►
Description
Don’t miss out! Join us at our upcoming events: EnvoyCon Virtual on October 15 and KubeCon + CloudNativeCon North America 2020 Virtual from November 17-20. Learn more at https://kubecon.io. The conferences feature presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects.
containerd Deep Dive - Akihiro Suda, NTT & Wei Fu, Alibaba
Join containerd maintainers to discuss the design of containerd’s core services and how to make use of them. This talk will dive into the plugin design and use cases for extending containerd. With knowledge of containerd’s service design, learn how to implement custom backend components such as special sandboxing and image storage. Finally, come away with an understanding of how to customize and configure containerd for your use case, whether through Kubernetes or a standalone deployment of containerd.
https://sched.co/ZexS
A
A
A
A
A
A
A
This
slide
compares
regularly
charges
it
and
discharges
it
with
regular
situation.
You
can't
inspect
fire
offsets
without
downloading
and
reading
the
entire
archive,
but
with
strategy
you
can
instead,
the
fire
offsets
immediately
right
after
reading
the
footer
structure,
because
the
footer
structure
contains
offset
information
of
all
files.
A
A
A
A
A
In
colorado,
1.3,
the
rudras
mode
did
not
support
imposing
resource
limitations,
as
in
docker
run
dutch
cpus
and
dokkara
dash
memory,
because
amphibious
users
cannot
control
c
groups
on
single
version.
One
host
and
yet
continuod
did
not
support
the
overlapping
snapshot
because
of
the
limitation
of
the
kernel.
A
A
A
A
A
A
A
A
B
B
Before
we
get
into
the
detail,
let's
recap:
the
architecture
look
at
this
picture.
Kennedy
is
for
the
very
more
grpc
service
for
the
internal
component
like
container
service
content,
service,
dev
service
and
so
on.
The
internal
components
maintain
a
very
low
level
resources
so
that
those
api
is
very
general
purpose.
B
B
The
answer
is
no
kennedy
care
care
more
about
the
interface
design,
so
they
know
so
they
know
how
to
work
with
your
very
special
backend
implementation
to
handle
the
contender
lifecycle
and
image
data.
So
you
don't
need
to
the
wrapper.
You
don't
need
to
do
the
hacking
thing.
Just
as
some
configuration
the
kennedy
will
know
how
to
use
your
very
special
backing
invitation.
B
B
B
B
Look
at
this
picture.
When
you
download
the
image,
the
client
will
call
the
div
service
to
read
image
layer
data
to
get
the
arc
to
the
tar
archive
and
then
unpacking
to
the
snapshot.
If
the
image
layer
is
with
the
gzip
compression,
the
kinetic
will
decompress
it
into
the
third
string
before
I
before
unpack.
B
We
know
gladiator
only
supports
gzip
compression
right
now.
So
what
if
the
user
want
to
use
the
z
standard
image
data?
So
what?
If
the
user
wants
to
use
the
encrypted
image
data
at
the
beginning,
kinetic
integrity
with
the
encrypted
media
type
in
the
code
base?
But
it
is
all
about
the
implementation
design
rather
than
interface.
B
B
B
B
B
B
B
B
Currently
provides
a
built-in
snapshot.
Plugins
and
one
varies
is
very
special.
Is
this?
Is
the
prophecy
plugin?
It
is
used
to
redirect
the
request
to
the
remote
external
snapshot.
Plugin
like
internal
plugin,
like
internal
snapshot,
plugin,
the
external
plugin,
has
to
implement
those
api
server.
It's
api
service,
so
that
the
currently
know
how
to
communicate
with
your
own
server
like
a
streaming
processor.
B
You
just
need
you,
you
don't
need
to
add
some
configuration
to
tell
community
how
to
connect
it
to
your
server.
That
is
that's
very
simple.
So,
right
now
we
have
cover
the
the
streaming
processor
and
the
snow
and
the
snaps
of
nsf.
So
snap,
external
plugin,
how
we
know
there
are
there,
are
new
generation
also
runtime
implementation,
for
this
case
kennedy
will
handle
you
by
the
wrong
entire
choose
practical.
B
B
B
Can
it
just
need
to
focus
on
the
the
political
to
know
how
to
talk
to
each
client
at
the
wrong
time?
So
kennedy
introduced
the
first
class
stream
api
for
the
long-term
authors,
so
the
runtime
also
can
implement
their
own
stream
like
a
divisor,
color
contender
and
far
crack
yeah
runners
also
need
to
implement
those
api
service.
B
So
they
cannot
know
how
to
talk
to
your
server,
and
we
know
shimo
is
the
background
long-running
process
before
connected
to
it
connected
need
to
fire
it
and
start
it
in
order
to
make
the
integration
easier
kennedy
bring
a
value.
Nand
convention,
for
example,
when
you
use
io
dot,
connected
dot,
run
noisy
dot
v2
as
human
m
connective
wheel,
converted
it
to
the
tier
binary
net.
It
will
call
canada
d.
B
Heaven
stream
hiven
monster
highway
2.,
so
you
will
use
the
binary
name
and
to
find
the
binary
path
and
start
it
and
someone,
and
someone
will
confuse
about
the
name
about
between
the
between
the
one
c
dot
v,
one
and
one
c
dot
way.
Two
both
are
the
runtime
v2
implementation,
but
the
1cv2
implementation
has
a
more
feature
than
once.
The
dog
v1
like
a
grouping
feature
one
one
cv2
sim
cam
manager
more
than
one
container.
B
B
B
And
we
know
condition
right
now
is
use
the
name
type
as
default
channel
to
export
the
container
send
output
to
the
receiver
like
a
docker,
sji
plugin,
but
we
know
there
is
it
required.
The
receiver
must
be
alive
if
the
receiver
is
done
for
some
reason
and
the
content
is
still
running
and
still
output,
the
logging
data,
the
manpower,
will
be
haunt
and
then
he
impacts
the
running
container.