►
Description
발표자료: https://drive.google.com/file/d/1phjKwPQp7fSqyDeHRQFAbGn5D0_H_U9G/view?usp=sharing
잘 짜여진 모놀리스는 효율적이고 직관적이며 운용하기 편리한 합리적인 아키텍처입니다. 대부분의 작은 규모의 회사는 모놀리스로 서비스를 시작하지만 이 중 대부분은 서비스의 규모와 인력이 점차 늘어남에 따라 효율적인 아키텍처를 장기적으로 유지하는데 실패합니다. 오히려 모놀리스라는 이름의 수년간 쌓아온 프랑켄슈타인에 세번째 팔과 다섯번째 다리를 용접하고 있는 자신을 문득 깨닫게 되는 일도 드물지 않습니다. 뒤늦게 개발도 운용도 정상적으로 진행 할 수 없는 하향 나선의 상황에 직면에 있음을 깨닫게 되어도, 이미 하루에도 수만명이 이용하는 서비스를 근본적으로 손 대는 것 또한 좀처럼 엄두가 나지 않습니다. 우리는 수년간 쌓아 온 모놀리스 서비스를 해체하고 클라우드의 쿠버네티스로 옮길 수 있을까요? 어떻게하면 서비스를 사용하는 수 많은 유저들이 알아차리지 못하는 사이에 우리의 서비스를 안전하게 이전 할 수 있을까요? 클라우드 네이티브로의 이전은 과연 우리들의 Velocity에 큰 영향을 줄까요? 이 세션을 통해 데이터센터의 모놀리스를 클라우드의 쿠버네티스로 성공적으로 이전하기 위한 전략과 클라우드 네이티브 서비스의 장점, 그리고 주의점들을 짚어보려 합니다.
A
A
A
Think
this
announcement
will
be
of
some
help
to
those
who
are
interested
in
dismantling
the
monolith.
I've
heard.
A
lot
of
people
say
that
it's
a
native
of
the
cloud,,
but
what
exactly
is
a
plus
native?
Those
who
are
planning
to
relocate,?
And
since
this
is
a
cfs
Topic,
1,
event,
I,
think
most
of
the
people
who
come
here
will
be
familiar
with.
What
Tess
is
on
the
9th.
I'm.
Also.
A
A
A
A
A
Often
used
as
a
key
to
configuring
microservices,,
so
it
will
be
very
helpful
if
you
read
this
book.
It
is
a
book
called
Microservice
Patterns
that
came
out
relatively
recently..
This
book
has
good
examples
in
Java,
and
if
you
look
below,,
there
is
a
webpage
operated
by
the
author
called
Microservice
Diao.
The
esophagus
is
well
shown,
and
if
you
click
on
each
esophagus,,
the
explanations
are
very
good..
A
A
A
A
A
A
A
You,
don't
even
have
to
think
about
it.
So,,
as
you
all
know,
famous
services
around
the
world,,
such
as
Nrae
Flex,
1,
Amazon
started
with
monolith,,
so
the
core
is
monolith.
It's,
not
a
light
and
bad
architecture.
In.
The
early
stage
of
development,
microservice
port,
more
efficient
architecture
can
be
viewed
like
this.
It.
A
A
A
Realize
a
lot
that
we
can
no
longer
develop
quickly
and
operate
efficiently.
So,
uh,.
We
have
a
monolith,
pod,
baseball
and
more
architecture,
and
requests
for
new
features,
keep
coming
in,.
So
from
now
on,
This
picture
on
the
right
is
taken
from
the
hub
of
a
book
called
The
Phoenix
Project,,
which
is
one
of
my
favorite
books.
It's,
a.
A
A
A
Because
the
base
is
large,.
There
is
no
problem
with
the
part
that
people
are
working
on
right.
Now.
You
can
release
it.
But
if
more
than
50
people
are
working
on
one
code.
Base,
all
changes
are
okay.,
It's,
very
difficult,,
so
I
don't
think.
There's
a
big
problem
with
the
change
up
to
this
point
by
using
2%
of
the
neck
or
using
some
other
gatekeeping
method.
Let's
proceed
with
our
release,
and
that
part
will
eventually
be
released.
To
production..
A
A
If
you
luckily
found
this
problem
before
it
was
released
to
production,
For
example,
at
the
time
this
was
discovered,,
it
was
discovered
during
this
release.
Testing
period.
Then,.
The
release
will
become
a
firm
phase,
longer,
and
new
features,
or
things
like
this
can
happen.
Later.
And
finally,
to
apply
new
technologies,.
It
is
necessary
to
Since
I,
have
to
attach
the
continuing
function
of
the
code.
A
Base,
I
keep
repeating
this
kind
of
work
for
a
long
time,
and
when
I
do
it,,
what
I
want
to
make
is
the
shape
of
a
pretty
person,,
but
at
some
point,
I
see
three
legs,
and
it's
about
the
8th,
and
like
this.
In
the
first
place,
There
are
often
cases
where
an
architecture
like
Frankenstein
that
you
didn't
want
comes
out.
I.
A
Thought
it
was
unfair
to
look
at
it
like
this,,
so
the
front
end
at
the
top
has
a
very
pretty,
cute
and
beautiful
look,,
but
at
the
bottom,
it
also
depends
on
the
culture
moon,.
So
you
can
see
this
terrifying,
look.
So
we
have
so
far.
I've
talked
about
the
problems
of
Bliss,
and
the
problem
of
executives
and
questions.
In
fact,.
It
was
a
problem
that
has
been
emerging
for
a
very
long
time.
A
A
A
Four
standards
are
as
follows.
If
you
look
at
the
left,.
You
can
see
the
Eundi
play
pfc,
that
is,,
how
many
weeks
and
where
you
can
play,
and
the
change
to
the
lead.
Type.
We
changed
the
code.
How
long.
Can
it
take?
Next?
Is
the
mtt
al
and
the
mtt
al's
min
time
trigger
penalty,,
I.e.
if
something
goes
wrong.
When
you
do
a
new
backhoe,,
how
long
will
it
take
until
you
get
the
problem,
back?,
Quiet
and
1
in
the
change
pen,
column?
How
often
do
we
have
problems
when
we
do
distribution?.
A
This
is
the
Korean
version
now,,
and
if
you
look
at
some
of
the
important
details,,
this
part
contains
the
information
about
my
hometown,,
which
was
in
the
book
Accelerator
that
I
briefly
explained
earlier.
You
can
see
it
as
the
same
part,
and
in
the
later
part,.
How
often
can
it
be
performed,
predictably,
with
the
best
effort?
This
part
explains
it
well.
Before.
A
A
A
A
A
A
A
A
Parts
to
consider
when
dividing
a
microservice
candidate
into
a
smaller
unit
of
service
in
this
monolith.
It
should
be
maintained.
Even.
If
you
move
to
a
new
environment,,
it
will
only
be
that
the
function
is
lost
or
does
not
work,
and
secondly,.
The
infrastructure
that
is
currently
in
the
data
center
needs
to
be
migrated,,
and
this
itself
can
be
a
big
topic.,
Because,
well,.
The
current
job
is
to
play
data
with
3g,,
and
there
is
a
certain
service
running
on
the
bus,.
A
A
A
A
A
A
Wouldn't,
it
be
possible
to
create
a
new
one.
According
to
the
Diver
and
Maiko
service.
Omo
crowd,
substitution,
Tax
data
center
All
leads
are
the
five
senses
of
the
past.
Cloud
microservices
are
our
bright
future
Utopia
Let's
all
think
this
way
and
write
all
new
things
like
this
Methodology
5
Now.
We
are
talking
about
bytes
above
the
big
bang
But.
A
A
A
A
A
From
now,
on.
I
think
it
will
end
in
half
a
year,,
but
I
said
1
1
this
or
it
will
end,,
so
it
could
be
longer,,
and
if
this
happens
again,
there
is
a
situation
where
I
have
already
signed
a
contract
in
the
whole
frame,,
but
I
have
to
keep
extending
this
contract,.
Otherwise
the
contract
period
It
seems
that
the
prediction
is
measured.
Incorrectly,
and
more
penalties
are
imposed,,
and
this
case
is
a
problem
that
can
be
seen
more
often
than
expected.
The.
A
Only
thing
that
can
guarantee
is
that
a
new
big
bang
will
become
a
big
bang.
That's.
Why
Martin
Fowler
just
criticizes
Pippin
Wright
like
this
and
ends
the
story.
It
's,
not
a
question,.
He
himself
suggests
a
methodology
for
these
three
good
days..
Give
me.
The
picture
on
the
right
of
this
is
a
tree
called
String
Blush
Gra,,
which
in
Korean
is
a
strangler.
A
A
A
A
Blade
is
removed
like
this
for
5
males,
and
the
stem
is
lowered,
like
this.
Then,,
the
stem
gradually
sucks
in
nutrients
and
lowers
the
stem.
Finally,
when
this
stem
reaches
the
ground,
it
takes
root
in
the
ground,,
and
the
stem
that
has
come
down
so
far
is
used.
As
the
original
host.
It
is
said
that
it.
A
A
A
A
A
A
A
A
A
A
Applies
to
moving
a
specific
function
that
was
in
the
existing
monolith
to
a
new
service.
And
at
the
same
time,.
If
we
have
to
develop
a
new
feature,,
then
this
new
function
is
possible..
It
can
be
seen
that
this
content,,
which
is
created
as
a
new
small
unit
service
and
no
longer
added
by
Kumho
Logics,,
is
also
applicable.
A
A
A
A
So
now
we
dismantle
the
big
monolith
and
go
into
the
service.
And.
Another
notable
point
is
now
the
green
part.
The
green
blowfish
is
when
a
new
b
car
is
needed,.
These
services
are
added
to
the
new
environment
instead
of
going
into
the
existing
monolith
like
this
4
So,.
If
we've
been
talking
about
the
strategy
of
how
to
release
the
logics
into
microservices
or
smaller
scale,
services,,
this
time
we've
transformed
an
infrastructure
into
a
new
crowd.
I'm,
going
to
talk
about
the
river
for
a
bit
about
how
it
can
be
moved.,
It.
A
A
When
moving
an
infrastructure
to
a
new
location,,
there
are
a
lot
of
cases
where
the
existing
monolith
is
newly
repacked.
So,
in
my
experience,
I
dismantled
the
monolith
a
little
while
coexisting
with
the
two
infrastructures,
and
I
remembered
that
these
parts
were
fun,,
so
I
did
this.
I
thought
it
would
be
nice
to
be
able
to
show
you
how
to
convert
a
little
high
level.
5,.
A
So
I'm
going
to
do
a
simple
hub,
here.
So.
There
are
two
strategies:
I
prepared,
here.,
The
first
strategy
is
the
etn
shift
strategy
that
many
people
know.
As,
the
name
explains,.
The
lift-to-shift
strategy
raises
the
existing
infrastructure,
as
it
is,
and
the
reporter
picks
up
all
the
new
infrastructure.
When
we
go
to
the
mart,.
We
now
have
that
forklift.
I
entered
from
one
place
and
moved
to
another
place,,
but
you
can
think
of
the
image
of
my
lift
and
shift
as
a
bit
of
that
meaning.
A
A
A
Since
the
content
of
the
database
continues
to
change,
how
to
quickly
switch
over
this
database
and
move
it
to
a
new
place,
and
the
story
seems
to
have
progressed
since
then.
Now,
public
services
provide
data,
migration,
services,
and
the
complete
procedure.
There
is
a
way
to
keep
pushing
this
until
the
end,,
so
I
use
this
way.
A
lot.
For
now,.
There
can
be
many
ways,,
but
the
important
thing
is
to.
A
A
Everything
from
the
presentation
layer
to
the
database
part
is
done
in
1
code
base,
in
this
case,
everything
is
packaged
very
neatly,,
and
what
should
I
call
this
macro?
Should
I
call
it
a
container?
I
just
put
one
big
container
as
it
is
with
4
covers.
Then,.
The
current
service
continues
to
happen
in
the
monolith,
and
what
we
packaged
and
uploaded
to
the
Coburn
Lattice
is
now
a
new
infrastructure,,
tested,
and
now
works.
Well.
It
seems
to
work,,
but
there
is
now
a
request
at
the
top
on
the
front
page.
Well,.
A
A
When.
My
test
is
over,.
Now
I
cut
the
traffic
of
the
monolith
that
was
going
to
the
existing
one,.
So
when
this
new
quest
comes
in,,
I
move
it
to
the
new
old
Edith
Inkra,
and
Cream
n
Now,.
At
this
point,
when
I
started
receiving
all
these
quests
with
the
new
infrastructure,
Seoin
now
quickly
commissions
this
source.
In
the
existing
on-premise
data
center,
and
from
then
on,,
we
will.
A
A
A
A
A
A
Thought
it
would
be
a
little
easier
to
see
it
as
an
api
service,,
so
I
specified
it
like
this.
So,.
If
this
operation
continues
repeatedly,,
you
will
buy
the
size
of
the
monolith
on
the
left
little
by
little,,
and
you
will
think
of
more
and
more
services
in
Kubo
Netis
on
the
right.
They
are
not
necessarily
separated
from
all
existing
races,
and
new
services
can
continue
to
emerge.
So
in
this
case,.
Unlike
the
power
of
the
Big
Bang
rewrite
mentioned
earlier,,
it
is
a
store
that
can
continue
to
add
new
features
little
by
little.
And.
A
One
other
thing,
A
noteworthy
point
is
that
if
you
look
at
the
left
side,
you
put
a
c
layer
in
your
mind.
When,
operating
with
different
infrastructures
like
this,.
There
is
a
distance
between
infrastructures
and
a
torque
sum,.
So
if
this
switchover
is
to
be
carried
out
little
by
little
over
a
long
period
of
time.
In,
this
case,
it.
A
Seems
to
be
a
good
way
to
pay
attention
to
performance
in
terms
of
optimization,
by
giving
a
cache
layer
between
inputs
like
this,
and
to
read
the
long
term
with
the
mindset
that
the
long
term
will
go.
Slowly.
So,
this
business
continues,
As
it
progresses,
as
a
result,.
All
the
monoliths
on
the
side
of
the
juice
will
be
dismantled
and
moved
to
the
covernitus
on
the
right..
There
is.
A
A
A
As
you
can
see
in
the
5-minute
page,
it
means
a
bit
of
a
structural
framework,,
but
when
we
do
this
kind
of
monolithic,
3-body
work
for
a
large
organization
and
microservice
work,,
there
are
too
many
teams
and
the
methods
used
by
each
team
are
too
diverse.
Because.
It
is
different,.
There
can
be
a
lot
of
difficulties
from
the
point
of
view
of
operating
this
among
themselves,,
because
some
teams
want
to
use
a
certain
pattern
and
another
team
wants
to
try
another
pattern,.
So
we
minimize
this
confusion
and
the
most
centralized
pattern.
A
It
would
be
great
if
we
could
provide
something
like
this
structural
framework
to
make
it.
So,.
There
are
actually
many
ways
to
make
yarn
like
this,,
and
the
most
representative
way
is
to
use
the
2-turn
rsd
opportunity
before
the
code,.
So
people
have
a
lot
in
common.
The
method
that
makes
it
easy
to
create.
The
parts
that
are
used
is
also
a
method
that
is
widely
used.
A
A
A
A
And.
The
most
important
thing
is
that
mood
testing
is
very
important
now,
and
there
are
many
people
studying
what
kind
of
testing.
And
before
2
Protoss
One
Leaves,
board
testing
is
a
must
for
new
services.
It's,
not
just
that
Ron
is
born
when
starting,,
but
in
my
case,
among
the
b8
clines,,
the
seed
is
always
eating.
No
testing
during
testing
and
performing.
A
A
1000
won
Now,
in
the
past,
I
used
an
open
source,
Destin
toe
called
Locust,,
but
I
can
see
the
event
with
ui
like
this,
and
when
I
put
8
ply,
the
head,
ls
It
can
be
used
as
a
mode,,
so
there
is
no
such
tool
that
was
used
very
conveniently..
You
can
see
it
in
the
beginning
of
load
testing,.
And
if
you
look
at
the
right
side,,
the
scale
goes
up
depending
on
the
load,.
Then
the
skill
goes
down
again,
and
then
the
cpu
usage
at
the
time
was
combined
and
Crucially,.