►
From YouTube: OM#6: Continuous Delivery with k8s - Tobias Schwab
Description
Tobias' bio:
Tobias Schwab is a Software Engineer with over 10 years of experience in Extreme Programming and Test Driven Development. Before co-founding PhraseApp he implemented Continuous Delivery Pipelines for projects such as Simfy, Wimdu, Wunderlist and Locafox.
Topic Summary:
"Don't we all want to deploy like Google does? We can! Making your application "container ready" is only the first step to deploy your application with containers. A far bigger concern is scheduling and orchestration of containers in a production deployment. Initiated by Google, Kubernetes is an open source container scheduler which leverages 15 years of experience running containers at scale. In this talk you will get a brief introduction to Kubernetes and learn how we use it to deploy PhraseApp. Also, I will share our best practices for production deployments with Kubernetes including topics such as * Building Docker Images * Configuration Management * Load Balancing * Deploying Microservices * Monitoring * Logging"
A
You
have
like
a
good
standard
to
to
to
build
your
application
from
so
to
get
started
with
Cuba
native,
and
this
is
ordered
by
piece
of
sweater.
Physically,
the
the
simples
way
to
get
started
is
definitely
a
google
complainer
engine.
It's
a
hosted
service.
Google
keeps
track
of
like
the
master
and
managers
professor
automatically
for
you,
you
don't
have
to
care
about,
updates,
etc
and
love.
A
nice
tool
to
get
up
locally
is
mini
cube.
A
It
actually
uses
a
bucher
docker,
which
is
the
way
to
run
dr
on
used
to
be
the
way
to
run
docker
on
expanding
windows
and
types
of
a
little
like
single
node
cluster
on
your
local
machine,
and
it's
really
nice
to
play
around
with.
There
is
a
neutral
cube
admin
which
is
part
of
cumin
and
14
and
which
is
a
really
convenient
tool
to
set
up
a
cluster,
a
master
and
and
to
join
notes
intellect
after
it's
much
simpler
than
we
did
it
before
and
at
the
very
bottom.
A
Is
cuban
embassy
hard
way,
which
is
the
github
repository
where
you
can
look
into
and
and
see
how
things
are
actually
working
under
the
hood?
It's
a
it's
a
repository
by
Kelsey
high
power,
and
you
can
basically
follow
all
the
steps
you
need
you
to
set
up
your
cluster
without
any
extra,
so
the
the
building
block
of
and
every
Cuba
native
class
of
when
it
comes
to
the
containers
and
what's
actually
running
in
a
cluster
of
hot.
A
A
So
dizzy
fur
off
and
the
feature
of
poplar,
they
all
have
a
dedicated
IP,
which
means
F
from
from
all
parts
again
effort,
all
the
other
pots
by
our
IP,
and
they
do
share
a
network
interface.
That
means
the
the
elastic
search
exporter.
44
elasticsearch
could
just
talk
to
the
local
interface
to
gather
information
from
Cuba
native
and
from
from
the
electrical
knew.
It
wouldn't
need
to
figure
out
where
the
the
actual
to
notice
running,
and
you
can
also
share
the
file
system,
which
means
they
don't
have
a
share
positive.
A
A
So,
and
the
next
thing
I
want
to
talk
about
is
services.
Services
are
eventually
policies
to
access
a
logical
set
of
pots.
What
does
that
mean?
And
you
don't
create
a
thermos
when
you
say:
okay,
these,
but
I
acted
the
furthest
and
these
are
removed.
But
if
such
the
labels
I
already
talked
about
attack
of
packs,
basically
which
define
the
service
and
for
that
to
visualize,
that
I
have
a
little
graphic
here,
and
these
are
the
these-
these
are
the
labels
I
introduced
with
the
pot.
A
A
Fish,
that's
not
good,
ok
here!
Is
it
too?
So
so
you
have
three
services.
The
two
are
quite
quite
picky,
these
two
one,
the
run
label
and
a
version
label,
and
so-
and
if
this
group
of
parts
basically
is
only
this
group
in
this
food-
is
this
group,
but
this
service
is
not
as
picky
doesn't
care
about
the
version,
so
both
of
these
groups
are
part
of
this
one.
So
this
is
like.
A
Here
is
a
manifest
similar
to
the
pot
manifest.
You
also
give
a
name
to
the
service
and
you
also
define
this
port
laughing
with
the
policy
I
talked
about,
so
in
that
case
the
the
hello
service
and
provide
some
port
80,
something
which
is
then
proxying
to
the
pots
in
their
logical
set
on
port
8080,
and
here
you
have
that
selector,
which
is
basically
the
all
thoughts
matching
this.
These
labels
are
part
of
the
service,
so
we
furnish
if
there's
also
a
nice
feature.
A
Also,
all
the
services
have
a
dedicated
IP
address,
so
you
could
talk
to
them,
and
these
IP
addresses
are
also
exposed
to
the
environment.
Then
you
start
a
fox
and
most
of
the
time
you
also
have
a
dns
server
is
running
your
classified.
It
means
you
can
access
key
services
by
name
so
from
from
all
the
parts
we
can
just
say,
curl
hello,
and
you
would
get
a
proxy
response
from
one
of
the
four
in
that
service
and
normally
that's
done
by
wrong
brother.
A
So
for
the
next
thing,
on
top
of
these,
two
core
concepts
are
deployment
and
or
that
there's
a
few
and
steps
in
between.
But
this
is
what
I'm
talking
about
today
and
one
of
the
deployment
and
I
already
said
that
the
pots
are
about
to
one
specific
node
and
what
I
didn't
say.
It
is
very
mutable.
So
in
theory
you
could
change
the
image
of
a
part,
but
you
shouldn't
do
that
and
and
to
do
a
deployment
from
an
application
perspective.
A
It
would
need
to
create
a
new
part,
then
wait
for
the
new
part
to
be
available
and
then
remove
the
old
pot.
So
that
could
be
something
you
could
do
already,
but
as
I
first
Cuban
I.
This
is
a
framework,
so
you
have
a
concept
of
deployment
so
and
in
the
deployment
everything's
done
for
you
and
like
you
want
it
to
be.
Basically
so,
and
the
appointments
are
a
way
to
declaratively
update
parts.
That
means
image,
which
means
you
want
to
deploy
a
new
revision.
A
That
means
changing
your
image,
because
you
always
put
your
version
in
the
image
or
the
environment
of
the
part.
That
means
changing
the
database.
Url
or
some
credentials,
or
something
like
that
configuration
files
and
also
labeled
so
behind
or
below
the
deployment
and
Cuba
net,
accuses
replica,
sets
and
wrap
the
Confessor
just
the
sixth
number
of
patras,
because
running
on
a
multiple
notes
in
your
process
and
as
I
said,
the
part
is
always
down
to
one
note
is
that
one
note
dies
and
the
part
will
never
come
back.
A
A
There
is
already
a
feature
building
through
benefits
which
are
contact,
map,
config,
Messer
way
to
store
basic
configuration.
Files
example
could
be
read
if
cones
and
genetic
on,
so
these
are
all
applications
which
cannot
be
configured
by
the
environment.
So
the
other
option
here
would
be
to
create
custom
images
way
of
getting
your
configuration.
So
with
that,
you
have
like
a
better
coupling
place,
kanokon
set
of
configurations
and
when
you
start
a
reddit
quad
for
example-
and
anything
just
say:
ok,
please
reference.
This
configuration.
B
A
So
that's
that
and
before
config
relations,
historically,
there
were
secrets
and
sequence
were
just
a
way
to
store
like
credentials
in
your
classes,
but
they're
not
encrypted
in
any
way
they're.
Just
like
a
64
encoded,
please
don't
spoil
them
by
accident
and
you
have
like
a
really
simple
password.
So
people
wouldn't
see
that
it's
base64
encoding
their
killable
restore
TLS
certificate
such
as
preferential
everywhere
yeah,
and
the
last
thing
regarding
to
when
it
will
want
to
talk
its
volume
volumes
are
the
way
to
persist
the
state
of
a
part
of
beyond
like
pot
lifetime.
A
That
means,
if
you
have
a
deployment
or
a
a
replica,
set
running
your
production
database,
maybe,
and-
and
you
don't
want
to
lose
your
data
when
the
onenote
dies
you
can
use
volumes
for
that.
Two
volumes
which
are
not
writable
are
in
the
config
map
and
the
secret
already
talked
about,
and
then
there
is
host
pass,
which
means
you
write
your
data
to
the
local
note
which
on
which
the
park
is
running.
A
So
that's
also
a
lot
of
good
and
good
use
case
to
to
persist
the
data
across
like
milk
areas
that
would
participate
data
across
Australia,
for
example.
If
you
wanted
to
run
elasticsearch,
for
example,
you
can
put
the
data
in
a
host
part
on
the
note
and
then
you
can
update
the
image,
you're
running
or
elastic
search,
and
we
pick
up
the
old
data
from
from
the
original
one
and
you
will
have
to
rely
on
on
that
or
later
base.
A
And
then
there
is
network
network
volume,
for
example
on
Google
compute
engine
or
Google
container
engine.
You
can
use
persistent
disk,
so
these
are
automatically
mounted
into
the
pot
and
you
can
use
them
to
store
your
data
and
in
case
the
notifier
fails.
The
part
would
be
rescheduled
on
the
new
mode
and
and
kuba
natives
would
make
sure
that
the
the
disk
with
data
from
the
old
nose
Mulder
into
the
new
one
on
ADA.
A
We
actually
have
a
similar
concept,
but
you
can
also
use
and
NFS,
for
example,
various
plugins
for
staff
and
all
kinds,
but
you
can
even
mount
each
repository
in
cydia
part,
and
so
so
that's
volumes
and
and
the
last
words
in
the
theoretical
part
will
be
about
the
architecture.
First
of
all,
as
I
said,
the
api
is
the
most
important
part
of
google
native.
Everyone
is
talking
to
vapi.
No
one
is
talking
to
each
other.
A
Basically,
all
the
communication
goes
through
the
API
in
the
HDD
left
the
prison
storage
of
Professor
and
back-end
in
the
in
the
normal
case,
and
what
you
set
up.
If
you
have
your
HDD
classes,
just
the
key
values
coach
and
hold
the
state
of
all
your
classes,
and
you
have
an
API
server
sitting
in
front
of
that.
A
And
then,
if
you,
if
you
wanted
to
add
a
new
knowledge
of
the
console,
you
would
would
started
a
cute
black
cube
with
the
obstruction
layer
in
some
places
and
the
container
rugged
pioneer
and
the
Kipling
would
go
to
vapi
and
say:
hey
I
mean
you
know.
I
want
to
be
part
of
your
pasta
and
I
have
like
a
gigabyte
of
ram
and
and
like
a
course
or
something
like
that.
There's
nothing
going
on
there.
So
so
he
would
like
store
that
information
via
API
and
exedy
and
then,
oh
and
all
of
a
sudden.
A
Someone
comes
here,
but
that's
me
basically,
as
a
user
of
cuban
in
four
different
user
and
said,
I
want
to
start
that
pop,
so
I
talked
to
the
API
say:
I
want
to
have,
but
running
I,
don't
care
where.
So
this
is
not
of
starting
up
to
this
and
then
the
scheduler
comes
into
play
and
the
scheduler
always
asks
the
API,
so
something
I
need
to
do,
and
then
the
scheduler
figures
are
okay.
A
There
is
a
new
pot
and
then
the
scheduler
looked
at
what
the
note
PA
report
basically
and
figures
out
where
to
put
that
that
pot
and
then
it
figures
that
ok,
note
1,
for
example,
has
enough
thread
for
for
the
pot
and
and
and
enough
like
resource
of
CPU.
Why
so
itself?
The
API?
Please
store
that
when
next
time
this
cute
little
it's
just,
it
should
be
modified
that
it
should
start
this
pot
and
when
this
process
started
here,
then
the
cupid
is
updating
the
API.
A
The
pot
is
still
running,
etc
and
stuff
like
that,
and
then
you
have
the
controller
and
the
controller
manager
is
sitting
on
top
of
the
scheduler.
The
schedule
is
basically
responsible
for
scheduling
parts.
To
separate
note,
the
controller
manager
is
responsible
for
ensuring
that
these
replicas
fact
these
deployments
are
actually
running
on
the
purpose
of
this
specific
amount
of
and
fixed
amount
of
pots
is
running
of
any
kind.
So
in
that
case,
let's
say
one
no
dice.
For
example,
the
comparable
manager
figured
out.
A
Ok
I
wanted
to
have
two
notes
of
that
kind
running
and,
and
it
would
just
tell
the
API
as
I
could
do.
Please
get
you
a
new
pot.
I
don't
came
where
and
then
the
scheduler
comes
in
figure.
1,
ok
need
to
do
something.
Ok,
so
that's
the
basic
architecture
there,
any
like
questions
already
regarding
the
architecture.
A
Yes,
let's
feel
like
in
your
venmo
look,
and
so
the
the
the
actual
here
is,
if
you
configure
in
your
manifest
that
you
are
going
for
this
specific
storage,
so
you
can't
stay
independent
by
using
NFS.
If
you
want
that
or
surface
like
that,
but
if
you're
on
AWS
or
in
Google
a
container
engine
or
other
I
think
there
is
also
a
plug-in
and
then
you
can
use
the
specific
ones.
But
then
you
were
a
little
bit
more
bouncy
or
hostess
than
than
you
are
and
then
you
cannot.
C
A
D
Just
a
note
on
that,
in
one
dot,
for
they
have
the
storage
classes
in
beta,
which
allows
you
to
auto
provision,
Gloucester,
GCE,
AWS
and
just
one
more
but
I
can't
remember
who
it
is.
They
allow
you
to
auto
provision
using
persistent
volumes
and
persistent
volume
claims,
so
you
can
actually
get
round
that
now.
So
you
can
still
reuse
your
manifest,
but
you
can't
reuse
your
storage
class.
D
A
B
A
First
of
all,
you
kind
of
do
any
changes
to
the
typical,
so
you
kind
of
like
start
new
pot
and
then,
if
something
dies
in
their
pasta
pot
size.
This
will
like
it's
only
the
process
of
the
path
I
from
the
Cuba
can
restart
it.
So
that's
not
a
problem
if,
if
actually
an
old,
sighs
and
then
you
have
bigger
problems,
because
then
the
scheduler
wouldn't
be
able
to
talk
to
the
API
and
the
controller
manager
wouldn't
talk.
A
E
A
D
So
you
can
actually
I
think
it's
one
to
one:
they
get
to
commit
to
run
multimaster
where
coordination
comes
now
and
then
se
de
it's
highly
recommended
by
the
Gubru.
Many
teams
that
you
run
SED
separate
nodes
outside
of
clusters
that
have
absolutely
nothing
to
do
with
Cooper
dennys,
except
for
API
calls
just
because
SPD
itself
uses
a
protocol.
It
provides
a
che
natively
and
then
you
don't
really
have
to
care
about
your
API
server,
which
is
the
master
or
the
scheduling
components.
D
You
need
to
start
the
scheduling
components
on
all
of
your
masters,
though,
coordinate
with
themselves
to
make
one
of
them
all
year,
and
then
you
can
run
a
che.
I
mean
you're
running
three
nodes.
If
you
lose
all
three
obviously
with
us,
but
I
want
to
do
have
one
of
them
alive
in
the
cluster
and
you're,
using
something
whether
that
be
an
H,
a
proxy,
whether
that
the
DNS
round
robin
is
obviously
not
the
most
effective,
but
it
works.
A
Obviously
like
not
a
like
the
proof
that
it
would
work
better
to
go
for
Google
container
engine
than
google
is
taking
care
of
it,
so
they
have
SRH,
which
are
suing
the
back
up
and
restoring
them
and
which
also
they
are
also
doing.
The
update,
so
updates
are
also
a
little
bit
tricky,
but
things
are
getting
better
with
every
single
release
in
this
place.
Okay,
so
cool
so
and
let's
sum
up
the
architecture
yet
because
I
actually
want
to
talk
about
continuous
delivery.
A
So
what
is
continuous
delivery?
It's
where
it's
the
way
of
building
testing?
Releasing
your
software
faster
and
more
frequently
and
I
want
to
I
want
to
say
that
there
is
also
people
confusing
continious
delivery
and
continuous
deployment
I'm
a
plan
of
taking
this
delivery,
which
means-
and
everyone
can
deploy
any
time,
but
not
every
single
committees
automatically
deployed.
So
if
you,
if
you
like,
deploy
once
a
week
or
something
like
that,
and
but
you
don't
have
like
a
tight
schedule,
basically
you
say:
okay,
we
worked
on
some
stuff,
it's
ready,
we
deployed,
then.
A
Theoretically,
that's
already
continuous
delivery,
and
you
wouldn't
probably
not
do
that.
But
it's
continually
we
just
meet
your
application
is
always
going
to
say
to
defend
deployed.
So
your
master
is
cracking
the
test.
You
have
everything
set
up,
you
don't
you
don't
have
to
like
spend
like
two
days
of
building
images.
One
thing
like
that
view:
you
have
some
automation
around
that
to
be
able
to
do
like
if
you're
about
comping
your
room
and
says
we
need
to
deploy
this
update.
A
You
say:
okay,
here
we
go
and
that's
conservatory
and
also
what
coaches
continuous
delivery
for
me
is
and
more
like
the
death
of
philosophy
of
you
build
it.
You
run
it,
so
you
don't
have
dedicated
relief
teams
who
are
doing
your
releases,
and
so
so
if
the
developers
are
actually
if
they
like
most
upon
request,
for
example
and
say
I
want
to
see
that
life.
Now,
then
you
can
just
go
to
appearance
boy,
oh
boy.
A
We
can
I
assumed
and
most
of
the
time,
you'll
also
have
a
continuous
or
you
always
have
a
continuous
delivery
pipeline
pipeline.
It
just
depends
on
what
you
want
to
automate
and
I
set
up
this,
like
example,
pipeline
near.
First
of
all,
you
have
to
build
set
and
which
means
I,
compiling
f,
respond
wing,
etc.
A
Basically,
and
so
the
novel
pipelines
and
can
look
something
like
this-
you
have
your
source
code
management
and
written
and
I
like
went
for
a
hosted
version
CFO
as
a
keeper
give
up
or
get
left,
and
then
you
have
some
see.
I
hosted
TI
service
checking
out
their
code
being
notified,
fire
reported,
Empire,
MF
p.m.
hosted
and
and
and
then
he'll
normally
run
you
you
nitor
p.m.
also
built
your
doctor
images
here
and
then
you
can
push
them
in
a
registry
in
that
case
of
Republican
hope
that
as
a
doctor
halfway.
A
You
have
might
have
an
effort
pipeline
or
minify.
Your
JavaScript
and
etc,
and
then
you
build
an
out
of
these
artifacts
people's
container
images
and,
and
you
push
your
container
images
into
some
kind
of
registry.
Basically,
so
how
can
you
do
that
with
Cuba
later?
First
of
all,
and
you
can
use
the
build
container,
I
will
give
in
the
short
demo
and
after
this
slide,
first
of
all,
you
can
use
the
milk
container
running
in
your
google
nexus
cluster,
which
gets
the
docker
from
the
notes
and
mounted
as
a
volume.
A
So
in
that
case,
I
can
just
say
I
want
to
mount
the
docker
command
of
the
host
of
the
note
into
my
container
and-
and
I
can
also
say
I
want
to
mount
the
docker
it
into
the
container.
So
that
means
from
inside
a
container.
I
can
talk
to
the
soccer
Damon
running
on
the
on
the
note,
basically
and
just
build
any
other
images,
and
it's
not
software
running
inside
buffers.
Just
controlling
your
talker
from
within
the
doctor,
basically
and
so
yeah
both
are
mounted
from
the
Cuban.
A
It
is
notes
and
and
then
you
can
easily
set
up
a
private
soccer
registry
and
in
that
case,
in
the
deployment
and
I
use,
persistent
storage
for
that.
So
that
means-
and
you
can-
you
can
have
fault
tolerance
e.
It
could
probably
also
set
up
from
the
thing
with
them,
with
AWS
block
storage
and
TKE.
If
that
normally
can
only
mount
a
volume,
one
right,
I
read
and
write,
but
multiple
times
read-only.
A
So
you
can
figure
out
if
you
want
to
have
like
a
set
up,
which
is
like
proxying
that
for
my
case
normally,
because
there
is
not
that
many
activities
in
my
classes
and
a
single
note
in
registry
was
I,
guess
assistant
this
work.
We
could
also
go
for.
Like
a
persistent
storage,
there
is
plugins
for
the
doctor
to
treat
you
right,
you
at
three
or
Google
storage,
etc.
But
you
can,
if
you
can
run
that,
like
in
your
in
your
in
your
husband,
maybe.
A
Yeah
in
the
directory
exposes
itself
via
service
by
a
note,
part
I
didn't
talk
about
that,
so
normally,
and
these
services
are
actually
a
way
to
communicate
from
pot
in
to
a
group
of
other
parts.
Why
our
IP
address
and
to
to
talk
to
these
services?
You
need
to
be
part
of
the
class,
so
you
need
to
be
inside
of
torture,
talks
of
them
and
for
that
like
the
default
privacy,
and
then
there
is
a
pipe
cause.
A
Note
part,
and
that
means
you
also
open
up
a
specific
port
on
all
the
nodes
in
the
cluster
and
all
the
Kraken
cubelets
error
at
the
the
Cuban
negative
note.
So
you
can
talk
to
these
notes
from
the
outside,
but
also
from
the
from
the
inside.
So
that
means,
if
I'm
running
inside
a
docker
container.
On
any
note,
I
can
just
talk
to
the
to
the
to
the
local
interface.
A
Well,
it's
not
correct
doctor
would
be
talking
to
the
interface
and
he
would
be
like
outside,
open
or
I'll
show
an
example
keep
confusing,
and
so
you
can
always
say
once
27001
and
then
your
specific
most
part
and
talk
to
the
registry
and
and
yet
and
the
service
is
running
on
all
the
notes,
but
that's
normally
for
all
of
them
and
yeah
and
I
want
to
give
this
short
demo.
Regarding
that,
so
I
set
up
a
little
I
set
up
a
little
test
project
and
what
I
have
here
is
a.
Let
me
sing.
A
A
Into
the
project,
in
that
case,
it's
a
public
application,
so
I
don't
need
any
credentials,
but
you
could
also
mount
in
credentials
from
your
Cuban
epic
configuration
or
from
you
from
your
secrets
here,
so
it
wouldn't
be
a
big
big
deal.
So
in
that
case
you
would
have
your
credentials
in
a
in
a
file
sitting
in
your
build
notes.
Basically,
images
we
could
read
them
from
from
the
file,
for
example,
and
here
I
pick
the
revision,
what
what
is
currently
being
built
and
then
I
run
at
dr.
A
bill
fh
and
that's
part
of
the
actual
code
and
I'm
of
the
repository
I'm
building
and
I
can
control
that
in
a
minute
and
so
I
run.
This
I
provide
this
script.
The
image
tag
it
should
be
using
to
tack
the
image
with
and
and
after
that
again
also
and
the
descriptives
responsible
of
pushing
the
image
into
the
crapper
okay.
So
is
that
clear,
most
of
it?
It's
not
that
I
complicated.
I
think
so
here
is
a
docket
build
script,
and
this
is
pretty
pretty
simple.
A
A
A
Otherwise,
it
would
be
restarted
when
it's
accident,
and
here
the
configuration
is
a
little
bit
more
complicated
than
and
then
the
one
I
already
showed
in
the
in
the
first
manifest
example.
First
of
all,
I
mount
something
from
from
the
host
from
the
node
into
the
container.
First
of
all,
as
I
said,
the
doctor
binary,
so
this
is
dependent
on
how
you
set
up
your
notes.
Basically,
this
is
working
on
Google
container
engine
and
might
not
work
on
core
OS,
because
soccer
is
somewhere
else.
A
So
this
is
already
a
little
bit
faster
specific,
but
you
can
probably
figure
out
a
way
to
do
that.
Then
I
market
akasaka,
that's
normally
that's
running
on
the
run
dr.
pocket.
So
this
only
defines
the
volumes
I
want
to
have
I
need
to
mount
them
then,
and
here's
a
list
of
containers
in
that
case
is
only
one
container
I,
give
it
a
name
default,
so
it
can
address
it
from
the
outside.
A
This
is
the
base
image
you
can
use
any
base
in
which
you
want
for
that,
and
now
it
comes
a
little
hecky
tricky
part
so
that
build
script.
I
showed
before
that's
it's
here
so,
but
if
you
want
to
put
that
in
a
in
a
pot
manifest,
it
needs
to
be
Jason
encoded,
and
this
is
for
today,
I
just
like
hack,
that
in
the
afternoon
today
and
this
basically
just
prints
out
the
content
of
the
film
script
and
then
Jason
and
coated
wire
via
Ruby.
So
there
is
a
dependency
on
Ruby.
A
If
there
is
anyone
knowing
about
a
more
clever
way,
I
try
to
take
you
with
it,
but
I
but
I
but
I
failed,
but
this
for
the
time
time
being
works,
but
you
can
also
have
like
a
more
abstract
level
on
creating
this
pot
manifest
here
yeah,
but
others
that
put
this
is
works
for
me.
Normally,
you
can
watch
program
something
around
it
or
generated
from
your
programming,
language
or
whatever.
Okay,
so
that's
that's
the
command.
Is
that
clear
to
anyone?
Other
the
command
works,
basically
I
just.
C
E
A
Gettin,
it's
gettin
a
little
bit
little
bit
complicated.
In
that
case,
the
Ruby
run.
Ty
needs
to
be
present
on
the
machine
where
I
am
running
that
script
from
because
what
I'm
doing
here
is
basically
I
called
your
stash
and
then
this
and
then
you
see
that
it's
I
copies
in
like
Jason
encoded
this
stuff,
and
this
is
to
script
basically
or
the
manifest
which
I'm
throwing
at
the
Cuban
in
disgust
and
say.
Please
fill
me
that
and
after
that,
if
you,
if
you
compile
that
script,
then
you
could.
A
A
Any
more
questions
regarding
that,
okay
and
the
last
bit
is
the
volume
on,
and
so
here
I
basically
reference.
These
two
volumes,
the
doctor
volume
with
the
socket
volume
and
just
mount
them
in
the
very
same
directory
as
I,
have
them
on
the
on
the
local
under
under
note
and
yeah.
So
the
little
basic
way
to
to
create
it
to
use
scuba
native
to
to
build
something
I
have
a
so
I
can
do
now.
For
example,
if.
A
Any
bill
can
flip
yomo.
So,
as
I
said,
this
is
just
like
spitting
all
this
in
line.
Json
and
I
cannot
just
pipe
that
cute
cattle
create
minus
f
dash.
So
that
means
just
like
sorry
just
a
second
okay,
so
I
evaluate
that
as
a
bash
script,
I
just
made
the
GMO
for
the
the
highlighting
works
better
for
me
and
and
then
I
like
pie,
that
is
cute
couple
create
and
what
you
cuddle
Creator
to
cuddle.
If
the
command
line
tool
to
talk
to
your
cluster
is
basically
create
a
sin.
A
Here
so
this
is
the
list
of
parts
which
are
currently
running
in
my
class.
I
have
some
some
old
stuff
lying
around.
This
is
the
one
I'm
interested
in
so
I
can
just
say
and
show
me
the
locks,
oh
by
the
way
and
okay
just
an
alias
for
to
cuddle.
So
if
you
typed
it
a
hundred
times
an
hour
or
so
it's
quite
a
time
saver.
So
what
I
now
want
to
do?
What
happens
actually
inside
the
container
look
at
the
locks-
and
it
looks
quite
nice.
A
A
C
A
B
B
A
Yeah
well
yeah.
In
that
case,
it's
always
using
hat.
If
you
wanted
to
use
a
specific
revision,
recommend
to
use
get
archives,
so
you
just
like
clone.
The
repository
affects
everything
from
with
lovely
and
archives
that
right
into
a
temporary
directory
and
then
build
that
from
that
temporary
directory
and
I
would
probably
move
all
that
to
hear
so.
A
So
this
is
just
like
a
really
simple
way
to
build
my
project,
but
you
could
set
up
like
I,
want
to
build
a
specific
branch
or
a
specific
feature,
something
like
that:
yeah
yeah,
so
the
the
hustle
around
here
the
is
or
the
good
clone.
If
basically
I,
don't
need
a
deep
clone
here,
for
example,
because
I
just
like
pick
the
most
recent
one,
I'm
only
interested
in
the
code
and
in
the
docker
build
script,
which
is
in
the
repository
okay.
So
that's
the
boot
that
so
that's
the
good
faith
yeah.
A
A
So,
in
that
case,
it's
just
a
little
shell
wrap
around
it,
so
you
can
see
what's
currently
going
on
in
that
case
also
all
the
time
every
time
I
run
it
and
that
alpine
image
installs
the
the
same,
go
and
get
binary.
So
that's
also
something
you
could
work
on
and
try.
Caching
gives
a
specific
bills
container
for
that
or
something
of
that,
but
I
didn't
want
to
make
it
to
accomplish
complicated,
so
yeah,
so
that
that
build
script,
maybe.
A
It's
basically
the
same
thing:
I
executed
on
the
command
line
and
dash
that
file
I
am
giving
in
pipe
it
in
cube,
clever,
create
in
the
in
the
last
bit,
I
extract
the
ID,
which
is
created,
and
here
I
wait
for
the
for
the
part
to
be
actually
running
and
no
depending
in
that
loop
and
here
I,
just
tailed,
lock,
basically
and
I
wait
for
the
long
tail
to
actually
work
yeah,
as
I
said.
This
is
just
like
a
little
example.
A
It
is
not
like
something
we
use
in
production
or
something
of
that
just
a
little
example.
To
show
you
that
there
is,
you
only
need
a
few
lines
of
desperate
or
you
I'd
recommend
to
use
a
programming
language
better
than
when
bash.
For
that
I
personally,
I
would
do
that
and
go
probably,
but
it
will
it's
quite
hard
to
get
that
into
your
CI
server,
which
I'll
be
talking
later,
and
but
it's
just
with
a
bunch
of
cache
files,
and
I
wrote
everything
I'm
showing
today
today
and
you
can.
A
A
Yeah
but
I
didn't
want
to
I
didn't
want
to
say
this
is
the
point
of
all
of
these
slices.
Cuba
native
is
not
a
product,
it's
a
framework.
If
you
have
things
on
top
of
that
like
something
or
someone
else
already
built,
you
can
use
that,
but
you're
not
limited
to
this
one
tool
or
these
free
tools.
It's
not
that
complicated
to
do
that
yourself
with
you.
If
you
have
a
different
idea:
okay,
cool!
Let's,
let's
talk!
Let's
talk
later,
okay.
B
A
So,
okay,
that's
the
business
and
all
the
tests
so
testing.
What
steps
testing
mean
already
said?
You're
the
test
integration
test
exceptions
pass
the
test,
normally
something
you
run.
Your
code,
like
I,
did
a
lot
of
Ruby
rearranged
development.
So
you
have
your
your
code
and
you
like
running
on,
like
a
check
out
of
your
of
your
coat
integration
tests
normally
like
they're,
not
that,
like
straight.
A
A
First,
you
might
need
a
kadai
debase
lefty
search
shredded
longer,
baby,
whatever
you
use,
so
that's
sometimes
a
little
bit
more
complicated
to
set
up
and
then
acceptance
test
and
in
that
area
I'm
talking
about
like
selenium,
maybe
even
like
a
manual
testing
throughout
clicking
through
your
application,
and
so
how
could
you
do
that
with
cuban
edit?
First
of
all,
first
of
all,
yarmulkes,
it's
quite
simple!
You
can
use
a
a
container
for
that
as
well,
so
similar
to
how
I
did
the
bills
in
the
inner
container.
A
You
can
also
obviously
run
your
tests
in
a
container
for
integration
tests.
When
you
need
more
and
you
can,
you
can
run
your
tests
in
the
pot.
So
here
is
the
the
exception.
I
always
always
say:
don't
put
your
production
database
in
the
pot
next,
your
application
server
for
testing.
This
is
the
perfect
just
like
a
solution,
because
then
you
have
your
application.
A
The
application
for
texting
me
field
production
database
need
to
be
readily
to
elasticsearch
and
you
put
all
of
them
in
a
pot,
and
you
don't
have
to
figure
out
where,
like
my
production
database
is
running
or
like
my
test,
database
is
running
with
my
test.
That
is
when
my
stan
soleski,
such
as
running.
All
of
them
are
binding
to
the
local
interface.
Basically,
so
if
you
compare
that
to
dr.
compose
or
something
like
that,
which
I
can
compose
you,
you
have
like
multiple
container
and
then
you
like
start
the
database
campagna.
A
And
then
you
start
your
test
container
and
then
you
meet
you
like
put
your
like
IP
of
the
database
components
from
how
into
the
test
container,
because
they're
not
running
on
the
same
network
interface.
Maybe
you
can
do
that
now,
but
I
don't
know-
and
but
in
that
case
you
don't
have
to
care
about
that
android
and
you
can
also
run
like
I.
Don't
know
tens
of
tests
of
parallel
because
they're
not
interfering
they're,
all
like
only
working
on
a
specific
network
interface
they're.
Not
writing
into
the
file
system
and
attentively.
A
Richard,
emulator,
and
so
so
so
normally,
and
how
we
references
the
database
rather
elastic,
such
as
by
environment
variable
so
so
I
would
end
in
a
viral
variable.
Ready
SQL
is
reddish
color
and
/
/.
Normally
it's
ready
if
I
write
it
in
production.
If
I
run
the
test,
it's
107
00
16
379.
Basically,
so
you
should
look
into
that
and
yeah
something
like
that,
but
you
get
and
you
can
probably
make
up
other
things.
So,
okay,
yeah
I,
want
to
want
to
want
to
give
a
demo
of
that
as
well.
C
A
Not
playing
The
Scream,
okay,
no,
it's
true!
It's
a
little
Go
application
of
excess
here,
I'm,
really
really
pragmatic
and
just
putting
in
and
the
posters
database
is
running
on
localhost,
so
Adam
I
don't
mind.
So
this
is
working
for
my
local
machine.
This
is
also
working
from
inside
the
pop
if
I
have
to
the
post
inside
pot
and
I'm,
just
creating
a
table
inserting
something
and
figuring
out
the
count.
The
bit
verbose
at
the
lot
like
go
or
not,
I'm
not
point
today.
A
A
So
what's
interesting
here,
it's
a
bit
more
complicated.
First
of
all,
the
first
that
is
echo
on
running
path
or
I,
know
that
actually
the
part
about
it
and
then
I
say
install
dependencies,
and
so
the
pendency
means
these
are
basically
the
dependencies
I
need
and
to
figure
out.
If
the
the
test
can
already
run
so
I
always
meet
Jake
you
for
some
extracting
off
I'd
eat
from
from
responses
from
Cuba
native,
then
I
need
curl
and
clothes
dress,
and
those
two
to
figure
out
with
myself
is
already
running.
A
I
could
also
look
in
and
check
the
API.
You
can
provide
and
health
check
or
something
like
that,
but
in
that
case,
I
want
to
keep
it
simple,
but
I
have
that
in
that's
brick
and
then
there
is
logic
which
are
waiting
like
wait
for
post,
read
it
for
last
research
by
a
CO
and
give
and
go
to
like
check
out
the
repository.
So
that's
the
install
dependency
thing.
Then
you
have
set
our
project.
This
is
similar
to
how
I
did
the
bills.
I
write
it
in
to
go
pass
clone
it.
A
So
that's
that
and
then
wait
for
services.
This
is
what
I
already
said.
First
of
all,
I'll
just
wait
for
like
TCP
port
for
five
four
three,
two,
the
posters
p
for
to
be
available.
It's
like
again
just
like
a
shell
script,
which
is
like
an
infinite
loop
and
doing
that
cut
and
to
the
in
to
the
host.
Then
I
create
a
test
database.
Every
every
test
run
will
run
from
a
scratch
from
scratch
database
pedigree.
So
it
will
nothing
be
nothing
will
be
there
great
a
database
and
then
similar.
A
I
wait
for
HTTP
wait
for
elasticsearch
to
be
available,
so
that's
here
similar
to
the
wait,
for
course,
and
yeah
then
elasticsearch
is
available.
I
don't
need
to
create
anything
here
and
then
I
here,
I
just
like
filter
out
the
the
vendor
dependency,
looks
or-
and
here
I
just
run
and
go
test
on
these
I'm
interested.
Ok,
so
that's
a
test
good,
any
questions
so
far.
A
So
this
is
actually
simpler
them
and
then
the
other
one.
So
the
same
pattern
and
F
before
I
have
the
alpine
a
jobs
bill
container.
I
I've
in
the
sh
and
here
I
just
configure
that
I
I
want
to
post
Chris
96
running
in
that
hot
anyone
want
to
have
elastic,
search
175
in
that
case
running
in
that
cot,
and
that
case
I'll
say
we
start
never
so
in
case
of
an
error,
just
don't
care
like
it
would
be
Marcus
an
error
and
generate
name
here
is
test
so
yeah.
A
A
So
that's
the
same
same
rapper,
foot,
basically
and
then
I
camera
and,
let's
see
what
happens
so
start
a
container
same
functionality.
In
that
case,
I'm
installing
quite
a
lot
of
dependencies.
I
was
quite
surprised
how
fast
that
all
goes
with
Alpine
on
on
Google's.
Well,
I
didn't
mind
like
caching,
something
so
you
see
the
cloning
and
goes
on
right
now,
then
I
waiting
to
postgres
post
was
available.
Elasticsearch
most
of
the
time
takes
a
little
bit
in
sa
elasticsearch
and
then
I
run.
A
C
A
The
image
I
didn't
image
here
you
mean
the
build
a
that's
with
us
69
seconds
like
three
and
50
megabyte
I,
think
it's
all
on
us
feel
85
megabytes
or
something
like
that.
You
can
compress
it
a
little
bit
so
and
it's
not
that
big,
but
if
you-
and
if
you
want
to
like
safe
and
with
something
like
that,
it
might
make
sense
to
do
you
create
a
prebuilt
image
of
something
like
that,
but
the
build
step
I
showed
before
you
can
also
use
that
to
basically
build
your
build
images.
A
B
A
E
A
Yes,
so
I
would
probably
be
template
like
the
elastic
circles,
with
a
version.
This
is
quite
similar
to
dr.
compose,
like
you
also
have
the
similar.
So
would
probably
add
these
templates
to
your
turbo
vettori
and
then
maybe
have
a
template
language
like,
for
example,
I.
Don't
know
if
you,
if
you're
familiar
with
go,
you
can
just
add,
go
and
go
templating
around
that,
and
then
you
have
like
a
generic
binary,
which
figures
out
the
dependencies
etc.
But
M
I
said
it
again.
A
Yeah
well
normally
I
had
is
so
so
no
no
and
you
create
a
part,
and
the
part
has
the
database
plastic,
rich
and
and
one
container,
which
is
running
the
test.
So
what
I
had
in
there
before
later
on
is
that
in
case
everything's
fine
and
that
a
container
which
is
running
the
tests,
you
could
also
mount
in
the
cube
cuddle
from
from
the
node
and
then
kill
em
yourself.
A
B
A
Yeah
I
look
into
the
children
job,
but
the
in
that
case
it's
not
a
big
difference.
I
think.
If
you
used
to
pop
like
like
this
or
do
the
job
thing,
the
the
job
sex
abstraction
basically
make
sure
similar
to
the
deployments
that
if
your
bills
fails
on
one
load,
it
could
be
rescheduled
to
another
node.
In
that
case,
that
wouldn't
wouldn't
happen,
but
em
like
for
like
a
build
scenario.
I
wouldn't,
but
you
can
also
use
just
for
that
purpose.
A
E
A
A
B
A
What
is
the
application?
It
has
a
click
me
button,
but
importantly,
I
don't
know
if
we
can
see
that,
maybe
you
can
open
up
the
source
code,
it's
just
using
javascript
to
think
so.
This
is
something
really
really
hard
to
test,
but
ok
really
really
hard
to
tell-
and
this
is
why
I
need
to
learn
in
for
it.
Okay,
and
to
show
you
that
it's
actually
work,
you
can
click
on
it
and
it
just
sets
you
quit.
Oh,
but
how
can
I
run
that
in
0.
A
Integration
template
first
of
all
the
integration
script
yeah,
so
with
a
comfort.
Okay,
I
have
a
three
containers
in
that
cave.
First
of
all,
the
app
container-
and
this
is
something
this
is
just
something
similar
to
the
1i
already
built.
So
this
is
just
an
image
I've
already
built
and
push
to
the
registry.
So
this
is
the
application
which
is
running
in
in
this
page
here.
A
Okay,
then,
I
have
a
selenium
and,
like
instance,
in
the
very
in
the
same
pot
and
and
I
will
have
this
default
thing,
and
the
default
thing
is
also
the
same
binary
here,
because
I
was
lazy
I
put
in
the
integration
testing
to
the
very
binary
which
is
actually
like
running
the
applications
if
a
go
binary,
everything's
building
into
the
binary.
So
when
I
start
that
binary
with
one
integration
equals
true
here
I
use
the
ends
variable.
A
Then
it
would
run
the
selenium
integration
tests
and
they
look
a
little
bit
like
this
also
hard
coded
just
just
just
for
the
convenience.
I
know
I
need
to
talk
to
the
f
URL
listening
on
localhost
inform
2000
and
this
many
new
URL,
also
in
local,
and
no
no
a
little
different
here
then
I
wait
for
these
services
to
be
available.
A
This
is
very
similar
to
the
touch
screen
but
like
it
will
over
both
because
it's
go
and
wait
for
them
for
five
minutes
to
be
available
and
then
I
like
starter,
would
get
any
driver
from
selenium
and
then
I
just
get
that
you
ll
hear
Clary
for
the
anchor
tag
and
then
click
it
and
then
I
hear
verify
that
it's
the
pessimistic
way.
Okay.
So
that's
that
and
yeah.
A
B
A
Like
this
I
think
I
already
at
that,
but
I
but
I
but
I
changed
this
I
had
a
shell
script
but
which
was
really
really
hard
to
do
to
get
that
thing
running.
For
some
reason,
something
was
listening
or
something
like
that,
so
I
don't
rely
on
on
on
that
at
all
the
only
thing
which
is
like
a
template
II
that
I
just
like
use
that
very
image
here
I
can
also
provide
from
environment
or
whatever
here.
B
B
A
On
it,
ok,
so
the
same
output.
This
is
the
offer
from
the
ghost
groups
or
wait
for
the
application
to
be
available,
and
it
already
like
finished
basically.
So
this
is
the
time
it
took
for
the
selenium
pot
to
come
up,
for
the
application
should
come
up.
Go
with
is
really
really
tough
and
to
run
with
em
the
the
test,
and,
if
you,
if
you
are
using
the
scripting
language
in
the
framework
around
selenium,
you
can
just
like,
like
clone
your
repository
on
them,
but
it's
a
again
really
really
nice
for
like
self-contained
acceptance.
A
A
No,
no,
the
business
I
put
in
then
the
goal
binary.
Basically
I
have
like
a
start
equal
time
now
and
I
have
a
logger
which
is
creeping
the
gesture
to
figure
out
how
long
the
whole
thing
actually
works.
Hello
I
need
to
wait
for
it,
ok,
cool,
so
so
that
was
that
integration
testing
any
questions
regarding
integration
tests.
A
B
A
Here
I
set
up
a
with
the
pipeline
plugin,
which
I
really
like
a
pipeline
which
is
doing
like
these
bills
passed
declaration
pages,
but
they
are
now
running
inside
by
cluster
and
and
I
can
set
that
up.
This
is
running
with
every
single
commit.
Basically,
I
can
like
show
you
how
the
pipeline
case
we
have
again
in
place-
and
you
see
this
is
like
the
same
script.
Are
you
this
is
like
very
famous,
but
I
just
showed
you
on
the
control,
but
now
this
is
hosted
in
in
the
Canadian
Cuban
little
cluster
itself.
A
That
means
Jenkins
is
running
every
pot
in
the
cluster,
so
this
is
actually
the
you
talk
from
script
running
in
Jenkins
to
the
darker
in
1k
cloning
pushing
into
the
registry
from
from
the
craft
itself
yeah.
So
that's
that
so
yeah
the
building
near
you
see
that
the
image
of
course,
so
it's
behaving
famous
and
I
just
showed
you-
and
here
like
this-
will
be
for
the
unica,
and
for
that
that's
the
one
thing
I
wanted
to
show
you
and.
A
A
It
depends
so
it
it
doesn't
have
to
be
a
production
cluster.
But
then
again
you
cannot
access
your
registry
from
from
in,
but
what
you
can
do
with
kubernetes,
for
example,
we
can
have,
let's
say
15
notes
or
something,
and
then
you
say
you
can
label
these
notes
as
well,
like
you
label
the
pot
and
say
these
are
my
built
notes,
these
five,
for
example,
and
their
lives
build,
and
then
you
can
add
restrictions
to
these
pots.
A
That
Jenkins
is
only
able
to
allow
to
run
on
this
build
mode
and
also
that
the
bills
around
the
running
on
these
personal.
So
so,
in
that
case,
you
can
like,
like
separate
first
of
all
like
on
a
motorized
and
these
from
from
from
the
other,
but
for
for
namespaces,
for
example,
means
like
you,
like.
You,
like
protection,
like
the
text
shouldn't
be
able
to
talk
to
the
production
database,
is
something
like
that
you
can
you
can
you
can
set
up
a
separate
namespace?
Probably
for
that.
A
So
you
cannot
just
like
talk
to
you,
production
database
and,
and
there
is
that
what
plugins
way
you
could
also
limit
the
the
road
to
not
be
able
to
talk
to
specific
things
so
but
yeah
and
as
I
said,
it's
not
a
production.
You
use
that
in
production,
and
you
exactly
have
to
do
it
like
this.
Just
to
get
you
an
impression
how
how
you
can
use
kubernetes
to
implement
like
some
kind
of
continuous
delivery,
so
you
can
also
use
a
public
registry
and
then
easily
build
cluster
and
push
into
that.
A
Although
two
more
things
I
want
to
show.
First
of
all
how
you
set
up
that
build
pipeline
it,
it's
really
really
simple:
it's
a
pipeline
plugin!
So
you
have
a
Jenkins
file
inside
your
repository
and
you
basically
only
set
up
the
project,
the
git
repository.
Please
fetch
it
and-
and
you
don't
have
to
configure
anything.
A
Everything
is
configured
here
like
in
Jenkins.
This
is
like
hard
normal.
So
here
said
the
environment,
where
the
image
repo
is
so
similar
to
the
one
here
check
out
it
and
then
I
run.
These
very
same
scripts
are
already
went
just
just
in
that
case
in
in
the
CIA,
so
that's
that
and
what
you
can
can
be
even
further
what
laughing
come,
if
was
it?
Don't
worry?
Is
that
hello
to
another
application,
where
the
pipeline
is
even
even
more
advanced
in
that
case
I
can
chew.
A
It
also
has
a
release
stage
here
and
in
the
release
page
just
a
few
seconds,
but
the
realistic
is
optional,
so
the
relief
can
also
be
just
a
script.
In
that
case,
we
need
to
cuddle
to
update
the
deployment
of
the
image.
Basically,
and-
and
in
that
case
you
see
here
that
it
actually
prompts
me
if
I
want
to
proceed
with
that
stage
or
not.
So
in
that
case
I
can
say
a
ball.
A
A
C
B
A
A
A
Yeah,
well,
you
would
probably
have
two
different
deployments:
I
didn't
talk
about
the
products
today,
and
so
what
you
have
a
view
at
one
deployment
running
and
the
department
creates
pot,
and
the
process
is
basically
reference
like
by
label
these
pots
and
for
a
be
testing.
You
could
have
a
different
deployment,
but
love
with
a
50
replicas,
but
only
with
five,
and
they
need
a
ten
percent,
and
if
they
have
to
save
tax,
they
would
all
be
part
of
the
service
or
something,
and
the
nice
thing
about
that
is.
A
This
is
in
the
inside
here,
for
example,
if
you,
if
you
have
that
deployment
you
could
have,
this
is
your
default
deployment.
Basically,
so
this
is.
This
is
what
your
end
users
look
at,
but
the
kind
of
separate
prevacid
here
and
here
so
you
can
directly
talk
specifically
to
only
these
posts
here,
for
example,
these
would
be
15
IV
these
55.
So
this
would
be
like
a
if
you
go
here.
A
A
C
A
Gold
ever
so
like
in
native
library,
so
you
don't,
but
the
authentication
is
with
a
little
bit
tricky,
sometimes,
for
example,
with
Google.
They
use
all
off
on
their
own.
Your
token,
so
I
think
the
newest
version
of
the
library
and
can
do
that.
There
refresh
tokens
and
all
that
kind
of
a
few
puddles
between
all
that
under
the
hood,
and
if
you
provide
my
little
Jason,
it
would
get
Jason
office
anyway.
You
can
throw
in
Jason
and
get
space
enough,
but
you
have
to
dependency
that
you
have
to
ask
you
cuddle
playing.
A
B
A
A
premier,
for
example,
it's
really
really
simple
setup
in
the
crust
as
well.
So
you
have
monitoring-
and
here
you
have
all
the
details
running
like
what
image
is
currently
running
and
then
you
can
like
expose
a
closed
all
the
metrics,
and
this
is
also
like
there's
an
API
here,
so
you
can.
You
can
use
that
to
cancel
your
appointments,
but
and
you
can't
go
as
crazy
as
you
want.
Basically.
E
C
A
A
But
this
is
what
I
showed
about
the
adjacent
file:
the
Jenkins
file
it
in
the
repository,
what
you're
building
it's
not
on
the
drinking,
so
it
picks
up
the
check
on
file,
and
then
you
define
all
the
pipeline.
Defining
your
codebase,
not
no
longer
think
convinced,
I
think
biggest
benefit
of
that
whole
pipeline
thing,
because
this
is
what
I
used
to
do
before
as
well,
and
ejection
cos
h
and
the
only
skip
ahead
on
Pegasus
run
at
changing
h
and
it's
so
11
to
do
something.
Fancy
then
each
put
you
configure
or.
A
Man,
okay,
yeah
I,
didn't
didn't
do
any
like
before
we
did
kubel
it
away,
we
went
occur
in
production
or
offensive,
so
we're
still
running
or
like
stateful
stuff
outside
of
responses,
so
you're
our
database.
For
example,
will
we
try
to
get
hosted
versions
of
for
database
which
we
use
early
estimates
made
of
yet?
And
but
we
still
run
our
own
ethics
ourselves?
I,
don't
secure
benefit
for.
A
For
that
well
configuration
management
we
tend
to
just
like
you
called
in
a
script
and
then
in
the
Buddha
note
and
I
just
start
a
container
with
a
would
like
specific
elasticsearch
image
and
like
set
up
some
methods
HP.
But
if
it's
throwaway
infrastructure,
when
this
node
fails
and
throw
it
away,
I
don't
like
trying
to
fix
something.
Configuration
management
like
that.
C
A
C
A
A
Buckets
behind
that
that's
possible
yet,
but
I
don't.
I
don't
know
about
how
how
the
locking
is
to
push
into
the
same
s3
bucket,
because
there
is
eventual
christmas
enorme
s3,
so
you
don't
have
any
guarantees
there.
I
think
having
it
on
a
disk
is
sis,
but
but
this
is
all
only
an
example
that
it's
a
self-contained
thing:
it's
not
and
you
can
still
use
quite
or
docker
hub.
You
can
also
use
them
for
building
a
fortress.
You
don't
have.
D
E
C
D
You
use
a
pull
to
ramir.
You
accept
the
images
that
you
need
to
run
the
cluster.
If
you
aren't
using
a
pull-through
mirror,
then
usually
you're
pulling
those
images
direct
from
what's
the
Google
GCR
Google
container
registry
or
the
docker
hub
or
Quay,
or
wherever
you're
pulling
them
from.
So
you
don't
eat
images
to
bootstrapper
cluster.
Your
cluster
starts
up.
Your
registry
starts
up.
You
now
have
access
to
all
of
your
images
and
then
the
cool
thing
about
Cooper
Nettie's
is
you
don't
have
to
care
about
timelines
and
all
that
stuff?
D
If
it
can't
pull
down
images,
it'll
just
sit
there
and
keep
trying
to
pull
them
down
until
a
10.
So
it'll
start
up
your
registry.
Your
registry
will
be
sitting
there
and
then,
as
soon
as
it's
available
and
the
content
is
available,
it'll
start
pulling
out
the
images
and
starting
all
those
containers
for.
A
A
F
D
A
A
So
so
you
can
talk
to
your
like
stop
running
in
a
container
from
the
node
of
which
the
container
is
actually
running,
and
the
nice
thing
about
the
north
point
is:
you
can
always
go
to
the
local
interface
and,
by
definition,
the
lower
my
defaulted,
a
local
interface
that
one
has
to
be
secure,
otherwise
you
have
to
add
either
security
exceptions
or
create
like
TLS
certificate
around
your
registry.
Those
roles
are
always
talking
to
bit
to
the
local
interface
I.
A
Let
it
out
but
I
think
it's
more
open
in
moorpark
and
all
the
notes
in
the
caster
and
and
with
the
don't
have
to
rely
on
the
DMF
thing.
It's
just
open
a
note
for
if
someone
talking
to
that
most
part,
if
you
specify
this
tax,
all
the
pot
related
to
be
like,
like
with
these
sites,
are
part
of
that
terms
with
a
note
part,
and
they
have
like
10
different
IP
addresses.
A
F
F
A
Yeah
but
I
don't
I,
don't
need
that
because
for
me,
like
ingress
and
load
balancing
around
that
is
talking
to
the
cluster
from
the
outside,
but
inside
the
building
through
the
building.
As
my
interface
into
the
classes
for
building
is
either
chip
cut
off
from
here.
But
I
don't
have
to
talk
to
this
port
from
the
office.
I
wouldn't
have
actually
here
anyway,
so
these
are
behind
a
firewall
and
all
that
kind
of
stuff
and
all
from
the
inside,
but
I've
already
in
the
inner
container.
A
E
F
D
From
the
note
itself,
so
that
there
is
there's
a
differentiation
between
so
usually
you're
running,
like
flannel
or
calico,
or
something
so
inside,
the
cluster
is
the
pods
and
and
the
actual
container
is
running
inside
the
pod.
But
you
can't
talk
like
if
you
are
sitting
on
the
node
itself.
That's
running
the
containers
and
you
try
to
talk
to
an
IP,
that's
inside
the
container
subnet.
D
You
aren't
going
to
get
it
even
talking
to
that
node,
even
though
they're
talking
on
the
same
interface
because
of
the
way
to
proxy
works
so
the
way
to
access
it
even
from
localhost
on
the
node,
not
localhost,
on
on
a
pod
but
local
host
on
the
note
is
that
you
have
to
expose
that
port
and
say
hey.
This
host
is
allowed
to
talk
to
the
container
that
is
running
on
it
or
that
is
running
somewhere
else
within
the
cluster.
D
A
F
A
E
It'll
look
a
mess
and
you're
going
to
give
it
to
be
in
it,
so
to
be
in
a
success
about
the
Guinness
I
website.
If
you
define
them
conclude
that
this
might
be
dangerous
office,
Indiana
remember,
it
is
not
able
to
know
that
created
messaging
and
unless
we
have
a
lot
of
interest
like
maybe
see
what
they
say
for
confirm
them
and
I
wasn't.
E
B
E
F
C
A
Who
lives
like
a
good
concept?
The
problem
is
your
everything
in
your
configuration
like
geese
without
any
manual
interference.
But
if
you,
if
you,
if
you
don't
want
to
have
an
ox
chi
on
call,
every
time
of
new
pot
comes
up
and
type
in
a
secret
to
decrypt
something
it
has
to
be
somewhere
in
the
in
the
SCD
configuration
of
the
crossbow.
So,
even
even
if
you
would
like
encryption.
C
B
D
D
Tennessee
with
in
coover
Nettie's
is
still
kind
of
a
little
bit
of
a
difficult
concept
and
we're
not
quite
there
yet
on
being
able
to
effectively
do
it
there's.
Definitely
no
you
eyes
or
anything
I
mean
you
can
hook
it
on
decks.
You
can
hook
on
ldap
and
do
some
other
crazy
stuff,
but
at
the
end
of
the
day,
I
mean
the
easiest
thing
that
you
can
do.
If
you
do
me
to
secure
secrets
for
subsets
is
run
separate
clusters
for
each
of
those
applications
outside
of
that,
I
mean
you're
really
looking
at.
D
Until
Dec
matures
a
little
bit
until
often
and
not
see
mature
a
little
bit,
if
you
have
access
to
the
cluster
with
coop
cuddles
for
the
most
part,
you
can
consider
that
person
to
have
access
to
all
of
the
secrets
all
of
the
resources
rules
yeah
exactly
so
it
it
is
getting
much
better.
And
with
Dex
and
the
author
and
off
the
end
point
I
mean
you
can
authenticate
much
better.
You
can
authorize
certain
actions.
You
can
restrict
users
to
certain
namespaces.
D
Certain
commands
certain
abilities
where
you
know
they
may
just
be
able
to
view
along
or
they
may
be
able
to
just
do
a
get
on
what
pods
are
running
in
this
namespace
I
mean
we're
able
to
do
that.
Granular
of
control
using
things
like
deck,
we're
not
quite
there
to
make
it
easy.
So
it's
definitely
getting
better
and
I
expect.
B
D
C
D
F
D
Somebody
very
recently
just
put
out
a
blog
post
on
storing
all
of
the
secrets
for
Cooper
Nettie's
within
Tasha
core
bulb,
which
does
allow
you
to
do
very
granular
access
control
to
the
keys,
even
for
machines,
but
also
gives
you
a
huge
advantage
in
that
you
can
do
a
lot
of
very
good
audit
logging
on
that.
So,
if
you're
very
concerned
about
your
secrets,
then
even
if
they
do
get
leaked
at
least
you
know
where
they
went,
and
you
can
change
them
very
easily,
because
both
does
a
lot
of
kind
of
Secrets
management.
A
Always
that
you
provide
your
users
who
use
to
cuddle
with
a
TLS
pacifica
stare,
so
you
have
just
like
the
authenticator
stuff,
like
yeah
fine
keys,
and
then
you
you
like
to
see
less,
can
shake
thing.
That's
awesome,
authentication,
not
just
for
like
security
and
but
I.
Take
that
you
off
at
least
from
my
understanding-
and
this
was
probably
the
same
time.
They
also
introduced
fine-grained
access
control
to
the
degree.
So
so
you
can
have
people
on
your
cabinets
after
you
cannot
create
anything
or
cannot
delete.
A
A
A
B
C
C
A
When
I,
when
I
way
bigger
concerns
with
the
networking
thing,
but
things
are
getting
better
there
as
well,
because
we
are
we,
we
came
from
and
like
and
a
my
security
group
based
approach
where
the
load
balancer
is
only
accessible
to
the
outside
none
other
component
and
then
the
application
only
it
gets
like
traffic
from
the
load.
But
it's
the
database
can
only
accept
like
traffic
from
the
application
services
from
there
and
when
you
go
to
give
it,
as
everyone
could
talk
to
everyone
right
now
and
but
there
is
a
like
Kaliko.