►
From YouTube: Keynote: Cloud Native at Alibaba Scale - Xiang Li
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
Keynote: Cloud Native at Alibaba Scale - Xiang Li
As one of the biggest Internet companies in the world, Alibaba provides thousands of online/offline services to various customers to support their business. A typical Alibaba cluster consists of tens of thousands of nodes and manages more than ten thousand heterogeneous applications. To evolve and innovate Alibaba's infrastructure at a rapid pace, Alibaba started its journey of cloud native a few years ago. We fully containerized our applications, launched Kubernetes clusters with thousands of nodes, integrated Kubernetes into our container platform, and adopted emerging projects like Dragonfly, Envoy, and Helm. We are building an ecosystem around cloud native technologies. This talk will walk you through our successful journey to cloud native, identify key scalability and usability problems we faced to scale Kubernetes and the cloud native platform up to tens of thousands of nodes, and present the strategies to overcome them. We believe that the engineering effort and the experience we have are valuable to companies small and large for their path of being cloud native.
https://sched.co/NrqG
A
B
Of
all,
thank
you
for
the
moderators
introduction.
One
way
has
delay
by
the
time,
but
it's
good-
and
you
know
that's
the
physical
together
taxi,
but
I
see
the
audience
can
in
one
of
another.
Today,
I
want
to
talk
about
the
image
earlier
for
cognitive,
what
we
have
done
in
ally
group
for
some
of
the
progress
in
our
contributions,
and
we
talked
about
cloud,
the
native,
the
first,
the
topic
in
Cyprus,
and
why
we
want
to
what
is
a
cloud
native,
why
cloud
native
and
so
what
we
want
to
do.
B
Rather
native
and
I
just
say
that
in
recent
years
some
experience
and
practice
and
projects
I
want
to
share
with
you
all
standing
of
cloud
native.
One
of
the
objectives
of
our
group
is
to
enhance
the
resources
utilization.
Judicial
efficiency
started
early
on.
We
use
the
virtualization
technology
such
as
containers,
have
a
lower
overhead
and
the
convenience
next
and
we
use
the
scheduler
so
that
we
can
better
to
have
a
lesser
package
rate
to
lower
the
fragmentation.
We
try
to
use
storage,
networking
technology
to
stabilize
our
resources
by
these
progress.
B
We
have
increased
that
Ally
groups
resource
efficiency.
Secondly,
another
goal
of
a
cognitive
is
to
improve
their
developers.
Efficiency,
that's
why
we
could
quickly
try
technology
such
as
proven
artists.
We
help
to
do
the
orchestration
for
the
developers
to
automate
the
deployment,
after
that,
we
try
to
improve
our
CI
CD
system
from
R&D
to
deployment
shorten
the
time
front
for
the
writing,
to
align
and
to
have
a
higher
iteration
efficiency.
B
At
present,
we
are
developing
new
technologies,
such
as
a
service
March,
and
the
middleware
is
downloaded
to
the
middle,
where
we
are
able
to
manage
these
capabilities
in
multi-language
scenarios
were
able
to
offer
such
share
capabilities
and
up
these
two
things
down.
Ally
group
infrastructure
is
becoming
more
open
and
the
in
open
field
in
our
community
that
has
a
back
integration.
There
are
two
benefits:
the
first
one
and
the
more
open
technologies
could
be
localized
in
Ally
group,
and
we
can
try
some
of
the
mountain
technologies
and
to
improve
the
other
groups
infrastructure.
B
Today,
if
we
want
to
do
the
cloud
native
and
the
we
can
I
start
with
the
kubernetes
to
do
your
cloud
native.
There
are
two
reasons
behind
it.
First
of
all,
kubernetes
has
become
the
core
of
cloud
native,
and
the
differentiation
of
the
resources
is
minimized.
No
manual
use
a
different
kind
of
cloud.
You
are
able
to
find
a
kubernetes
service.
B
On
the
other
hand,
kubernetes
has
a
lot
of
a
business
to
the
appellant
and
the
such
answer:
AI,
machine
learning
and
the
mini
kind
of
ecosystems
in
kubernetes
under
kubernetes,
and
it
has
become
a
colossus
and
the
second
secret
of
kubernetes
is
also
platform
for
platform
builders
and
in
it
is
kind
of
platform.
Where
that
you
can
expand
yourself
by
kubernetes,
you
can
build
up
a
more
sound
and
complete
platform
based
on
kubernetes
and
a
platforms
are
catered
to
your
business.
B
That's
why
we
want
you
to
start
your
cloud
native,
frank,
kubernetes
and,
of
course,
why
any
group
a
dot
to
the
natives.
We
encountered
many
challenges.
The
first
thing.
We
think
that
could
not
is
used
to
be
very
young
project,
whether
it
is
able
to
hold
a
large
volume
business
of
valley
group,
we're
so
diversified
the
business
scenarios,
especially
shopping
festival
and
about
11
and
the
peak
usage,
and
we
make
some
exploration,
experiment
improvement
on
kubernetes.
B
We
want
to
spread
the
to
do
upstream
and
I
hope
that
sir
and
we
want
to
enjoy
the
benefits
of
the
technology
together
with
Twitter,
and
we
want
to
have
a
coal
construction
and
then
we
can
share
this
benefits
to
more
communities
and
the
stability
stability
is
something
ongoing
process.
This
is
a
highly
sophisticated,
and
then
we
try
to
optimize
it
this
one
and
the
next
day.
B
B
We
have
our
own
dispatcher
from
the
within
the
user
fertilize
the
way,
and
we
have
our
own
key
to
be
dispatcher
and
of
course,
and
though
we
will
also
use
that
in
our
people
to
justify
the
title
and
of
course,
but
notice
can
work
very
well
and
I.
Think
that
could
market
is
able
to
support.
Most
of
the
scenario
worldwide,
so
scale
scalability
is
no
longer
a
problem
to
kubernetes.
B
The
second
thing,
I
want
to
say
at
allé
through
to
what
we
are
exploring,
is
the
how
to
transfer
it
into
kubernetes
system
and
at
our
group
that
this
is
not
something
done
overnight.
We
have
experienced
the
several
generations
of
architectures
at
the
beginning.
We
only
use
the
kubernetes
to
support
container
resources
at
Ali.
Our
group
we
have,
we
have
a
system
container.
An
assistant
container
is
more
like
a
virtual
machine
and
the
path
assistant
on
the
upstream
user
VM
to
use
a
system,
a
convener
and
the
signal
a.
A
B
We
hope
that
so
we
want
to
introduce
two
bananas
and
we
want
to
have
these
proven
artist
resources
to
it,
and
it
will
automatically
manage
this
assistant
container
in
order
to
incorporate
well
with
the
cloud
native
and
the
two
are
simplified
and
the
lightweight
our
cistern,
and
we
use
a
lightweight
container
into
a
port
and,
of
course,
in
task
system.
By
automatic
way,
we
can
put
it
into
kubernetes,
so
that
meant
making
a
system
becomes
very
cloud
native.
A
B
Now
I
talked
about
is
a
scalability.
It
is
very
good,
but
one
of
the
things
one
of
the
biggest
challenge
is
the
maintenance,
as
well
as
it's
a
ambience
operation.
We
have
a
lot
of
Aqaba
natives
cluster
management
and
on
the
platform
we
have
more
and
how
could
ensure
stability
of
the
kubernetes
cluster.
This
is
very
important.
It
is
something
we
want
to
talk
about.
The
first
thing,
I
want
to
say
how
to
manage
kubernetes
master
and,
of
course,
we
have
some
very
innovative
way.
B
That
is
the
kubernetes
alternative
and
the
first
of
all,
we
have
a
kubernetes
cluster
and
then
those
non-starters
knows
we
can
manage
these
notes
by
proven
artist,
automation,
and
that
is
two
common
artists
manage
kubernetes
to
stabilize
the
users
onto
the
negative
signal.
A
kubernetes
is
run
on
the
workers
and
it's
the
service
without
influence
the
service,
as
we
must
to
enhance
our
management
over
these
functions.
You
need
to
be
very
cautious
in
within
a
group
in
upgrading
node
number,
we're
also
scaling
up
our
scalability
and
though
a
better
way
to
manage
them.
B
Certainly,
we
try
to
have
a
stadium
and
Static
and
productive
and
the
new
version
of
kubernetes
tested
in
a
new
environment
apply.
You
need
to
stress
out
the
cluster
in
the
real
working
scenario
and
then
promoted
to
productive
environment
and
I.
Think
that
such
cool
way
and
that
can
give
you
one
layer
of
prevention
to
make
an
easier
and
civilized.
B
Angelina
and
that
there
are
several
good
questions
and
the
first
one
is
that
from
new
areas
from
the
these
new
areas,
and
there
is
no
burden,
historical
burden
and
such
as
AI
machine
learning
or
something
we
have
done,
and
that
is
a
mini
service
architecture,
controller
architecture,
all
of
which
a
very
good
opportunity
for
us
to
train
a
cloud
native
and
the
many
companies
are
now
transitioning
to
the
cloud,
and
this
is
the
chance
for
you
to
become
cloud
native
in
such
a
position.
You
need
to
upgrade
your
architecture.
B
Why
not
make
it
even
better
by
the
cloud
native?
Suddenly,
we
need
to
find
out
the
pin
point
to
try
to
promote
the
cloud
native
in
early
groups
and
in
some
large
volume
and
resource
allocations
very
important,
but
to
some
startup
companies.
Smaller
companies,
developer
happiness,
developer
happiness
is
one
of
the
reasons
we
need
to
use
developer
efficiency
to
promote
a
cloud
cloud
native
for
transition,
and
then
your
maintenance
and
operation
system
will
become
a
more
robust
from
code.
Writing
to
the
full
production
will
be
shortened
so
that
you
can
make
your
developer
happier.
A
B
If
you
already
use
kubernetes
or
cloud
may
make
it,
you
can
share
more
your
experience,
ally
group
like
ours.
We
keep
here
cloud
native
and
the
dis
communities
are
back
up
and
to
localize
the
cloud
native
at
your
own
site.
I
want
to
say
coop
notice
and
a
cloud
native
has
been
verified
and
oppressed
in
the
tribe
in
any
group,
and
it
brings
a
lot
of
benefits
through
our
company,
and
we
also
believe
that
cloud
native
or
our
kubernetes
could
could
work
for
everyone.
B
It
now
work
for
any
group
I
believe
it
is
able
to
bring
the
convenience
and
benefits
to
all
of
you.
Please
try
coronet
kubernetes
and
cloud
native
now.
I
want
to
talk
about
some
contribution
at
the
source,
open
source
and
we're
going
to
have
the
new
version
and
the
eg
CDs
we
put
forward,
and
so
we
have
a
lot
of
developer
as
our
our
partners,
as
I've
shown
in
the
slide
in
release.
We
have
two
hundred
improvements
and
we
focus
on
the
performance
of
etcd
such
as
storing
writing
reading,
especially
as
include
critical
scenario.
A
B
Container
and
it's
about
the
containers-
realization
container
d
is
one
of
the
earlier
containers
and
the
most
author
companies
haven't
habit
and
in
order
to
sanitize
it
and
within
our
company
we
try
to
have
the
control
D
and
at
present
we
already
have
the
continuity
in
operation
at
the
fifty
thousand.
A
note
and
I
believe
this
is
the
largest
our
continuity,
our
deployment
and
the.
Surely,
with
this
improvement,
we
have
over
one
hundred
improvement
as
well
and,
of
course,
and
we
have
our
own
open
source
projects.
B
Dragonfly
contributed
a
to
C&C
aqueous,
absolutely
production
users
use
it
and
recently
and
we
make
it
from
Java
to
go
language
and
the
for
better
contribution
to
the
community.
Apart
from
these
things,
I
are
already
mentioned
at
kubernetes.
We
also
try
something
very
innovative.
Lee
I
want
to
mention
one
that
is
a
kubernetes
multi
tendency.
As
known
to
all,
and
probably
you
will
arm,
you
will
have
a
lot
of
capabilities,
but
there
are
some
challenges.
B
First
of
all,
the
tenant
as
you
sharing
the
same
ap
server,
some
performance
problems
and
API
does
not
offer
their
strong
isolation.
Some
global
results
difficult
for
you
to
share
across
containers
and
the
you
cannot
secure
a
the
namespace
at
you
all
wish.
We
want
to
offer
a
multi
tendency
capabilities.
Every
user
can
have
virtualized
the
foster.
A
To
provide
my
sir
independent
API
server
api
server
and
just
like
to
use
the
local
coordinators
and
to
create
master
level
resource-
and
here
are
several
challenges,
for
instance,
at
a
button
at
a
base
actually
for
the
resource,
we'd
like
to
share
them
in
order
to
improve
efficiency
and
other
effects,
and
therefore
we
use
the
communities
architecture.
And
how
can
we
ensure
that
for
the
users
they
can
create
a
synergy
with
the
base?
A
Actually,
we
have
a
super
named
and
to
do
the
same
synergy
and
actually
to
do
multi-tenancy
based
on
kubernetes
is
quite
a
challenge
and
we
have
the
veer
agent
function
and
in
order
to
enable
the
county
for
communications
between
kubernetes
and
attendances,
and
absolutely
we
have
down
the
PLC
within
Alibaba
and
also
we
have
move
it
upstream
and
also
it
has
steady
approval
of
the
motor
tenancy
and
also
in
the
future.
We
will
open
our
code
to
upstream,
and
we
welcome
all
of
you
to
participate
in
the
build
of
Redis
faction
and
also
for
this.
A
Well.
This
is
another
project
urban
cruise.
This
is
a
open
source
project
as
well,
and
actually
it
galvanized
the
alibaba's
best
practice
in
the
recent
years.
The
open
course
it's
its
goal
is
to
automate
everything
on
kubernetes
and
wait
about
to
provide
automation,
platform
and
also
tools
and
I,
enable
people
to
better
construct
their
own
platform
on
kubernetes
and
also
for
the
workload
part.
A
We
have
already
do
the
open
source
and
we'd
like
to
provide
a
better
workout
management
capability
and
English
for
you
to
manage
the
the
kubernetes
and
other
other
clusters,
and
you
can
have
you-
can
scan
a
QR
code
and
to
see
our
github
page,
and
here
we
also
have
a
rich
at
school
and
you
can
pay
and
actually
the
password
for
days
of
which
recruits
here,
zero,
six,
two
six.
So
if
we
are
interested
in,
you
can
join
our
wait.
Red
group
and
absolutely
have
three
different
capabilities.
Alarm
is
the
broadcast.
A
It
is
a
great
help
and
why
you
want
to
do
the
cleanup
and
also
operation
installation
every
node
same
can
use
a
broadcast
drop
to
enable
you,
and
also
we
have
the
site
concept
to
manage
service
mess
and
also
to
manage
such
containers
at
scale,
and
actually
it's
the
couple,
the
management
and
also
decouple
the
lifecycle,
and
to
really
maintain
the
advantage
over
server
smash
and
also
have
the
advanced
estate
or
set.
This
is
a
controller
and
it
is
a
hazard
to
major
improvements.
The
first
is
in
order
to
increase
the
stability
of
deployment.
A
We
have
the
influence
of
updates,
so
you
don't
have
to
create
a
new
part
every
time.
So
here
you
can
upgrade
apart
locally
and
you
just
can
relate
lower
the
border
of
scheduler
and
also
why
you
do
the
upgrading
of
a
code
and
various
the
process.
There
is
a
sequence
and
also
honors
certain
scenarios.
This
difference
is
really
useful.
A
Before
at
some
other
scenarios,
we
can
really
manage
the
process
and
therefore,
in
the
advance,
to
state
process,
we
have
the
maximum
available
support
and
also
to
to
enable
to
have
a
better
product
management
and
also
equation,
and
we
have
open
source
the
three
functionalities
and
also
in
the
future.
We
will
open
source
for
more
and
more
functions
from
Alibaba
and
the
way
delight
to
welcome
you
on
board
and
we'd
like
to
know
what
our
developers
want
and
to
keep
to
to
keep
pace
with
the
development
of
community.
A
So
I'd
like
to
color
for
all
of
you
to
be
on
board
with
us
and
for
our
technology
development.
We
also
have
made
a
lot
of
contributions
to
the
community
and,
together
with
themselves,
we
have
opened
a
online
course
for
cloud
native.
It
is
a
free
of
charge
and
will
galvanize
more
than
20
professionals
in
cloud
native
and
also
it
is
aligned
with
our
stake,
a
a
test
and
after
learning,
of
course,
you
can
pass.
A
For
instance,
this
year
we
have
been
in
four
cities
and
we
have
eight
hundred
attendees
and
also
a
half
by
eleven
thousand
more
than
eleven
thousand
OnLive
years,
and
we
have
many
interesting
topics
in
all
these
meters
and
apart
from
meetups,
we
also
have
write
some
articles.
For
instance,
every
week
we
have
a
weekly
report
and
also
article,
and
we
will
galvanize
all
the
interesting
news
and
also
informations
about
the
latest
cognitive
technologies
in
this
crowd,
native,
weekly
and
I.
A
Welcome
all
of
you
to
look
to
find
us
in
deaf
hub
and
also
you
can
contribute
to
coordinate
of
weekly
as
well
and
from
alibaba's
perspective.
We
believe
that
resonators
needs
everyone
and
everyone
needs
to
participate
in
this
journey,
to
devote
a
cult,
to
define
and
shape
the
feature
of
a
cognitive
and
also
to
grow
the
cloud
Native
community,
not
only.