level
interface
in
CRM,
so
the
benefits
are
very
straightforward.
A
There
is
extensibility
and
of
course
we
can
have
feature
velocity
because
now,
for
example,
developers
new
hyper
in
rocket,
they
can
focus
on
their
own
implementation.
Do
not
care
about
too
much
things
about
kubernetes
and
crew.
Blade
and
most
important
thing
is
that
there
will
be
code
maintainability.
So
we
don't
have
to
use
that
wallet.
Hell
internal
interface,
anymore
and
I
need
to
say
that
color
is
being
really
tortured.
Everybody
in
the
stick
note
for
a
long
time.
So
we
don't
need
that
anymore
for
more
detail.
A
You
can
check
this
poor
request
by
Fisker,
my
colleague,
my
colleague,
and
all
of
you
into
an
implementation
about
GRE
there.
So
the
color
run.
Amin
specification
is
actually
quite
simple,
just
a
group
of
fractions,
so
you
can
see
there
is
a
management
for
pod
sandbox
for
the
wall
and
for
container,
and
we
have
streaming
API
and
for
image.
It's
quite
quite
simple,
so
you
may
be
curious.
What
is
dynamics
a
little
pod?
No,
it's
not
hold.
A
The
standard
box
actually
describes
how
you
can
implement
isolated
you,
how
you
can
isolate
public
cloud
environment
for
the
dumping
darker.
It
should
be
a
infer
container
with
a
group
of
so-called
poor
label.
C
groups,
that's
all
and
in
hyper.
However,
it
will
be
a
lightweight
huge
machine.
So
it's
about
how
to
implement
that
title:
isolation,
layer,
it's
not
about
the
boat
and
for
the
container
part.
However,
there's
nothing's
nothing
special
for
daugher.
A
It
can
be
directly
mapped
into
the
dock
container
and
for
Hyper
they
will
be
mapped
into
a
group
of
namespace
containers
controlled
by
the
hyper
start,
atom
automation
before
so
now
we
have
come
to
a
random
interface.
We
have
a
group
of
the
a
group
of
a
group
of
definitions
about
how
are
you
can
do
so?
How
can
the
sea
I
work
with
head
container?
So
if
question
comes
just
implement
it?
That's
why
we
had
a
friendly
project,
which
is
now
official,
Cobra
his
project
and
helping
released
with
kubernetes
one
plastics.
A
It
has
already
passed
an
artistic
percent
of
the
node
end-to-end
conformance
test
and
I
mean
in
the
next
few
days.
We
will
reach
the
goal
of
100
percent
and
we
are
using
the
sceneline
network
here.
Everything
is
exactly
the
same
and
you
are
what
you
are
using
today
and
we
also
supported
the
poll
levels
of
management
and
we
we
also
support
a
mix.
The
runtime
only
statement
note,
so
you
can
run
had
canary
darker
and,
on
the
same
note
freely,
and
of
course
it
we
can
deploy,
we
can't
deploy.
A
Frank
TV
criminal
is
by
using
could
be
demeaning
and
in
the
future,
we've
also
have
unique
kernel
support,
which
is
now
official
topic
for
the
google
Summer
of
Code
in
evening
news
in
this
year.
So
please
stay
tuned.
So,
first
of
all
how
Franklin
works,
it
is
actually
quite
simple.
We
can
see
kubernetes
at
two
parts,
the
first
part
of
the
so
called
controller
controller
panel
control
panel,
the
management
node.
A
So
this
is
where
the
scheduling
and
the
workload
application
actually
happens,
and
this
is
also
where
the
concept
report
is
actually
played
in
the
most
important
role.
But,
however,
we
also
have
note
where
the
cupola
T
running
so
in
that
place,
pod
will
be
transferred
and,
for
example,
the
couplet
will
watch
the
change
of
the
pole
list
in
SIDS,
or
it
will
notice
what
happened
program.
If
the
politic
really
deserve
the
crew
blade
men
cycle
loop
will
deliver
is
supposed
to
circle.
A
The
generic
runtime
and
Janica
run
ham
is
responsible
to
generate
the
configurations
of
port
and
containers,
and
then
it
will
call
the
container
random
interface
functions
to
use
that
parameters
to
startups
post
and
box
and
container.
So
from
now
on
from
that
formula
place
from
that
lie,
there
were
no
code
anymore,
there's
no
problem,
only
Stanton
box
and
containers.
So
please
note
here
and
after
you
call
the
Condor
random
interface,
the
implementation
will
be
responsible
for
doing
the
rest
of
things.
For
example,
you
darker
there
will
be
a
dr.,
Shimron
e.
A
It
will
translate
to
the
CRI
into
the
darker
API,
so
it's
like
a
broker,
so
that's
how
we
implement
frankly
it
just
like
another
broker
which
irresponsible
to
chattel
a
the
CI
into
the
hyper
demon.
Api.
So
quite
simple.
Well
now
is
a
is
a
key
point:
how
to
implement
this
kind
of
a
runtime
seem
like
takashima,
Frankie,
cryo,
rocky
low
rock
relays,
everything
okay,
the
most
important
thing
is
lifecycle
management,
for
example.
A
If
you
are
using
a
boat
it
creating
a
Polish
name
in
the
floor,
and
it
is
two
containers
inside
so
the
workflow
is
like
first
of
all,
ROM
closed
and
boxed
and
then
created
container
a
start
container
a
and
then
create
container
B,
and
it's
not
gonna
be
so.
This
kind
of
workload
workflow,
however,
however,
in
the
docker
run
time,
will
be
likely
to
run
post
any
bugs
in
darker.
It
actually
mapping
to
start
a
special
container,
which
name
in
the
infra
container.
A
This
container
will
be
used
to
hold
a
network
namespace
of
the
whole
code
and
then
Eva
called
docker
to
start
to
container
for
you
and
then
join
those
two
containers
to
the
network
name
space
of
info
container.
So
that's
how
it
works.
So
you
can
see
here.
You
will
actually
start
three
containers
for
this
pod
right.
That,
for
Hyper
is
not
that
case
in
hyper.
You
both
first
of
all
the
wrong
wrong
wrong
host
and
box
will
create
a
lightweight.
A
He
works
machine
for
you,
and
then
you
will
create
a
starter
container
inside
and
start
content
being
side
so
down.
There
were
no
such
kind
of
unified
container
running
inside
and
you
will
have
a
lot
of
IP
which
machine
with
two
containers.
So
that's
why
I
mentioned
before
I'm
hyper
container
is
actually
a
better
influence,
implementation
for
the
concept
of
a
boat
and
the
for
container
part
lifecycle.
However,
there's
no
difference
between
darker
and
hyper.
A
So,
for
example,
if
you
call
the
correct
container
and
the
container
state
will
be
transferred
to
now
to
credit,
if
you
call
stock
container,
it
will
be
transferred
to
running
and
so
on.
So
this
is
first
first
part
and
the
second
part
is
streaming
so
trimming
is
I,
think
the
most
complicated
part
for
any
kind
of
container
runtime
and
in
older
version,
when
we
have
no
control
interface
implemented.
Things
like
this.
A
There
are
some
problems,
so
first
of
all,
crew
plate
will
actually
become
the
bottleneck
and
on
the
other
hand
you
can
also
see
at
sea
here
the
run
time
machine
is
also
standing
in
the
cradle
critical
path
of
this
AP
AP
ICO.
So
this
actually
brought
us
very
severe
performance
problem
and
there
are
a
lot
of
streaming
API
running
at
the
same
time
and
on
the
other
hand,
they
are
a
lot
of
code
application
among
the
runtime
machines,
for
example,
dark
and
rocket.
A
They
are
almost
handling
this
kind
of
streaming
API
in
the
same
way,
but
they
have
to
implement
everything
in
there
on
their
own
side.
No
no
code
share
live
here.
So
that's
why
in
encounter
encounter
one
interface
version,
we
change
the
design
of
the
streaming.
The
basic
idea
is
quite
simple
that
we
will
start
a
separate
streaming
server
running
instead
of
handling
everything
by
corridor
of
the
runtime
machine.
So
there's
no
business
for
kool-aid
anymore.
This
store
will
be
responsible
for
handling
every
stream,
assuming
request
and
let's
see
how
it
works.
A
So
again,
the
API
server
will
send
a
sender
connection
to
crew
blade
and,
and
this
time
it
could
read
only
call
account
a
random
interface
to
the
country,
want
a
machine
to
launch
a
HTTP
server,
a
standard,
HTTP
server,
and
it
will
return
the
URL
of
this
streaming
server
back
to
cooperate
and
then
a
read
and
learn
trigger
redirect
response
to
API
server.
So
from
this
time,
the
equalizer
have
established
a
fresh
new
connection
to
the
streaming
server.
So
there's
no
business
of
criminal
cooperate
anymore
and
this
seeming
server.
We
are
responsible
to
handle.
A
All
of
the
streamer
requests
happen
next
and
that's
how
frankly
works
today
actually
and
the
couplet
we
all
call
is
the
our
X
cubed
API
and
the
frankly
runtime
service
will
be
responsible
to
return
the
URL
observer
streaming
server,
and
then
it
has
the
established
connection
to
the
streaming
server.
But
please
note
why
the
streaming
server
can
handle
the
extreme
a
request
that,
because,
as
implementation
with
the
I,
we
are
responsible
to
implement
those
three
streaming
run.
A
Hi
factions
only
three
functions
and,
what's
more
important,
CI
has
provided
us
with
a
very
useful
common
streaming
library
to
use,
including
the
HTTP
server
part,
the
handler
part.
They
were
just
read
from
part,
so
we
don't
need
to
take
care
of
this
streaming
server,
which
workers
on
how
we
implement
these
three
method.
That
would
be
much
simpler
right
so
and
that's
how
streaming
works.
The
third
thing
is
the
portable
resource
management.
A
It
is
very
important
in
the
kubernetes
1.6,
because
it
is
how
we
enforce
the
curious
classes
and
eviction
in
kubernetes,
and
that's
that
is
actually
the
core
idea
of
the
Borg,
the
cordial
buggy,
that
this
suit
makes
deploy
different
kinds
of
workloads
with
different
cut,
with
different
priorities
in
your
cluster.
So
there
can
be
preemption.
A
So
that's
why
we
have
three
kinds
of
queue
as
classic
synchronize,
the
first
of
all
Garret,
Hill
and
bustable,
and
best
effort
here
is
example
of
the
bust
Pol
Pot,
and
you
can
see
here
it
has
a
resource
request
and
resource
limit
and
they
are
not
equal.
So
that
means
this
pod
can
use
more
resource
than
its
resource
request,
because
the
scheduler
only
knows
about
requests.
So
that's
why
it
could
possible
and
the
kubernetes
implementing
this
model
by
using
the
so-called
poor
levels
to
groups.
A
So
like
the
example
here
this
pod,
when
this
pod
is
created,
cubelet
will
be
responsible
to
calculate
our
stammering
of
the,
for
example,
the
memory
limit
of
all
the
containers
in
the
in
that
pod
and
cackle
a
summary
on
what
this
value
into
a
special
secret
root
on
the
machine.
So
if
you
already
the
value
to
memory
limit
impasse
here,
so
that's
how
all
level
sequel
works,
and
this
is
very
useful
for
the
result
counting.
A
So
so,
the
outside
system
carry
this
value
and
it
also
imposed
an
important
for
charge
container
overhead
to
the
pole
instead
of
node,
because,
for
example,
the
streaming
server
I
mentioned
so
in
future
with
all
started
streaming
server
in
study,
spot
paula
was
the
group's,
so
it
cannot
consume
more
resource.
Then
the
pole
is
allowed
to
do
right
and
it
also
the
case
for
the
contradiction.
We
should
create
it
by
by
darker.
So
it's
it's.
A
It's
very
useful
and
this
cuddle
model
should
be
supported
in
frankly,
but
I
think
you
may
only
my
realize
that
frankly
does
not
need
to
use
this
oops.
The
group
for
a
polio
management,
but
please
note
that
you
cannot
start
a
virtual
machine
without
resource
limit.
It's
not
like
the
container,
so
every
resource
limit
should
be
said
before
this
poly
sandbox
starts.
That's
why
we
need
to
read
a
value
from
this
poor
level,
state
groups.
A
So,
for
example,
we
will
read
the
memory
memory
bytes
at
the
virtual
machine
as
a
memory
or
watch
machine,
and
we
were
also
rated
their
CPU
CPU
Kota
to
calculate
virtual
virtual
machines.
If
you
number
for
later
use-
and
you
may
also
found
that
if
any
of
these
value
on
a
stage
in
a
pod,
we
have
to
use
default
value
and
that's
not
good.
So
they
encourage
people
to
to
say
to
resource
limit
if
they
want
to
use
a
hype
container
at
the
runtime.
A
So
that's
a
resource
part
and-
and
the
next
thing
is
a
network
part
and
actually
just
a
few
months
ago.
This
is
impossible
because
the
thing
I
is
still
evolving,
which
cannot
step
out
of
the
wash
machine
plan.
Now
it
can.
It
can
do
that,
and
but
please
note
you
cannot
start
the
virtual
machine
before
you
can
ask.
You
cannot
start
working
machine
without
a
network
configuration
and
I
said
have
a
later:
it's
not
like
container.
A
You
cannot,
for
example,
and
join
it
a
container
to
another
namespace
or
use
NES
enter
tool
inside
to
configure
its
network
is
impossible
for
the
virtual
machine.
So
here
is
our
workflow,
which
is
a
little
bit
different
from
the
standard,
seeing
our
network
flow
so
first
or
we
will
create
a
network
namespace
for
standard
box.
This
is
the
standard
seeing
our
network
puggy
in
workflow,
and
we've
also
called
plugins
that
our
postmaster
to
configure
its
namespace.
This
is
also
the
same
as
seeing
right,
but
next
thing
different.
A
We
will
read
the
network
information
from
that
namespace
and
cache
it.
So
this
card
information
looks
like
this
step,
for
example,
is
interface
name
and
Matt.
Kuchar
IP
address
gave
an
abridged
name,
so
we
will
read
these
kind
of
information
from
the
namespace
and
without
checkpoint.
Is
the
namespace
passed
for
future
usage,
but
it
is
actually
teardown
and
we
will
use
this
catch.
The
network
information
to
configure
our
sandbox
machine
later
and
we'll
also
keep
scanning
the
the
thing.
A
It
will
break
the
current
model,
current
model
of
a
crew
blade.
We
need
to
refactoring
everything.
So,
that's
why
you,
frankly,
visa
probably
mix
the
run
times
any
content
which
claim
to
use
a
host
names,
main
space
of
all
provision,
node
or
have
the
spatial
annotation
will
be
running
automatically
by
using
a
building
or
a
nerd
talking.
A
So
this
cut
of
this
doctrine,
we
will
also
use
the
same
singer
network
plug-in,
so
the
dock
container
here
and
a
hub
container
can
communicate
with
each
other,
and
it
is
also
enable
you
to
mix
the
run
or
microservices
be
like
this
application,
because
hot
container
will
provide
you
with
a
dependent
kernel.
So
any
applications
which
can
running
can
be
running
in
your
universe.
A
Machines
of
the
commission
can
be
migrated
to
head
container
easily
without
extra
effort
and,
at
the
same
time,
it
also
run
your
newly
created
micro
services,
in
the
background
same
time
on
the
same
machine.
So
that's
good
and
it
also
enables
so-called
high
resource
infinity
model
which
internationally
in
box,
so
anyone.
So
anyone
remember
the
quadrille
book.
A
Yes,
it
encourages
you
to
deploy
mixed
the
workloads
with
mixed
priorities
inside
so,
for
example,
in
this
case,
you
can
just
run
your
critical
port
in
your
head
container,
the
guarantee
report,
so
it
will
be
protected
and
isolated
by
using
a
hypervisor
from
the
noisy,
neighborhoods
and
all
right
hand.
You
can
also,
at
the
same
time,
in
running
your
based
a
for
the
pod
or
bust,
upholding
dark
around
hands,
so
they
can
be
evicted
or
preempt
in
the
future
will
easily.
So
that's
how
frankly
works,
but
frankly,
the
axe
only
one.
A
Only
apart
from
the
whole
picture,
if
you
really
want
to
Utica.
Ladies
in
a
multi-tenant
environment,
because
we
need
to
do
B-
is
the
multi-tenancy,
the
isolating
Network,
the
persistent
volume?
That's
why
they
established
the
hyper
net
is
projected
in
a
fully
open
source,
so
the
architecture
of
High
Plains
projects,
a
quite
simple,
with
just
a
make
use
of
some
existing
projects
like,
for
example,
the
king
stone.
A
You
can
start
to
manage
over
multi-tenancy
and
we
have
over
top-level
sauce,
which
name
is
network
and
the
tenant
who
network
marketing
is
1
to
N
and
for
network
part.
It
is
actually
the
network
it
is
matching
to
the
nutria
moat,
so
the
crew
blade
will
be
responsible
for
to
call
the
master
likes
that
uphold,
and
this
request
will
be
delivered
to
a
small
team
and
running
your
machine,
which
name
is
the
crew
bees
tank.
A
So
it
will
either
broker
between
crew
blade
to
the
nutrient
API
translator
actually
and
for
persistent
volume,
part
which
is
more
either
because
we
are
using
high
quality
or
standard
box
one.
So
we
can
directly
attach
a
block
device
by
casting
the
volume
to
these
10
bucks
and
done
so
and
that's
why
we
modify
this
thing.
They're
planning
to
do
that
and
this
actually,
what
is
supporting
the
high
price?
A
Now
we
have
kubernetes
1.6,
the
cre
has
become
released
and,
frankly
has
become
an
official
project.
So
now
we
are
actually
refactoring.
It
is
to
use
frankly
directory
and
we
were
also
change.
This
could
be
stank
demon
standards,
the
UI
plug-in
and
we
were
also
reflects
volume
plugin
to
replace
our
homes
in
their
world
plugin
and
we
will
use
GPR
to
manage
or
network
and
we
will
use.
The
IP
is
a
very
constant
Kingston.
Plugin
for
the
bulletin,
in
part,
so
for
anyone
interested
in
this
project,
please
pay
attention
to
over
happening.
A
This
happen,
ADC
repo!
Well,
so
am,
let's
retrace
summary
part,
and
today
what
I'm
told
what
what
I'm
talking
about
it
actually
about
contraband
ham
interface
and
it
simplified
the
most
tricky
part
of
country,
one
having
integration
work.
For
example,
it
eliminates
the
potencial
runtime
api
incorporate
and
it
also
provide
a
very
clean
runtime
lifecycle,
including
the
post
and
box.
The
container
the
image
API
is
very
good
and
it
also
provided
with
have
some
interesting
tools.
For
example,
the
check
point
for
us
to
store
some
auxiliary
data
in
the
run
time
stream
style.
A
So
we
don't
need
to
call
API
server
or
crew
blade
to
get
this
information
or
write
a
temporary
file
on
the
note,
2,
regular
load
data
and
for
the
streaming
part
is
much
better
because
it
actually
leaves
implementation
to
the
run
time.
Machine
so
all
depends
on
how
you
implemented
this
kind
of
stuff,
and
it
also
provided
us
with
a
common
streaming
server
library.
A
So
there's
no
too
much
problem
on
our
implementation
side
and
today
also
based
on
the
so-called
happiness
project,
which
actually
I
mean
it
is
the
kubernetes
plug-in
that
makes
the
real
innovation
possible,
for
example,
a
third
party
resource.
We
use
that
to
manage
over
a
network
object,
and
this
thing,
I
network
I,
need
to
say
it
really
simple,
because
we
need
to
be
to
implement
two
methods
there
and
it
a
very
powerful
and
on
the
other
hand
it
also
investigated
the
kinetic
model.
A
But
we
need
to
say
it
is
impossible
to
use
that
for
runtime
other
than
darker
and
dissipating,
and
that's
why
we
said
kubernetes
is
a
very
great
project
because
it
actually
enable
more
possibilities,
and
that's
also-
and
so,
if
you
are
asking
something
that
can
works,
maybe
you
can
use
one
key
for
that.
But
if
you
are
fighting
something
that
you
want
to
build
your
own
distributing
solution
system
like
hyper
natives.
On
top
of
that,
you
want
to
be
relying
on
that.
Then
you
want
to
use
something
out
of
dependency.
A
I
say
it
should
be
kubernetes,
and
that
is
also
why
I
think
the
success
of
country,
random
interface,
it
actually
the
success
of
oxygen
project
itself.
It
is
because
the
users,
the
users
of
kubernetes
or
things
like
that,
should
never
care
about
the
details
that
were
runtime
stuff.
That's
why
I
think
docker
has
donated
the
content
B
to
C
in
C
F,
because
it
also
realized
that
fact
we're
all
users
who
never
care
about
the
runtime
stings.
Well,
thank
you.
Thank
you
for
coming
and
this
end
of
my
speak.
A
I'd
understand
oh
yeah.
Actually
there
is
a
park
inside
a
hyper
demo,
so
I'm
actually
most
of
the
field
end-to-end
test
it
caused
by
that
they
are
using
some
images
from
Google,
and
this
images
are
different
from
what
we
are
unit
8.
For
example,
there
is
no
past
in
the
past,
defining
their
image
or
their
binary
past
it
different
from
the
normal
places,
and
things
like
that.
So
we
need
to
do
this.
The
special
case
in
your
hyper
start.
We
have
standard
auto
patch
hyper
start
with
your
with
that
yeah.