►
From YouTube: Let Developers and Operators Focus on What They Know Best - Rafal Gajdulewicz & Julie Zhuying 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
Let Developers and Operators Focus on What They Know Best - Rafal Gajdulewicz & Julie Zhuying Li, Google
Kubernetes has steep learning curve for application developers and operators to learn new concepts to manage multiple tools and artifacts to build deployment pipelines. This problem is worsened by the dynamic and heterogeneous nature of cloud and hybrid environments, where components constantly get upgraded, scaled and destroyed. In this talk we would like to demonstrate a workflow letting each role focus on what they do best by separating app and ops artifacts and versioning them in separate repositories. This separation allows CI/CD tools to effectively keep the desired and actual state in sync while simplifying complex processes required when delivering multiple applications. Finally, we will show how the OSS Cluster Registry can be used to manage deployment and experimentation when managing multiple Kubernetes clusters in different environments.
To learn more click here: https://sched.co/FuJY
A
Hello,
everyone.
Thank
you
so
much.
My
name
is
Julie
I'm,
the
interaction
designer
at
Google
I'm
here
today
with
my
colleague
Rafael,
who
is
a
software
engineer
and
we
both
work
on
Google
kubernetes
engine.
Today
we
are
going
to
talk
about
our
approach
for
faster
and
more
reliable,
tormund
and
release
process.
This
is
designed
to
streamline
developers
and
operators
workflow
by
integrating
some
of
the
monster
frequently
Institute's
available
during
this
talk,
we're
going
to
take
for
the
problem
space
commonly
adopted
industry
tools
and
our
proposed
the
solution
with
a
demo.
A
For
the
system,
and
so
you
can
improve
in
monitoring
and
telemetry
for
operators,
this
includes
testing
applications
in
different
environments,
monitoring
the
health
and
security
of
the
application
and
implementing
necessary
upgrades
and
updates.
Please
note
hands
on
the
structure
of
your
organization.
These
tasks
may
be
performed
by
one
or
multiple
engineers,
and
for
this
presentation,
that's
assuming
this
tasks
are
worth
bankers
and
operator
roles.
A
So
that's
now,
let's
look
at
where
these
tasks
are
performed
in
a
general
development
and
release.
Workflow,
first
developer,
right
codes
in
their
local
environment
have
a
reviewed
and
committed
to
the
version
control
repository
next.
They
requested
employment
and
past
environment
and
handed
over
to
the
auditors.
A
Operators
fantasize
go
to
what
is
to
be
the
point
and
where
to
toil
to
then
they
build
the
deployment
and
publish
the
artifact
after
which
they
modified
the
kubernetes,
which
was
manifest
and
apply
the
changes
to
the
test
environment.
Once
deployment
is
done,
operators
will
notify
the
developers
and
developers
then
verify
the
deployment,
perform
manual,
testing
and
correct
any
arrows
in
their
code
configurations
and
performance.
Pacification
after
a
release
is
ready.
Developers
request
a
deployment
to
production
than
handed
back
to
the
operators.
A
Our
preachers
then
repeat
the
process
as
deploying
to
the
testing
environment.
They
decide
what
is
to
deploy
and
where
to
deploy
to
and
then
build
and
published
artifacts
and
finally
deploy
all
the
artifacts
to
the
production
environment.
This
is
both
laborious
and
tedious
process.
As
you
can
see,
all
the
core
tasks
are
only
a
very
small
percentage
of
the
entire
flow
which
God
our
thinking.
C
Thank
you
reading,
okay,
so
do
we
describe
the
workflow
and
noted
that
it
could
be
simplified,
but
how
can
we
do
that?
The
cognitive
echo
system
is
pretty
complex,
as
you
can
see,
and
for
most
problems,
there's
more
than
one
tool
that
you
could
use
to
solve
it.
The
solution
would
like
to
describe
today.
You
realize
some
tools
that
you
should
be
familiar
with,
or
even
using
a
daily
workflow,
so
it
should
be
possible
for
you
to
easily
gradually
adapt.
C
It
will
show
you
how
you
can
use
version
control
systems
to
magic,
your
deployments
and
be
the
source
of
true
for
deployment
state.
How
do
you
see
I
see
details
to
reduce
the
number
of
manual
steps?
I
will
introduce
a
new
tool,
open-source
cluster
registry
that
can
help
you
manage
between
the
mapping
between
applications,
stages
and
clusters
and
a
similar
declarative
way,
I'll
be
using
gates
in
the
demo
later
today.
C
But
the
general
pattern
works
with
any
other
version
control
system,
we'll
use
it
for
storage
for
application
coding
configuration
it
should
be
familiar
and
also
for
2090.
First,
the
files
used
to
deploy
your
app
I
don't
want
to
get
into
the
single
versus
multiple
repos
discussion
here.
So
I
assume
applications
have
their
own
rep
or
oppose,
but
we'll
have
a
search
info
repo
for
manifests.
C
Then
we'll
use
the
CI
CD
job.
So
there
are
many
of
those
with
various
degrees
of
integration
of
kubernetes,
but
the
features
I
was
seeing
today
are
on
being
a
like:
the
possibility
to
define
a
custom
tags
things
bash
script
and
a
possibility
trigger
tasks
from
a
good
comet
which
all
of
those
tools
you
provide.
C
Okay
and
now
I
have
questions
so
how
many
of
you
use
more
than
one
kubernetes
cluster
in
a
date
that
they
work,
okay,
fu
and
how
many
of
you
are
familiar
of
class.
The
registry?
That's
due!
Ok,
thanks,
so
I
will
introduce
so
class
the
registries
and
kubernetes
open
source
project.
It's
basically
I
create
these
custom
resource
that
you
can
install
and
use
to
track
to
the
list
of
the
clusters.
You're
interested
in
it
starts
in
for
information
about
the
cluster
itself,
the
API
server
there
are
certificates
and
it
has
all
the
standard
communities
metadata.
C
So
you
can
use
labels
to
query
and
filter
the
cluster
engine
you're
interested
in
it
also
lets
you
define
custom
operators
and
controllers,
but
will
not
talk
about
this
here.
So
cluster
registry
helps
us
a
few
things
one
is
we
don't
need
to
know
the
mapping
between
application
environments
and
clusters
by
heart
read
the
docs
or
try
to
infer
it
from
other
sources.
We
can
rely
on
tauriel
registry
starting
it.
C
Another
is
that
the
cluster
registry
is
the
source
of
truth
that
lets
operators
make
changes
to
environments
without
the
need
to
update
individual
app,
config
files
or
in
front
of
elders.
Imagine
unit
from
tomorrow.
You
have
to
run
your
products
on
two
clusters
or
you
want
to
replace
the
disk
luster
with
a
new
version.
How
many
things
you
do
have
to
change
in
your
current
code,
so
yeah.
So,
let's
look
at
the
workflow
Julie
presented
before
I
highlighted
some
of
the
steps
that
could
be
automated
or
could
benefit
from
integration
with
tools
I
mentioned.
C
C
Can
we
simplify
it
using
the
by
first
following
a
convention
that
map's
the
source
code
versions
to
artifact
versions
and
to
by
using
craft
registry,
and
my
demo
I
will
use
git
hashes
to
tag
my
artifacts,
but
any
versioning
scheme
works
as
long
as
you
can
maintain
relation
between
source
and
artifact
built
can
then
happen
automatically,
as
you
imagine,
and
the
deployments
are
triggered
by
comets
to
the
entire
Evo.
I'll
show
this
in
the
and
details
later
now.
C
Cluster
registry
can
also
help
us
in
a
few
things,
so
first
yeah
it
tells
us
what
to
deploy
and
where
like.
If
we
have
a
version
which
pastures
to
the
PI
to
and
then
we
can
get
the
details
of
the
pastures
that
we
acquire
to
do
the
actual
deploy.
And
then
there
is
one
more
step
and
we
included,
but
you
shouldn't
be
doing
this,
but
if
you
want
to
follow
immutable
deployments,
which
are
a
good
idea,
you
shouldn't
be
rebuilding
your
app
between
the
plans,
two
different
stages.
C
D
D
E
D
C
So
I
provided
a
quick
demo
to
you,
have
to
combine
the
tools
and
conventions
mentioned
before
to
optimize
your
deployment
and
development.
Workflow
I
wanted
to
make
sure
that
the
demo
works
without
connectivity
to
the
cloud.
So
all
the
things
I'll
show
you
run
locally
and
I
will
use
a
single
cast.
Everything
just
be
environment,
free
local,
git,
repos
and
I'll.
Do
the
CI
CD
steps
locally
using
an
excel,
so
we
will
take
a
look
at
a
simple
microblogging
application
that
lets
you
post
serve
messages.
C
The
app
has
two
components:
a
front-end
which
shows
the
list
of
method,
messages
in
HTML
and
back-end
that
provide
the
storage
and
query
API
both
and
kubernetes,
and
shared
three
main
stages
of
the
rope
and
pipeline.
So
we'll
have
a
local
environment.
When
we
go
from
code
we
build
the
code
and
run
it
and
then
fading
and
fraud,
which
is
pre-built
and
periodic
containers
which
do
a
container
registry.
C
E
C
You
see
I
included
a
text
that
identifies
the
environment
and
also
the
sidebar
will
change
colors
when
we
go
to
staging
thanks
for
10.
So
it's
clear
well
which
environment
I'm
using
okay.
So
let's
go
through
a
simple
change.
Workflow
so
we'll
add
a
feature
where
we
show
the
timestamp.
The
time
has
it
that
has
elapsed
since
the
message
were
created,
so
our
be
a
back-end
team
already
did
that
and
they
include
the
timestamp
I
want
to
make
it
to
change
the
front-end,
to
include
a
message
like
that
in
this
screenshot.
C
I'll
have
to
be
the
CEO
since
I
this
is
all
offline.
So
let's
pretend
then
all
commands
that
start
with
CI
are
run
on
a
CIC
ito.
So
first
we
build
the
artifact,
so
I
built
the
docker
image
and
push
it
to
the
registry,
and
then
we
trigger
the
staging
deployment.
So
I
find
the
version
that
needs
to
be
applied.
C
C
And
switched
abroad,
and
what
happens
is
that
we
have
an
error
like
something
is
wrong:
I'm
I'm,
not
an
operator
now
so
I,
don't
I'm
not
trying
to
debug
the
thing
in
production.
I'll
do
what
the
first
reasonable
thing
so
I'll,
just
roll
back
the
deployment
and
we'll
try
to
fix
everything
later
Robeck.
C
How
do
you,
how
do
I
roll
back?
How
do
I
know
what
to
deploy?
Now?
It's
actually
easy,
because
we
have
an
infrared
bow
that
stores
all
the
things
that
happened
to
all
environments.
So
if
we
look
at
the
log,
I
updated
the
staging
to
the
new
version,
then
I
updated
brought
to
the
new
version
and
then
I
just
did
get
reverse
commands
and
applied
the
manifest
and
now
I'm
at
the
previous
production
version
that
was
known
for
work.
C
So
what
can
we
do
that
with
you
before
we
mentioned
that
automating,
your
workflow?
Will
save
you
time
and
you
can
spend
the
time
improving
other
aspects
of
your
product.
So,
as
you
know,
it's
critical
that
your
production.
As
for
your
errors,
so
how
can
you
try
to
deploy
a
new
version
of
the
app
and
test
it
in
production
while
minimizing
the
impact
of
the
box,
so
that
less
people
see
the
box
or,
and
one
of
the
methods
of
solving
this
problem
is
called
Canarian,
where
you
deploy?
C
The
new
version
to
a
small
subset
of
production
servers,
verify
the
impact
of
the
change
it,
whether
it's
metrics
manual,
testing
or
other
signal
we
can
use
to
see
if
the
new
versions
working
and
then
we
decide
whether
to
proceed
with
the
rollout
to
all.
Servers
are
to
rollback,
and
this
should
be
especially
if
your
rollouts
take
longer
than
what
I
showed
here.
This
saves
you
valuable
time
when
everything
is
broken,
so
let's
try
adding
a
canary
to
our
work.
D
D
C
And
try
canary
likes
ya.
So
again
we
all
try
to
promote
a
brand
known
wrong
version
from
staging
the
canary,
but
that
is
you.
Should
you
should
now
see
how
this
works
in
production?
How
we're
all
of
a
new
version
that
might
be
broken
works
so
as
its
we
see,
that
prod
is
fine.
Most
of
the
requests
we'll
get
product,
so
they'll
be
fine,
but
they
request
the
fit
the
canary
are
still
broken.
Okay,
can
I
let's
kill
the
cannery
and
there's
not
much
more.
You
can
do
as
an
operator.
C
We
just
need
to
let
the
developers
figure
this
out.
So
now,
as
human
a
developer
comes
back
to
the
office
after
a
conference
and
I
learned
that
there
was
a
terrible
bug
that
happened
and
I,
don't
really
know
what
was
the
state
of
the
environment
and
what
happened.
So,
how
do?
How
do
I
know
reproduce
the
back
in
my
local
environment,
because
that's
where
I'm
the
most
effective
without
report?
Oh,
that's,
easy!
I,
all
the
changes,
even
the
ones
that
happened
after
we
did.
C
F
D
C
We
can
run
yeah
using
any
tools,
you're
familiar
with
and
should
be
easy
to
resist
it
and
see
what
happens.
Okay,
we
see
the
error.
Let's
check
out
what
happens
with
the
back
end.
Okay,
so
it
looks
like
the
back
end.
Team
didn't
apply
the
change
that
they
prepared
to
the
pradhan
firemen
and
it's
still
not
there
now,
so
we
can
either
wait
for
them
to
like
stop
all
releases
of
the
front
end
until
they
release,
or
we
can
add
a
we
can
take
this
on
our
end-
to
support
both
versions
of
the
back
end.
C
I
can't
is
still
not
returned,
we
have,
but
the
fronton
works
so
looks
good.
Will
it
work
28
with
the
new
back-end?
So
let's
get
the
last
version.
We
have
ok,
looks
like
AdWords
the
back
and
returns
a
date.
We
don't
see
the
sorry
a
very
refreshing,
so
we
should
see
the
date
a
video.
Let
me
hard
reset.
D
D
D
C
So
those
wouldn't
be
there
if
you
properly
pushed
your
changes
to
some
upstream
but
I'm
just
doing
everything
locally,
and
now
we
see
that
fraud
works
and
canary
also
works,
but
the
back
end
is
still
not
updated.
So
let's
pretend
that
our
friendly
back-end
team
deployed
their
changes
and
we
can
safely
check
if,
if
the
canary
now
should
work
well
in
display
the
date
or
the
crowd
will
not
work
and
not
I
mean
will
still
work
but
not
display
tonight.
So
my
fraud.
C
C
Okay
hope
rods
and
cannery
come
on.
Cannery
have
the
new
version
of
the
code.
So,
as
you
see,
I
hope
this
book,
we
found
it
useful
and
the
idea
we
were
able
to
follow
along
because
I,
probably
a
bit
too
quickly
better.
As
you
see
the
structure
you
proposed
allowed
us
to
automate
some
Republika
double
tasks:
moving
from
static
environment,
clamping
to
declarative,
easy
range
tools.
C
It
also
allowed
the
operator
to
add
a
new
deployments
that
without
changing
the
app
in
any
way
they
could,
the
operators
can
change
the
infrastructure
and
the
entire
eco
and
the
developers
don't
have
to
make
any
changes
the
application
code.
And,
finally,
it
also
made
it
much
simpler
for
developers
to
refer
this
issues
because
by
mismatch
state
in
the
remote
environment,
which
is
usually
quite
tricky
so
now
I'll.
Let
Julie
summarize
Oh.
D
F
D
D
B
D
D
A
D
F
C
When
I
built
the
docker
image
I,
just
look
at
the
gate
comment
a
little
bit
from
and
I
Ducker
packet,
the
drainage
I'm.
Just
using
this
for
some
publicity
you
could
follow
a
versioning
scheme.
Like
you
know,
you
can
use
yeah
pretty
much
any
scheme
for
as
long
as
it.
For
example,
you
can.
If
you
want
to
do
something
else,
as
your
artifact
version,
you
can
tag
your
gift
comets
with
the
version
you
assigned.
C
C
D
D
B
Either
for
those
reticulum
and
faster
than
I
thought,
I
would
show
you
cfop
switch
on
studio,
slow
cooling,
shade
three
Imperial
short
other
quadrant
without
actually
George
putting
down
wedgie
dancer.
So
his
room
and
take
a
lesson
it
was
just
devastated.
I
didn't
push
her
Thomas
a
that
I'd
seen
him
father,
so
I
don't
see
change
to
that
user.
They
were
watching
ficus,
impermanent
Eliza,
my
job
I'm
against
Marcy
harder,
their
internal
change.
Second
issue:
that
was,
she
would
see.
B
D
E
C
I
mean
that's
right,
yeah
of
course,
so
again,
I
I
wanted
to
make
sure
this
works.
Luckily,
so
in
here
I
build
it
luckily,
but
you
could
imagine
that
the
CI
thing,
the
CI
a
build
step
is
run
on
a
and
then
we
have
a
Google
cloud
builder.
There
is
a
number
of
tools
that
build
your
docker
images
remotely.
So
this
could
happen
work
there
and
then,
when
it,
the
process
is
done.