►
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: A Journey to a Centralized, Globally Distributed Platform – Katie Gamanji, Cloud Platform Engineer, Condé Nast International
For over a century Condé Nast International has set the benchmark for print and digital publishing. With brands like Vogue, GQ, Wired, Condé Nast Traveller under our umbrella we are operating in more than 12 markets across different geographies, including Russia and China. Whilst serving content to million of customers daily (220m unique users per month), it has become paramount to have a scalable, fault tolerant and highly available infrastructure, combined with a robust delivery process. Originating with a fragment technological ladscape, Condé Nast International is progressing with a remarkable project to embrace cloud native principles and deliver a centralized, globally distributed platform that will further emancipate our international teams.
This talk aims to highlight Condé Nast International's journey to deliver a multi-cluster distributed Kubernetes platform, with a centralized management mechanism and self-service CI/CD process.
https://sched.co/MRz4
A
Hello
coupe
con.
My
name
is
Katie
Comanche
and
I'm.
A
cloud
platform
engineer
for
Conde
Nast,
International
I
couldn't
be
more
precise,
but
one
year
ago
this
day,
I
joined
Conde
Nast
to
contribute
to
a
project
that
aims
to
create
a
centralized
globally
distributed
platform.
Today,
I'm
going
to
talk
you
through
how
exactly
we
achieve
this
Conde
Nast
international
has
set
the
benchmark
for
print
and
digital
publishing
for
more
than
110
years.
A
We
operate
brands
that
are
fashion
and
luxury
oriented,
and
you
might
be
familiar
with
some
of
the
brands
like
vogue,
GQ,
Wired,
Vanity,
Fair
ad
and
many
more.
These
are
just
a
few
of
the
brands
we
have
in
our
portfolio
with
the
headquarters
in
London.
We
operate
on
an
international
scale
in
1
in
11
countries
we
have
62
web
sites
overall
and
they
generate
300
million
unique
users
every
single
month.
So
far,
we've
raised
a
total
of
1.5
billion
page
views
every
single
month.
Thinking
about
this
there's
gonna
be
700.
A
570
articles
viewed
every
single
second,
as
you
can
see,
this
is
quite
a
considerable
customer
base.
We
have
out
them
so
how
exactly
the
journey
towards
a
centralized
platform
begun
in
the
past,
the
global
market
teams
acted
like
independent
business
units
at
the
time.
This
was
the
right
thing
to
do,
because
it
provided
the
right
environment
for
growth
and
development.
However,
from
an
engineering
point
of
view,
with
a
with
evolution
of
different
deployment
mechanisms,
it
was
clear
that
the
markets
have
quite
a
centralized,
centralized,
individualized
and
personalized
tech
stack.
A
A
As
such,
two
years
ago,
Conde
Nast
International
decided
to
roll
out
a
unified
platform
across
all
regions
while
embracing
the
cloud
native
principles.
So
how
exactly
does
our
tech
stack?
Look
like
right
now,
all
the
websites
hosted
on
our
platform
are
going
to
be
fronted
by
fastly
fastly
is
our
CDN
provider
and
having
a
CDN
provider
allows
us
to
reduce
the
load
on
our
back-end
systems
and
optimize?
The
time
to
deliver
the
content
to
our
users,
supposedly,
we
have
an
editor
that
has
written
an
article
that
has
not
even
been
cached
by
our
CDN.
A
In
this
case
an
X
cache,
miss
response,
header
is
going
to
be
returned
and
the
request
is
going
to
be
forwarded
to
traffic
traffic.
Is
our
ingress
controller
that
manages
the
external
access
to
the
services
within
our
communities
clusters?
In
the
current
setup
we
have
to
load
balancers
per
namespace,
one
for
public
and
one
for
restricted
access.
Once
traffic
has
an
ingress
endpoint,
the
request
is
going
to
be
forwarded
to
the
global
web,
app,
which
is
a
multi-tenant
application.
It
will
be
able
to
get
the
desired
content
and
deliver
to
the
user.
A
A
We
are
a
kauravas
house,
meaning
that
we're
using
tecktonik
installer
for
our
cluster
deployment
and
using
EDD
for
our
storage
of
cluster
data
using
tectonic
allows
us
to
have
a
self-hosted
control
plane
for
kubernetes,
while
plugging
in
into
different
cloud
providers
to
consume
the
compute
and
networking
resources.
As
such,
we
avoid
vendor
lock-in
while
deploying
self-hosted
multi
master
multi,
node
100%,
open
source
kubernetes.
At
this
point
in
time,
we
operate
more
than
100
instances
in
AWS,
which
is
our
cloud
provider
by
choice.
A
We
make
use
of
auto
scaling
groups
to
make
sure
that
we
scale
and
down
based
on
interaction
of
our
customers,
with
our
brands
all
destroyed
is
the
infrastructure
here
is
mainly
deployed
using
terraform
having
infrastructure
as
code
at
the
bases,
looking
into
other
tools
that
we
have
in
our
cluster.
This
is
just
a
few
of
the
open-source
tool
that
you
use
to
provide
authentication,
logging
monitoring
and
many
more.
We
use
helm,
which
is
our
de-facto
deployment
package
manager
to
kubernetes.
A
Our
platform
is
live
since
October
last
year,
when
we
first
launched
Gigi
France.
Since
then,
we
launched
10
more
websites,
making
a
total
of
11
and
so
far
we
aim
to
migrate
34
websites.
At
this
point
in
time
we
serve
75
million
pageviews
every
single
month,
which
is
quite
impressive
for
a
platform
that
has
been
there
for
around
eight
months.
A
So
at
this
point
we
have
a
platform
that
is
centralized.
That
is
scalable.
Do
we
need
to
think
about
global
distribution?
The
answer
is
yes.
Looking
at
this
map,
we
can
see
that
we
operate
in
areas
which
are
quite
geographically
dispersed.
I
would
like
to
put
a
focus
on
China
and
Russia
from
this
two
countries
alone.
We
have
19
million
unique
users
every
single
month.
They
represent
17%
of
our
entire
digital
readership.
A
It's
quite
a
considerable
customer
base,
and
this
was
a
highlight
that
conde,
nast
international
should
be
prepared
technically
to
meet
the
demand
in
these
areas
as
well,
deploying
to
countries
like
Russia
and
China's.
What
lies
the
need
to
have
a
infrastructure
that
is
cloud
agnostic
when
the
project
for
a
unified
platform
was
kicked
off,
we
did
not
have
a
clear
vision
of
one
cloud
provider:
that's
going
to
be
used
for
all
platforms
and
all
regions.
Again,
this
enforces
our
decision
to
go
for
health
self
hosted,
given
a
tea
solution.
A
Another
reason
for
global
distribution
is
origin
latency
if
you'd
serve
all
the
content
from
one
origin
or
one
server.
It
was
set
of
servers
at
least,
and
the
cache
is
not
available.
Then
we'll
have
to
pull
the
content
from
our
servers
every
single
time.
This
will
introduce
a
latency
to
the
end
user,
and
this
is
not
something
we
aim
for
what
we
aim
for
is:
market
proximity
and
the
highest
user
experience
for
our
customers.
As
such,
it
was
a
business
decision
to
replicate
our
clusters
in
five
different
regions,
making
a
total
of
nine
clusters.
A
A
China
alone
has
two
dedicated
clusters
as
staging
and
a
production
one,
while
in
Dublin
we'll
have
two
more
clusters
to
host
our
internal
tools.
As
you
can
see
in
chat
well
in
Russia,
we
do
not
have
a
replica
of
our
tech
stack.
This
is
because
there
is
no
major
cloud
provider
that
we
can
use
to
do
so.
The
current
strategy
is
to
serve
the
content
from
our
European
servers,
while
having
a
CDN
which
is
fashion
oriented.
A
So
far
we
have
a
platform
that
is
globally
distributed.
We
need
to
think
about
how
our
developers
interact
with
it.
One
of
the
objectives
to
roll
out
a
unified
platform
was
the
fervor
empower
our
local
and
international
teams.
The
cloud
platform
team
believed
that,
having
the
self-service,
solid,
well-defined,
CIC
deep
process
is
a
good
step,
and
actually,
in
order
to
achieve
this
so
to
deploy
taki
bonitas
clusters,
the
teams
will
require
a
da
Christ
version
of
the
application.
This
is
going
to
be
stored
in
key
dot
IO,
which
is
our
docker
image
registry
by
choice.
A
Secondly,
the
team
will
be
provided
with
a
template
helm
chart
to
deploy
the
application
to
the
cluster.
They
will
be
able
to
provide
a
configuration
file
to
specify
settings
like
amount
of
replicas
the
amount
of
CPU
and
memory
they
require
as
well
if
they
want
to
scale
their
service
horizontally
by
using
an
HP
object.
A
Once
these
two
dependences
are
available,
a
circle
say
pipeline
is
going
to
be
triggered.
The
circle
sign
pipeline
is
again
fully
customizable
by
the
teams,
so
it's
quite
common
to
see
application
specific
workflows
within
it
once
the
application
is
live.
The
team
will
be
able
to
login
into
the
cluster
and
interact
with
it
at
this
point
in
time,
every
single
team
is
dedicated,
one
namespace
where
they
have
admin
privileges.
We
make
use
of
resource
quotas
to
make
sure
that
cases,
officers
or
virtualizations
are
not
happening.
A
Another
thing
I'd
like
to
mention
here
is
the
are
back
rows
in
the
current
setup,
everyone
have
read-only
privileges.
This
includes
the
dev
teams,
as
well
as
the
cloud
platform
or
the
admin
team
for
the
dev
teams.
They
will
be
able
to
see
all
the
resources
in
the
cluster
except
the
secrets
in
our
namespaces
or
the
core
components
for
kubernetes.
This
provides
a
perfect
incentive
for
Inter
team
collaboration
and
knowledge
sharing.
A
Going
for
global
transformation
is
not
easy.
It
comes
with
challenges.
I
would
like
to
focus
on
developer
experience,
conde,
nast
international
is
a
digital
first
company.
However,
we
take
the
developer
experience
as
a
key
indicator
of
how
we
build
and
deploy
our
platform.
We
always
try
to
optimize
and
improve
the
process
of
deploying
to
communities
as
well
as
the
way
the
developers
interact
with
the
cluster
service.
A
Continuity
is
something
else
we
have
to
deal
with
when
we
migrate
services
to
a
new
platform,
we
have
to
ensure
that
the
content
is
going
to
be
available
with
minimal
disruption
or
actually
no
disruption
to
our
users.
This
is
possible
for
us
by
using
varnish
configuration
files
which
allows
us
to
apply
changes
to
the
cache
with
immediate
effect.
A
Upgrades,
unfortunately,
are
still
another
pain
point
for
us.
This
is
because
we
have
a
self
hosted
solution
for
kubernetes.
We
do
not
have
an
automated
sophisticated
way
to
do
upgrades.
However.
We
always
try
to
improve
the
process
with
every
single
release.
I've
mentioned
China
and
Russia
and
I
would
like
to
mention
it
again.
These
countries
are
challenging
to
deploy
infrastructure
into,
however,
conde,
nast
international
has
a
strategy
to
approach
the
demand
from
these
two
areas
and
in
terms
of
fervour,
development,
we're
thinking
about
tracing
observability
and
service
mesh.
A
A
Indian
I'd
like
to
say
that
conde,
nast
is
truly
on
a
path
to
embrace
cloud
native
principles.
The
journey
is
far
from
completion,
however,
so
far
we've
managed
to
create
a
cluster
that
has
a
centralized
management
mechanism
and
is
globally
distributed.
However,
what
makes
this
platform
so
successful
is
the
fact
that
we
have
a
solid,
well-defined,
CI
CD
process
that
further
empowers
our
local
and
international
teams
and
their
product
out
like
take
this
opportunity
to
say
that
we're
hiring.
Please
talk
to
a
member
of
my
team,
which
is
in
the
audience
today,
myself
or
visit.