►
From YouTube: Fly Your Containerized Environments by Joint Work of Harbor and Dragonfly - Jia Zou & Allen Sun
Description
Join us for Kubernetes Forums Seoul, Sydney, Bengaluru and Delhi - learn more at kubecon.io
Don't miss KubeCon + CloudNativeCon 2020 events in Amsterdam March 30 - April 2, Shanghai July 28-30 and Boston November 17-20! Learn more at kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects
Fly Your Containerized Environments by Joint Work of Harbor and Dragonfly - Jia Zou, VMware & Allen Sun, Alibaba
With the emergence and development of Kubernetes, it's becoming possible to run and operate large-scale containerized applications and services in enterprise environments. How to securely and effectively manage the lots of container images produced in the enterprise organizations and distribute them to the large-scale runtimes with less time and efforts when starting applications or services on demand is still a challenge. In this presentation, we'll introduce a joint solution from the open source trust cloud-native registry Harbor and the open source intelligent P2P based file distribution system Dragonfly to address the above challenge. The main contents include: 1. Manage your container images in a secure and effective way 2. Distribute your container image in the P2P way by layers 3. Pre-release your container applications from the image registry 4. Let's see the fantastic demo
To learn more: https://sched.co/GrYh
A
Let's
start
hello
good
afternoon,
hi
everywhere.
Thank
you
for
coming
here
to
join
our
session.
My
name
is
Steven
so
engineer
from
America
China
this
session.
We
have
another
speaker,
Allison
engineer
from
Alibaba
Group
today.
We
are
so
glad
to
be
here
to
share
a
idea
of
how
to
accelerate
her.
Your
containerized
environments,
by
joining
the
work
of
harbor
and
a
dragonfly
yeah
I'm
on
meet
Inner,
Harbor
and
Ellen
is
the
material
of
a
dragonfly
with
the
development
of
communities.
A
It's
the
pot
pot
or
wrong
and
operate
a
larger
scale
container
at
the
service
in
the
enterprise
environment,
but
they're
rusty
or
some
big
challenge
we
needed
to
face,
including
how
to
manager
the
numerous
continue
images
produced
in
your
organization
securely
and
efficiently
and
the
how
to
distribute
to
them.
So
the
largest
car
run
times
with
less
time
on
the
Mount.
Ok,
we
hope
we
can
find
the
answer.
The
you
need
to
propose
the
solution.
A
Hurry
the
agenda
of
today's
talk,
four
sections
into
two:
first
I'll,
give
you
a
quick
overview
of
what's
Harbor
and
what
you
can
do
with
Harbor.
Then
I
will
give
you
a
quick
introduction
about
the
checkered
flag
in
the
sector
session.
Next
I
will
talk
about
the
integration
between
Harbor
and
the
Dragonfly,
and
how
does
the
idea
Harper
to
accelerate
the
continued
writing
environment
in
the
last
section
will
show
a
demo
to
demonstrators
as
in
errors
of
the
solution.
Okay,
our
first
part
Harbor,
a
quick
survey
who
knows
Harbor
already
please
to
raise
your
head.
A
Oh
okay,
pretty
good,
so
I
think
I
can
talk
a
less
what's
harbor.
Okay,
harbor
is
the
open
source
trot
state,
a
cloud
native
registry
project
act?
Actually
the
so
Logan
shows
what's
harbor
want
to
be
okay,
open
source,
Harbor
work,
evil
as
the
open
source
project
and
make
open
and
a
transparent
and
the
maintained
by
the
global
community.
Okay
and
their
trust,
hater
yeah
Harbor
its
committed
to
make
the
a
major
measurement
Securi
effective
and
are
reliable,
okay
and
the
cloud
need.
He
was
that's
what
we
want
to
aim.
A
Praise
serve
image
at
the
charts
for
the
cloud
Andy.
Do
you
very
much
like
a
corn
eighties?
Okay
and
I
want
to
share
more
information
about
harbor,
its
initial
initiated
by
remember
China
team
and
now
matenda
by
the
global
community,
and
it
her
had
already
been
adopted
by
many
enterprise
products,
and
it's
lessons,
though,
with
apache2
and
hurry,
it's
our
key
to
harbor
repo.
It's
so
simple,
go
harbor,
slash
harbor!
If
you
have
any
questions
and
any
new
requirement,
please
come
to
our
github
repo
and
we're
ticket
seriously.
A
Okay
Harry
is
a
currently
situation
for
the
Harbor
community.
So
far
we
have
got
six
thousand
and
one
hundred
replies
I
started.
You
could
have
a
repo
with
to
twenty
thousand
the
tunnels
and
when
sudden
understand
one
hundred
of
works
and
additionally
coulda
to
say
there
are
more
than
one
hundred
contributors
to
in
contributions
to
harbor
project.
We
really
appreciated
the
contribution
from
the
community.
Okay,
next
I'll
quickly
go
through
what
a
Harper
had
her
pre.
You
too
many
meant
your
image.
A
Okay,
the
first
access
to
control
yeah
in
harbor,
the
content,
that's
the
continued
image
and
her
matures
is
isolated
by
projects.
Okay,
the
rule
based
access
control
is
applied
to
restricted
access
into
the
content
yeah.
When
you,
sir
everyone
used
to
want
to
access
a
Content,
sir,
it
must
be
added
into
the
project
as
a
member
and
the
garage
relevant
rules.
A
Currently
Harbor
has
three
predefined
rules
when
is
cast.
That
means
only
have
a
read
permission
comparing
to
the
talker
action.
It's
a
talker,
Pro,
okay
and
another
way.
It's
the
developer.
Developer
has
read
and
the
rights
of
formation
that
amis,
a
team
ever
can't
talk
or
pro
push,
imager,
okay-
and
there
is
another
way-
is
an
homage
traitor.
Besides
the
write
permission,
they
can
set
some
management
settings
it
to
the
project.
This
is
just
a
row
based
access
control.
A
A
Okay,
yeah
content,
trust.
Okay,
when
you
put
an
image
from
the
registry,
you'll
want
to
make
sure
the
image
is
the
one.
You
know
public
repair,
the
creator
and
it's
not
a
temperature
right
by
how
so
we
needed
to
send
the
imager
Harper
leverage
open
source
project
not
worry
to
support
the
content,
trust,
ok,
notary
it
also
a
sensor
for
hotel
project.
The
image
will
be
send
when
you're
pushing
imager
to
harbor
and
when
you're
pulling
the
image
of
the
signatures
this
or
will
be
read,
fried
and
digits
the
will
be
returned.
A
A
A
When
ability
scanning
can
hire
you
to
look
at
those
one
ability
based
on
Mitra
Savita
base
in
harbor,
we
leverage
another
open
source
project
layer
to
support
a
vulnerability
scan
you,
okay,
the
CVD
teacher
will
be
retrieved
from
Marty
passivity
to
base
like
a
redhead,
ARPA
and
so
on,
and
those
the
CBT
therapy
updated
regularly
to
make
sure
the
CVD
turret
latest.
Oh
okay
and
the
in
Harbor
you.
There
are
several
ways
to
trigger
the
scanning
process.
A
Yoky
ask
a
scale
scheduled
scan.
You,
for
example,
scan
image
at
one
o'clock,
each
Sunday.
Okay,
you
can
also
enable
a
event
based
mode.
That
means
the
west
of
the
image
it
posted
to
harbor
it
over
P
SCADA,
immediately,
of
course,
always
supposed
to
manually
record
enemies.
You
can
trigger
it
at
any
time.
You
you
want
Oh.
A
Additionally,
you
can
enforce
a
control
policy
to
block
the
puddings
which
do
not
match
the
threshold
at
your
site.
For
example,
you
can
set
the
policy
only
image
with
the
vulnerability.
Sir
sorority,
you
know
a
70
level
lower
the
meteor
can
be
allotted
to
pull.
If
one
amy
has
some
high
availability
high
level,
one
Betty,
the
pudding
will
be
blocked.
Okay,
that's
a
control
policy.
A
This
is
a
screenshot
of
photo
vulnerability
scan
after
the
image
is
the
successive
scare.
There
will
be
a
scan
result
of
report
summary
to
show
how
many
a1
ability
found
with
different
severity,
and
if
you
wanted
to
check
more
details
of
Harper,
you
can,
as
there
is
that
ETO
page
for
you
to
check,
which
is
a
VE,
is
found
in
the
imager.
A
Ok
replication.
This
is
a
definite,
a
feature
of
Harper
yeah.
Sometimes
you
may
deploy
multiple
registries
and
you
wanted
to
keep
the
Cotentin
second
eyes
the
muscles
deployment
registry
instance.
So
we
have
a
replication.
You
can't
FL
replication
policy
is
the
source
instance
and
status
I'm,
afraid
we're
too
late
narrow
down
which
image
can
be
replicated
from
the
source
to
the
target,
and
there
are
several
trick
mode
for
you
to
select.
You
can
schedule
a
set
schedule
for
the
replication
and
you
can
impose
that
you
know
the
event
based
the
model.
A
That
means,
if
new
image
it
Porsche
to
harbour
is
it
can
be
immediate
early
replicate
that
was
the
target
was
okay
here,
I
want
to
mention
the
replication
model.
Actually,
the
foster
replication
will
be
a
that.
We
call
it
initial
replica
didn't
meet.
The
existing
saw
similar
were
P
totally
replica
to
the
target
ax,
but
after
that
it's
verb
attorney
to
incremental
mode.
That
means
only
the
in
Kurnool
image
of
the
dilated
image
can
be
replicated
to
the
target.
A
A
Each
replication
rose
you
can
edit
or
trigger
it
in
the
portal
and
of
the
trigger
history
will
be
recorded
the
in
the
history
list
and
you
can
check
out
the
status
and
even
the
log,
okay,
yeah
yeah.
Besides
the
supporting
image
management,
we
are
her
Machado
poster.
Now,
with
the
same
experience
with
image
management,
you
can
easily
manage
your
charts,
iver
harbor,
the
hermit
rally,
also
isolated
in
the
project.
A
You
can
manage
ur
the
chart
and
the
you
can
manager
the
chart
of
versions
under
the
chart
and
you
can
attract
the
teachers
of
the
chart,
including
the
metadata
of
the
Charter,
the
dependents
of
the
chart
and
the
value
of
the
chart.
Okay
in
the
future,
we
are
go
further
to
provide
a
mode
of
the
month
of
each
other
to
come
to
converging
image
and
the
Hermit
charts.
A
Let
me,
including
you
know,
a
counter
trust
for
the
harm
chart
of
an
abyss
gunning
for
the
Charter
and
the
replicator
for
the
chart,
even
export
a
chart
as
a
offline
package
or
importer
of
lab
actually
into
a
hover,
and
if
you
want
to
learn
more,
please
join
our
harbor
tippity-tap
session
at
4:30
in
story,
A
or
B,
as
harbor
is
already
a
home.
Try
to
rip
away
compatible
with
the
harm
command
line.
A
Yoky
use,
Hermer
command
to
add
a
harbor
as
in
to
your
proposal
released,
and
you
can
then
to
search
fetch
install
it's
totally,
ok
and
we
also
support
us
a
harmful
plugin.
You
can
install
the
pusher
plugin
today
you
can
upload
the
harm
chart
from
the
terminal
used,
come
on
Lynette,
okay,
a
shade
of
Libra
deployed
I,
actually
harbor,
to
support
your
enterprise
productions.
The
never
it'll
be
delivered
up
a
harbor
hermit
chart.
A
A
The
Belk
waist
up
low
once
is
a
harbor
of
service.
The
coast
officer
will
provide
the
EPA
also
radiation
and
also
technician
service
and
the
web
services
for
the
US
even
extend
extended
the
charter
service
and
the
toposurface
were
run
some
background.
The
jobs
for
the
harbor
and
the
mid
server
will
cover
the
global
configuration
for
harbor.
We
also
leverage
the
several
set
of
product
component.
A
I
have
measured
before
use
the
NGS
for
a
PR
routine,
a
military
for
image,
storage
and
retrieving
under
Charter
Museum
for
a
charter
storing
and
indexing,
and
one
base
can
use
Claire
content,
also
for
not
only
for
connect,
roster
and
yeah.
We
also
have
a
packaging
component
to
provide
multiple,
deep
olive
oils.
You
can
use
Takemoto
to
deploy
harbor
to
your
download
environment
and
you
can
use
harbor
herb
charity
Pearl
Harbor
to
kubernetes.
A
B
Hi
everybody
nice
to
meet
you
guys
too,
to
shares
a
distribution
system
dragonfly
before
my
presentation.
I
would
ask
you
several
questions.
First,
one
have
you
ever
container
realized
all
the
ligands
the
applications
in
your
enterprise?
If
you
have
done
that,
please
raise
your
hand.
Oh
that's
great
something
positive
and
the
second
pass
phase
is
your.
Our
images.
B
Are
your
images
size
on
average
is
large,
then
maybe
one
gigabyte
whoa
some
guys
have
images
larger
than
one
gigabyte
and
you
met
some
ish.
You
mean
some
issues
when
distribution
images,
yeah
yeah
yeah,
for
example,
yeah,
so,
distribution
efficiency
like
a
very
large
imager.
You
need
to
spend
lots
of
time
to
download
it
and
we're
downloading
tasks
influence
so
working
applications
on
the
same
node.
It's
the
issue
we
may
be,
you
may
met
you
may
meet
so
to
tanco
to
tackle
all
these
problems.
Maybe
you
can
consider
dragonfly.
B
Okay,
dragonfly
is
an
open
source,
a
p2p
based
image,
a
distributing
system,
but
the
p2p.
This
part
is
a
little
different
from
BitTorrent.
Some
other
protocols
and
the
currently
dragonfly
is
a
is
a
sandbox
level
project
in
CN
CF.
Oh,
let
let
me
introduce
more
features
about
dragonfly.
First
dragonfly
is
focused
on
serving
distribution
programs
in
C
and
C
F.
What
dimensions
of
distribution
we
can
divide
it
into
three
parts?
So
stakin
are
the
first
parties
that
defeat
the
distribution
efficiency.
B
We
use
p2p
based
a
media
distribution
to
makes
a
node
to
construct
a
p2p
network
to
blah
blah
beautiful
data.
You
fell
your
image
layer
and
the
second.
We
provide
the
passive
CDN
to
avoid
repeatable
downloads
from
your
data
center
to
the
remote
registry.
If
you,
if
your
images
are
stored
in
dog,
hub
or
Google's
registry
hub,
if
you're
lots
of
lots
of
nodes
are
trying
to
pulling
images
from
set
I
think
there
will
be
lots
of
repeatable
downloads.
B
If
you
have
our
CDN
manager
to
handle
this,
and
that
distribution
will
be
much
more
efficient
and
the
second
how
to
try
to
control
flow,
your
distribution
or
do
the
distribution
flow
control
as
what
I
suggest
and
now
sometimes
image
downloading
tasks.
We
are
influenced
the
working
applications
in
the
container.
Him
know
how
to
avoid
this,
how
to
how
to
take
care
of
the
disk
I/o
and
the
network.
B
Speed
from
the
working
applications,
CC
is
also
what
we
need
to
take
into
consideration,
because
our
team
provide
continuous
service
in
single
ababa
infrastructure
team,
so
daily
containers
gear
is
more
than
one
1
million,
so
there
are
lots
of
image,
pooling
tasks
sphere.
So
if
the
image
pooling
and
their
services
provided
by
the
running
continuous
on
the
same
node
say
say
interrupt
each
other,
that
it
isn't
definitely
a
nightmare.
So
we
need
to
need
to
avoid
that
and
the
third
one
is
as
a
distribution,
security,
part
and
just
image
distribution
security.
B
We
need
to
take
into
considerations
encrypting
things
and
something
like
that
last
about
not
at
least
I,
think
that
implies
quite
easy
and
simple
to
use.
It
is
totally
non-invasive
to
your
content
engines
like
Talk
Talk
engine
like
container
D,
actually,
actually
in
Alibaba,
we
have
another
container
engine
which
is
named
a
by
paunch
container
I,
think
it
is
provides
more
strong
isolations
than
the
general
container
engines.
Also
punch
container
is
also
open
source
on
github.
B
B
B
So
so,
with
the
number
of
growth
yeah,
you
can.
So
if
the
efficiency
improves
a
lot,
the
second
graph,
it
is
about
distributing
statistics
in
Alibaba
actually
last
year
in
Alibaba,
we've
already
distributed
3.4
2pb
data
in
just
one
month.
Maybe
you
know
that
pebble
11
the
most
popular
shopping
festival
in
China
yeah.
B
Yeah,
what's
that
texture
of
dragonfly,
let's
show
us
a
little
bit.
This
is
a
very
simple
act:
temperature
of
dragonfly.
We
can
divide
this
in
this
active
chain
to
three
parts.
The
first
part
on
the
top
is
the
Dragonfly
control
controller.
One
can
control
job
of
live
as
a
restful
api
and
maybe
in
five
minutes.
We
wish
your
demo
to
see
how
to
control
dragonfly
with
hubber
and
does
it
middle
layer.
It
is
a
super
node.
We
can
see
in
these
app
temperatures
use
a
sever
super
Noda.
B
What's
the
usage
of
DF
get
DF
get
that
plays
a
role
of
a
peer.
This
is
peer,
will
do
the
downloading
scene
and
the
uploading
scene
why
to
download,
because
this
node
needs
image
why
to
upload,
because
it
is
also
a
peer
to
provide
uploading
services
for
as
a
node
and
that
what
is
dragonfly
demon,
DF
demon.
It
is
a
proxy
for
container
engine
to
for
images.
We
know
for
container
engine
how
to
provide
the
non
invasive
services.
It
only
can
recognize
pearl
images
from
a
registry.
B
All
the
protocols
based
on
the
services
provided
by
the
registry,
so
DF
daemon
plays
a
role
of
registry
mirror.
Then
it's
we're.
It
will
read
request
from
the
container
engine,
then
translated
into
two,
so
DF
get
for
the
DF
get.
It
can
only
react
nice
very,
very
general
file,
distribution
scene
and
the
DF
kit
and
the
Supra
know
that
will
construct
a
peer
network
to
transfer
data.
Just
a
fire.
No
image
is
that
at
all
yeah-
and
here
is
a
very
specific
actor
about
supernova
supernova.
B
B
First
is
the
p2p
scheduler
how
to
scheduler
peers
in
the
network,
how
to
choose
a
scheduling
algorithm
and
how
to
make
the
peer
status
consistent
with
that
after
stage
and
the
download
task
processor
something
like
that,
and
it
also
provides
a
CDN
functionality
for
the
users,
because
in
a
data
center,
if
we
all
the
nodes,
we
are
poor
images
from
the
outer
remote
registry.
Why
not?
We
use
a
CDN
to
handle
it,
so
the
CDN
manager
will
do
the
cache
function
for
the
image
of
fires
and
series.
B
Apart
for
the
transmission,
we
need
to
limit
the
transmission
rate
to
guarantee
some
quality
about
some
specified
amount
and
the
data
encryption
data
encryption
and
some
intelligent
flow
control.
Actually,
in
your
production
environment,
ok,
we
still
have
the
preheater
functionality
more
teachers,
CC's
walker,
workflow
of
dragonfly.
Imagine
here
is
a
node.
Oh,
who
wants
to
poor
image
how
to
construct
the
p2p
network
by
the
cluster
first,
it
must
have
send
out
a
image
polling
request
and
it
we
are
singed
and
requested
to
the
super
node.
B
Actually
it
is
done
by
the
DF
demon
and
DF
get
so
when
the
super
Noda
receives
the
request.
It
will
judge
whether
this
image
is
also
is
already
cached
by
myself,
if
not
it's
weird
downloaded
from
the
remote
registry.
At
this
time
it
praised,
plays
the
role
of
CDN
and
the
except
for
step.
So
super
Noda
will
reply
to
the
note
which
peer
meat,
which
peers
have
the
blocks.
B
Actually,
at
the
first
time
only
the
hoopin
node
has
the
blocks,
but
when
the
time
goes
on,
when
one
gets
the
block,
then
another
one
gets
a
block.
Then
in
the
network's
years
two
peers
have
the
blocks
and
it
we're
transports
the
file
blocks
among
the
other
peers
at
the
last
of
user.
We
will
finish
the
polling
when
a
blocks
down
are
downloaded,
and
this
is
a
very
simple
app
Church
about
dragon
flying
the
ecosystem.
B
We
can
use
harbor
to
do
the
immediate
store,
reducing
and
the
super
node
and
the
dragonfly
to
the
distribution
scene
and
the
kubernetes
and
the
DF
keep
they
have
demon
and
the
DF
get
can
be
deployed.
Fire
demons
that
actually
in
Alibaba
we
use
punch
container
as
a
container
engine.
Ok,
this
is
a
very
brief
introduction
about
dragonfly.
If
you
need
more
details,
please
try
to
find
the
eternal
kid
yeah
and
the
next
is
about
how
to
integrate
harbor
and
a
dragonfly
to
fly
your
container
right.
The
environment,
Cinderella.
A
Ok,
we
don't
have
enough
time,
I
accelerate
the
process
in
this
part,
I
want
to
talk
about
the
integration
between
the
harbor
and
the
Rockefeller.
That
means
this
solution.
Its
aim
accelerator,
the
containerized
humor
meant
okay.
After
the
introduction
of
the
two
projects,
I
think
you
may
be
familiar
with
the
capability
providing
by
Harbor
and
dragonfly
Harbor
Harrison
managed
manager,
the
image
with
a
secure,
efficient
and
reliable
way,
and
the
dragon
franca
optimize
the
distribution
of
a
meter
by
a
p2p
way
so
properly.
A
You
may
have
noticed
that
there
is
a
reverse
and
strong
complementarity
between
the
two
projects
with
a
major
distribution:
optimization,
a
sponsored
by
dragonfly
Harbor
computer,
a
highway
to
distribute
of
the
a
required
imager
to
the
latter
scare
or
run
times.
Actually
that's
the
original
intention
to
create
as
such
idea.
A
So
what's
the
main
idea,
a
hopper
will
identify
the
truck
a
flat
surface
with
the
api's
of
dragonfly
and
the
mid
turns
of
surveys.
That
means
the
studies
in
harbor
side.
This
is
a
very
loose
copper
way,
just
like
the
image
from
harbor
side
can
be
transferred
to
the
cache
of
dragonfly
to
make
it
already
before
the
runtime
a
podium.
That's
the
preheater
process
and.
A
A
Okay,
this
is
the
integration
architecture
over
real
I.
Don't
ever
in
order
to
talk
more
about
that,
because
we
have
measured
a
lot
of
you
in
the
previous
slides
Dragon's
Lair
will
provide
a
API
to
light
a
harbor,
provide
an
API
to
talk
to
her
a
tool
Harbor
and
retrieve
a
harbour
will
pass
really
the
imagery
information
to
drag
a
flat.
Wrongful
I
will
use
this
image
image
information
to
retrieve
a
account
in
the
data
of
the
imager
tool,
their
content,
Akash.
A
A
A
Let's
go
for
it's
from
Harborside
use
your
username
and
password
logging
into
the
harbor
portal,
as
there
will
be
a
distribution
section
to
manager
the
immediate
III
distribution
related
stuffs.
There
is
a
provider
list
all
the
support.
It
also
supported
the
distribution.
A
provider
will
be
available
here
and
based
on
the
driver.
A
We
here
we
can
see
the
track
a
flat
driver
based
on
the
driver.
You
can
create
a
you
know,
a
distribution
instance.
We
could
give
you
a
user
name
or
something
to
go
fishing
and
more
important.
Is
that
the
point
here
we
imported
the
point
of
a
checkered
flag
and
the
way
to
not
be
by
any
authentication
mode.
So
here
we
select
now
and
to
save
the
service
instance.
A
Ok,
you
can
update
you
can
able
to
say
about
the
instance
and
once
the
instance
is
ready,
let's
bronze
the
image
slag
so
ago,
a
coca-cola
meter
with
2018
tag.
We
want
to
preheat
this
emitter,
just
click.
The
button.
Ok
result
is
returned
to
the
pre
header
its
launch,
and
then
we
can
pack
it
to
the
history.
All
the
preheating
process
will
be
recorded
as
a
hitter
II.
A
In
the
this
list
you
can
track
the
status
Kirk
it's
running
after
where
the
city
theater
will
be
turned
into
a
success,
and
you
can
get
more
information
from
the
list,
for
example,
how
much
the
time
spent
on
this
preheating
process.
In
future,
we
can
support
policy
based
preheating
trigger
away
okay,
it's
Allen's
time.
Okay,.
B
B
So
here's
the
background.
We
provide
a
ten
note
to
do
the
experiment,
how
the
distribution
will
be
improved
and
here
is
a
two-terminal.
They
are
both
on
the
same
note
of
the
ten
notes
we
just
to
see
what
the
status
sounds:
anode
and
sir
DF
gate
logger
rotating
okay.
We
see
this
node,
we
have
no
images
and
we
have
the
registry
mirror
setting
in
the
doc
engine
and
it
is
to
the
local
TF
demon
you
know
and
we
can
and
we
can
find
a
process
which
is
DF
demon.
B
B
B
It
really
takes
lots
of
time
for
the
40
seconds
to
make
it
actually
for
what
consumes
us,
what
candle
part
walk,
consume
so
much
time.
Actually,
two
parts,
the
first
one
is
the
downloading
part,
and
the
second
one
is
how
to
come
on
how
to
uncompress
see
all
these
images.
Yeah.
We
see
it's
40
42
seconds
yeah,
it's
the
images
of
1.53
get
bytes.