►
From YouTube: Faster Container Image Distribution on a Variety of Tools with Lazy... - Kohei Tokunaga & Tao Peng
Description
Don’t miss out! Join us at our next event: KubeCon + CloudNativeCon Europe 2022 in Valencia, Spain from May 17-20. 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.
Faster Container Image Distribution on a Variety of Tools with Lazy Pulling - Kohei Tokunaga, NTT Corporation & Tao Peng, Ant Financial
Pull is known as one of the time-consuming steps in the container lifecycle. One of the root causes is the current OCI Image Specification because it can't run containers before the entire image contents become locally available. In this session, Kohei and Tao introduce state-of-the-art OCI-alternative image formats discussed in the community for speeding up pulling images, especially putting the focus on eStargz and nydus. These formats enable "lazy pulling" that allows container runtime to startup a container without waiting for the entire image. Over the past year, lazy pulling has been available on more and more container-related tools including Kubernetes, containerd, CRI-O, Podman, BuildKit, etc. So they will show how widely these image formats can be used on these tools in daily activity. eStargz has been proposed to OCI Image Specification as an extension and nydus has so to the next version of that spec (a.k.a. OCIv2). The talk will also share the discussion status.
A
A
First,
I'm
gonna
give
you
a
brief
introduction
to
lazy
blink
playing
continent.
Image
is
known
as
one
of
the
time
consuming
steps
in
the
container
lifecycle.
This
has
been
affected.
Various
kinds
of
use
cases,
including
batch
job
execution
and
building
images,
their
efforts
to
minimize
the
size
of
container
images,
but
not
all
images
are
minimizable.
A
A
A
A
A
A
A
First,
let's
take
a
look
deeper.
Look
at
each
third
digit
image
format:
easter
egg
is
an
image
format
for
legibling.
This
is
backward
compatible
to
the
current
ocr
image
specification,
so
this
is
lazy,
poolable
from
oci
compliant
registries,
as
shown
in
the
following
figure,
and
even
legacy
and
lazy
playing
agnostic.
Runtimes
can
run
this
image
as
a
non-laser
image.
A
A
If
installation
comes
with
performance,
optimization
and
content,
verification,
prefetch
optimization
allows
important
files
to
be
prefetched
to
avoid
network
overhead.
Ethernet
is
discussed
in
continental
storage's
snapshot
project.
This
is
based
on
started
image
format
proposed
by
crvs,
but
comes
with
performance.
Optimization
and
content.
Verification,
as
shown
in
the
following
figure
it
started,
is
comparable
to
gzip
and
usable
as
a
valid
oci
or
docker
image
layer.
Each
file
is
compressed
as
a
separated
member
and
the
offset
to
each
member
in
the
layer
is
recorded
in
toc
file
appended
at
the
tail
of
this
probe.
A
This
enables
random
access
to
fire
entries
and
http
range
request
is
used
for
the
random
access
to
the
layer.
Blob
stored
on
the
registry.
Estrogen
supports
prefetch
of
files
that
are
likely
accessed
during
container
startup.
These
files
are
called
prioritized
files
before
continuous
setup,
container
runtime
prefetches
them
to
mitigate
the
network,
related
overheads
for
reading
them.
A
A
This
shows
the
time
to
take
photo
python
3.7
image
among
three
types
of
formats
legacy
images
cannot
start
container
before
the
entire
image
contents
become
locally
available,
so
startup
takes
accordingly.
It
started.
No
opt
is
eastern
image
without
prefetch
optimization.
This
shows
the
time
to
take
for
startup,
but
there
is
a
random
performance
drawback
because
of
the
overhead
of
on-demand.
Fitting
ethers
can
mediate
this
by
profiting
note
that
creating
container
takes
longer
than
optimized
estrogen,
because
prefetch
occurs
during
this
step.
A
Glassfish
server
image
also
shows
performance,
improvement
of
pool
by
e-sergeant
image
and
same
as
the
python
image.
Optimized
eastern
visit
mitigates
runtime
performance
drawback
of
on-demand
fetching
lazy.
Blink
can
be
enabled
on
kubernetes,
as
shown
in
the
following
figure:
plugins
are
needed
for
cri
runtimes
search,
sergey
snapshot
is
used
for
quantity
and
sergey
store
is
used
for
cryo.
A
We
will
introduce
these
plugins
later.
There
is
a
real-world
use
case
of
cern
for
speeding
up
analysis
pipeline
k3s
supports
discipline
of
ether
digit.
This
patch
is
already
matched
to
the
main
branch,
so
will
be
included
in
version.
1.22
third
snapshot
provides
a
kind
note
image
that
enables
lazy
playing
of
ethers
event.
This
is
a
demo
of
running
hello,
world
golang
ci
as
a
workflow
on
k3s.
A
For
continuity
search,
the
snapshot
plugin
enables
laser
blink.
This
is
implemented
as
a
remote
snapshot.
Plugin
of
quantum
d,
as
shown
in
the
following
figure.
Sergey
snapshot
mounts
each
solid
layer
on
the
node
using
fuse.
Each
file.
Contents
are
downloaded
from
the
registry.
Lazily
started,
snapshotter
communicates
with
quantity
via
dlpc
api
and
provides
field
mounted
e-suggested
layers
to
quantity.
A
A
This
is
a
demo
of
lazy
playing
of
hello.
World
is
started
image
on
continuity
with
another
gtl.
The
upper
console
shows,
lazy,
playing
of
ether,
did
it
and
the
lowest
shows
normal
non-relatable
each
stage
that
enables
content
to
run
the
container
without
waiting
for
pool
completion?
So
it
finishes
in
about
5
seconds.
A
For
pottermore
and
cryo
storage
store,
plugin
enables
lazy
blink.
This
is
implemented
as
an
additional
layer.
Stop
logging
available
from
podman
version,
3.3.0
and
cryo
version
1.22.0
additional
layers
to
allow
sportsman
and
cryo
to
acquire
container
image
layers
without
pulling
them
from
the
registry,
as
shown
in
the
following
figure
sergeants
to
amount
extracted
each
suggested
layer
on
the
host
using
fuse.
A
Each
file
contents
are
downloaded
from
the
registry
lazily.
Third,
this
tool
communicates
with
sportman
and
cryo
via
file-based
api
and
provides
fuse-mounted
e-suggested
layers
by
using
cryo.
With
this
configuration
you
can
perform
lazy
playing
on
kubernetes.
If
you
use
podman
with
star
digit
store,
you
can
perform
lazy
playing
on
your
time
now
as
well.
A
This
is
a
demo
of
lazy
pulling
of
hello
world
estrogen
image
on
podman,
the
upper
console
shows,
lazy
playing
of
ether.
Did
it
and
the
lower
shows
normal
non-lasable
portman
finishes
running
its
third
in
about
five
seconds
already
exists.
Message
shows
that
the
podman
acquired
image
layers
from
the
studies
store.
A
Buildkit
experimentally
supports
latency
playing
of
e30
that
formatted
based
images
since
version
0.8.0
in
general.
Every
build
starts
from
preparing
based
images
by
from
instructions
of
dockerfile.
When
you
build
an
image
using
large-based
images
that
aren't
cached
on
the
node,
they
need
to
be
ports
here
and
it
can
take
a
long
time.
This
lazy
playing
feature
is
trying
to
solve
it.
You
can
use
this
feature
via
docker
build
x,
as
shown
in
the
command
example.
A
A
variety
of
tools
in
the
community
start
to
support
creating
e-surgery
images,
buildkit
and
canonical
support.
Building
e-suggested
image
buildkit
supports
building
is
as
one
of
the
compression
modes
you
can
use
this
feature
through
local
build
x.
Compression
equal
e
strategies
option
enables
this
feature:
kyanko
supports
building
easterners.
A
A
A
A
A
Among
these
implementations,
we
are
proposing
its
strategies
to
oci
image
specification,
because
each
stage
is
comparable
to
the
current
oc
image.
Spec.
We
are
proposing
this
as
backward
compatible.
Extensions
to
the
current
ocl
spec
feel
free
to
join
the
discussion
from
the
next
slide.
Tao
introduced,
nida's
image
format.
B
B
B
Firstly,
it
can
support
ziplining
with
medium
prefixed
policies.
While
we
do
when
we
do
poor
images,
nadas
image
from
standard
region
3,
we
want
to
make
sure
that
even
if
the
network
goes
down
accidentally,
we
can
still
access
the
data
locally.
So
we
have
so
we
we
want
to
pull
the
data
prefix
data
before,
even
if
the
application
doesn't
access
it
because
it
can,
it
may
access
it
later
on.
B
So
we
support
provisional
privilege
and
we
have
different
privilege
policies
such
as
semantically
or
hintedly
or
biochemical
just
background
prefetching
and
another
thing
we
support
is
chocolate,
duplication
with
natural,
build
cash.
What
is
chuckling
about
duplication?
We
we
can
increase
it
to
duplicate
the
ratio
of
a
computer
image
not
very
dramatically
and
with
native
build
case
we
can.
B
It
improves
image
conversion
from
a
central
oci
image
to
let
us
image
very,
very
largely,
and
we
support
end-to-end
data.
Integrating.
That
means
we,
we
can
optionally
verify
the
data
contents
when
we
read
any
data
locally
or
from
this,
which
is
to
make
sure
that
it
is
actually
it
is
actually
the
data
that
is
being
built
into
the
image.
B
The
the
application
will
never
know
will
never
know
it,
but
we
want
to
provide
the
capability
to
find
out
such
such
situations,
so
that
in
that,
as
we
can
optionally
verify
it
and
let
us
support
reproducible
image
building
the
ambience.
We
can
build
another
image
from
source
heading
machine.
It
will
be
you
it
will
not
be
affected
by
the
conversion
environment.
B
A
B
That
the
lettuce
is
usable
with
a
standard
cloud
registry
also
nadas
has
switched
container
ecosystem
intuition.
It
works
with
kubernetes
docker,
container
d,
build
kit,
hardware
and
drink
fly
and
different
runtime
different
container
runtimes
such
as
currency
and
color
containers.
It
is
very
resource
efficient
and
production,
ready.
It
costs.
B
So
this
is
architecture
of
anas
in
the
middle
we
have
a
fuse
demon
and
it
provides
both
endpoint
and
verticals.
In
the
port
field.
Fuel
enterprise
allows
the
traditional
zoning
container
to
exist.
The
image
data
directly
and
the
water
ifs
does
the
same
for
cataclysm
and
it
has
shared
uncompressed
local
cache.
So
the
docker
cache
is
uncompressed,
though,
when
being
pulled
from
from
remote,
remote
storage,
and
it
will
you,
it
only
needs
to
be
uncompressed
once
and
it
will
be
shared
by
different
data
streams.
On
the
same
host.
B
B
B
It
divides
each
image
into
two
parts,
one
barcode
three
metadata
layer:
it
is
circumcised,
verifiable
by
imbec
definition,
and
also
it
has
chunk
chunk,
share
data
layer,
so
different
emitter.
Different
images
can
share
the
same
data
chunk
on
the
data
layer
also
to
be
oci,
speak
competitive
to
to
be
hostile
compatible.
We
introduced,
we
just
changed
one
field
in
ocie
manifest,
so
we
as
we
introduced
a
meter
new
meter
type
for
the
another
probe
and
and
the
rest
of
the
manifest,
is
compatible
with
the
outside
artifacts
and
the
swing
spec.
B
So
that
means
it
is
compatible
with
which
is
three
functionality
as
well,
and
it
is
vital,
widely
working
with
many
container
registers
such
as
in
many
colored
registries,
and
it
works
with
docker
registry
and
also
it
works
with
harbor,
and
this
is
there's
a
putting
benchmark
for
where
we've
done
for
nadas,
as
we
can
see
for
oci
image,
the
end-to-end
time
for
oci
the
oci
image
creation,
their
time
increases
and
their
as
image
size
increases,
but
with
net
us.
The
time
is
mostly
consistent
or
below
4
seconds
for
different
images.
B
And
the
that
nets
works
with
variety
of
tools.
First,
we
provide
another
stem
short
plug
stem
shoulder
plugin
for
container
d.
So
this
is
a
link
for
the
stem
shoulder
and
with
with
this
plugin
that
works
with
building
meeting
container
compute
container
ecosystem
components
such
as
kubernetes
very
quickly.
It
works
with
ci
control,
it
works
with
nerd
console
engine
and
it
works
with
ci,
remote
or
ctrl
ctr
remote
tool.
B
B
B
B
B
B
Point
we
are
working
with
the
oci
community
to
push
all
the
artifacts
many
manifest
spec
to
the
to
the
community.
This
manufacture
functionality
is
tuned
back
and
the
stone
map
and
oci
image
manifested
to
an
artifact
many
manifest.
So
we
can
connect.
Let
us
let
us
manifest
to
the
original
oci
image
manifest
and
it
will.
This
will
help,
which
is
to
check
the
relation
between
different
images
and
we
we
are
propos
inside
the
pr.
We
are
proposing
lettuce
image
as
an
artifact
type
is
called
cncf.
A
B
B
Okay,
we,
let's,
let's
do
some
recap:
we've
introduced
both
strategies
and
lettuce
in
this
presentation
and
the
reason
we
do
it
is
this
mutual.
We.
We
are
aware
that
the
outside
images
are
large
and
slow.
We
need
we
need
to
do
the
business
pulling
to
start
containers
quickly
and
the
prefetching
is
a
good
friend
to
daisy
pulling
because
we
we
can
get
data
before
even
before
their
container
actually
use
it,
and
we
we.
B
For
for
for
your
strategy,
it,
its
key
feature,
is
background
by
capability
and
it's
an
extension
to
the
existing
oci
image
spec,
for
that
is
more
future
looking
and
it's
a
proposal
to
the
next
generation
of
ci
units
bank
and
here's
a
pair
are
some
information
for
the
strategic
project.
It
is
100
also
a
compatible
image
format
for
nasa
pune,
and
it
is
a
sub-project
of
cncf
graduated
project
container
d.
There
is
ending
for
his
resource
and
stack
slang
page
also.
There
are.
There
are
pre-converted
images
here
and
for
the
analysis
project.