►
From YouTube: Building an Enterprise-ready Lambda Experience (with Hands-On Demo) - Alexandre González Rodríguez
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
Building an Enterprise-ready Lambda Experience (with Hands-On Demo) - Alexandre González Rodríguez, BBVA Next Technologies & Héctor Rodes López, Adhara
To create a lambda experience is easy, to do that in a mission-critical environment is not that easy. We will talk about two parallel journies: how we implemented this at BBVA (an international 100 years old bank with more than 130 thousand employees, 10% of them being IT professionals). We´ll cover all the main decisions that we took in our path to serverless and the tradeoffs that we accepted. As a complement to this, we will implement in stage a simpler version of this experience using out of the box Kubernetes features like CRDs. There are certain thoughts that you need to keep in mind when you are planning a serverless solution in your company: usability, extensibility, security, resiliency, to use a market solution or build our own, logs, metrics… in the end, how do we make this tool part of our toolset seamlessly and production ready from the first minute.
https://sched.co/MPeC
A
Hello,
everyone,
so
we
are
going
to
talk
about
lambda
and
say
how
we
will
a
lambda
experience
and
how
you
can
build
your
own
lambda
students
at
home.
Sorry,
not
not
me.
Hector.
First
actor
is
a
CTO
Vadodara
and
say:
I
am
Alex
Gonzalez
and
I
head
up
back
at
BBVA.
Next
and
I
am
currently
working
at
BBVA,
which
is
a
bank.
So
first
we
want
to
start
like
breaking.
Myth
is
250
dryness
pain
and
nobody's
taking
siesta.
So
please
don't
fall
asleep.
Today.
Ok
with
us.
A
We
end
up
with
a
10th
of
last
year
and
then
you
left
in
January,
but
he
was
so
involved
and
I
am
so
involved
and
our
thing
he
is
so
involve
as
well
in
lambdas.
A
little
bit
of
context.
Bbva
is
a
is
a
bank
it's
from
the
financial
services
industry.
The
presence
is
in
in
more
than
10
countries.
We
have
data
centers
on
premise.
We
are
using
the
clouds,
of
course,
but
we
have
on
premise
important
data
centers
in
Europe
and
in
America,
and
we
have
around
10,000
IT
professionals
from
these
10,000
IT
professionals.
A
A
A
B
Okay,
first
thing
is:
okay,
we're
talking
about
the
enterprise,
really
it's
a
it's
a
difficult
term,
just
plain
I,
think
or
maybe
too
easy,
but
I
usually
try
to
look
at
Wikipedia.
If
you
find
that
term,
there
probably
is
the
best
definition
you
can
find
in
our
case,
I
can't
remember
if
we
didn't
find
that
that
term
in
a
Wikipedia
it
was
something
like
a
crappy
thing
that
you
cannot
use,
or
it's
not
meaningful.
B
C
B
Every
it's
something
that
we
cannot
discuss.
This
is
entered
verbally
and
after
that
very
shoot
money,
of
course,
but
I
think
money
is
everywhere.
So
it's
not
only
about
enterprises,
and
this
thing
that
I
don't
understand
too
much,
but
it's
funny
so
trying
to
figure
out
what
Enterprise
really
is,
or
at
least
what
we
understood
when
we
were
building
all
these
kind
of
services.
Is
that
when
you
try
to
build
something
that
need
to
be
deployed
or
to
serve
to
company,
probably
you
have
some
kind
of
Industry
constraints.
B
In
our
case,
it
is
well
mainly
financial.
It's
a
financial,
regulated
industry,
so
security
confidentiality
that
allocation.
That
was
something
a
kind
of
nightmare.
As
an
example,
Meru
can
store
data
in
Mexico,
but
Mexico
only
can
store
data
Mexico
and
sometimes
can
store
data
in
Spain,
but
Spain,
yes,
but
not
France.
It's
an
adverse.
So
again,
it's
a
lot
of
things
that
you
need
to
solve
that
every
application
of
this
bill
in
the
bank
that
will
be
provide
service
to
our
clients
need
to
be
solved.
B
So
it
was
worried
to
saw
in
the
foundation
and
all
the
services
that
they
were
building.
So,
instead
of
spending
a
lot
of
time
and
money
trying
to
solve
this
on
every
application,
it
was
very
soldiers
in
the
basis,
of
course,
depending
on
the
company,
you
are
going
to
have
some
rules
to
follow
some
tools
to
to
use.
In
the
case
of
bit
of
VBA,
you
know
it's
100
years
company,
a
lot
of
things
are
there.
You
need
to
use
those
things.
Some
of
them
are
awesome.
B
Some
of
them
were
awesome
at
some
point,
maybe
two
new
years
ago,
not
now
and
at
the
end
you
need
to
integrate
with
other
really,
maybe
it's
not
about
the
core
system,
but
it's
difficult
to
make
something.
That
interest
interacts
perfectly
with
with
everything
so
and
talking
about
the
land
experience
or
the
experience.
One
of
the
lessons
we
learn
failing
a
lot
in
the
in
the
first
years,
probably
the
two
first
years
of
our
developments.
B
We
spend
a
lot
of
time
trying
to
solve
all
the
security
issues,
all
the
things
that
were
related
with
being
a
financial
industry
service
and
what
we
provide
to
our
developers
was
something
that
maybe
it
was
not
the
best
experience
or
it
was
not
so
useful
for
them.
They
used
our
tools,
but
it
was
like
because
we
need
it.
We
don't
have
another
option.
Maybe
it's.
If
you
can
build
something
different,
it
will
Liber
and
we
were
factor
everything.
We
were
build
out
our
services
from
scratch.
B
Those
services
is
what
what
I
expect.
So
we
decide
to
to
make
this.
We
want
to
the
land
experience
for
our
developers,
similar
to
public
cloud,
offering
yes,
that
simple,
maybe
simple,
to
explain
not
so
simple
to
implement
and
based
on
this
or
these
two
principles
have
any
mind
all
the
things
that
we
need
to
solve.
That
is
related
to
be
financial
industry,
BBA
related
to
development
and
trying
to
offer
this.
We
started
at
Jordan
that
we
are
going
to
try
to
explain
you
so
how
we
did
it.
A
So
as
a
sector
mentioned,
the
first
thing
that
we
wanted
to
do
with
this
service
was
offering
a
proper
experience
when
you
use
it
so
we're
on
the
bank.
We
probably
know
some
windows
with
a
lot
of
input
fields
and
all
that
at
the
API
it
looks
more
or
less
the
same,
but
engaging,
and
with
this
with
this
UX,
we
were
trying
to
simplify
that
so
at
the
tunnel.
We'll
have
this
contact
with
the
our
clients,
with
our
developers
from
the
bank.
A
We
are
not
going
to
change
that
and
all
the
complexities
inside
so,
for
example,
an
API,
a
lambda
function
for
us
looks
like
this.
If
you
are
familiarize
with
lambda
sing,
AWS
is
the
same.
The
only
thing
that
changes
in
the
ISA
context-
okay,
but
the
idea
is,
you
have
a
class
and
this
class
has
a
function.
This
is
going
to
be
then
two
point
for
the
for
the
lambda
function.
One.
A
Yeah
one,
you
need
to
start
with
Java
if
you
want
users,
so
the
thing
is
we
are
going
to
have
an
output
which,
in
this
case,
for
the
example
is
in
string
and
we
are
going
to
have
an
input
which,
in
this
case,
is
in
string
as
well.
Okay,
the
context
is
something
that
we
use,
for
example,
if
we
want
to
to
allow
you
to
connect
with
other
services
from
this
ecosystem,
so
you
can
just
use
the
context
and
create
a
client
for
for
producing
these
ecosystems
without
dealing
with
it.
A
A
This
looks
more
or
less
like
this.
This
is
the
real
one
for
the
example
that
we
are
going
to
show
later,
it's
a
little
bit
more
simple,
because
we
don't
have
a
namespace,
and
the
big
difference
here
is
that
everything
that
you
want.
You
want
to
become
a
function,
every
piece
of
code
that
you
want
to
become
that
becomes
a
function
in
the
future.
It's
going
to
go
through
a
pipeline.
Okay,
you
cannot
say
please
they
play
this
code.
A
For
me,
we
are
going
to
go
to
the
pipeline
and
this
pipeline
is
going
to
check
the
security
concerns
that
we
can
have
is
going
to
have
traceability
and
a
lot
of
other
things
that
it
kind
of
looks
like
a
Greenfield
is
lambda
service,
but
the
dent
is
a
brownfield
because
we
need
to
interact
with
all
these
burning
pieces
that
we
have
already.
Okay,
okay,
so
I
forgot
me
to
show
you
this.
This
is
a
real,
a
real
function.
Okay,
we
have
some
out
data.
A
A
Don't
the
other
things
that
we
have
is
getters
and
setters,
because
you
know
Java,
and
this
is
the
experience
that
we
want.
Okay,
we
have
some
code,
we
push
it
somewhere,
something
happens
and
you
are
going
to
get
back
a
URL,
and
this
URL
is
going
to
be
your
function
service.
You
don't
care
about.
How
many
instances
do
you
have
at
the
of
the
function
or,
if
you,
if
the
function
is
sleeping
or
whatever
you
just
have
a
URL,
that
you
call
and
the
function
is
going
to
work.
A
This
is
for
the
for
the
sake
of
the
demo.
We
simplify
it
a
little
bit
okay,
so
instead
of
having
these
pipelines
that
are
going
to
build
a
function
for
you,
this
this
example
just
receives
a
field
which
is
the
code.
Oh
sorry,
which
is
a
proper
function
that
you
want
to
run,
and
we
say
then
to
point
from
this
function,
which
is
a
class
and
the
method
that
you
want.
That
is
going
to
be
done
to
point
in
the
bank.
A
What
we
have
are
things
like,
for
example,
the
stack
which
limits
the
number
of
resources
that
you
can
consume
or
make
some
other
configurations
that
you
need.
You
don't
need
to
take
to
take
care
of.
We
do
it
for
you,
so
our
second
decision
after
we
decide
that
we
want
this
appropriate
UX,
and
we
have
this.
This
contract
sign
with
our
clients
is
what
do
we
have
inside
and
for
what
do
we
have
inside?
A
We
were
checking
what
we
had
on
the
cloud
native
computing
foundation
and
other
in
other
projects,
or
we
can
decide
and
go
with
our
own,
so
the
state
of
the
art.
At
that
time,
which
was
last
year,
we
were
evaluating
open,
Wisco,
confess,
Connecticut,
less
and
some
other.
Some
of
them
are
really
good
and
they
were
offering
part
of
what
we
wanted,
but
not
all
of
it.
So,
for
example,
for
us
was
really
important
to
deal
with
multi
region
and
multi-tenancy
the
way
BBVA.
Does
it
and
said
to
integrate
with
all
these
pipelines?
A
Remember
the
partners
were
already
there.
We
don't
want
to
change
all
the
bank.
We
can
just
change
piece
by
piece
compatibility
with
the
RPC
and
some
some
other,
so
we
already
had
the
implementation
of
something
like
Google
App
Engine,
which
was
dealing
with
some
of
this
project
of
this
problems
already
like,
for
example,
logging
tracing
security
and
so
on,
and
we
had
some
certified
stacks
like
you
can
just
run
Yama
in
this
version
with
this
library
and
whatever.
So
we
put
them
together
and
we
end
up
making
a
homemade
project.
So
this
is
our
decisions.
A
Okay,
first
one
we
took
the
UX
and
we
we
offer
this
proprietary
to
the
client.
We
are
instead
creating
something
from
scratch,
reusing
another
tool
that
we
create,
which
is
App
Engine
and
we
are
using
the
certified
stack
setting
hot
all
right
at
home
and
they
are
certified
by
security
and
all
the
other
pieces
of
the
bank
and
the
koala
becomes
now
like.
Imagine
that
cannabis
to
comes
to
the
market
and
fixes
all
the
problems,
not
problems
really
fixes
all
our
concerns.
We
can
start
using
it
and
nothing
is
going
to
change.
A
So,
basically,
if
we,
where
on
greenfield,
wanted
to
build
it
wanted,
say
euro,
but
the
context
was
this
so
8-bit
Super
Mario
is
fine,
it's
a
really
fun
game,
and
probably
some
of
you
play
it
and
I.
Don't
know
if
there
are
some
Millennials
here,
but
the
thing
is
I
am
millennial.
No,
it's
not
the
like
about
him
being
a
millennial.
A
A
B
Next
thing
we
we
started
thinking
about
it
was
about
the
control
plane.
What
I
mean
when
I
say
in
control
plane
is
at
the
end.
We
are
deploying
a
function
for
you,
so
we
need
to
to
manage
the
lifecycle
of
that
function
to
a
start-up
server
or
some
kind
of
binary
or
whatever.
That
will
execute
what
you
have
built
and,
of
course,
dealing
with.
Maybe
it
need
to
be
scared,
or
maybe
it's
a
struggle.
We
need
to
do
a
start-up
again.
Of
course
we
will
reuse
in
communities.
B
So
almost
everything
was
really
easy
based
on
that,
but
at
the
end
it
was
something
that
we
need
to
deal
about.
Let
me
use
this
slide,
because
I
think
it's
easier
to
explain
with
this.
We
have
two
services
running
that
looked
more
like
the
left
side,
so
at
the
end
would
be
the
control
plane
that
was
interacting
with
a
lot
of
a
lot
of
server
on
things
around
gateways,
load,
balancers,
firewalls,
the
kubernetes
itself,
deploying
things,
and
after
that,
with
a
simple
URL.
You
just
call
that
you
are
more
or
less.
B
But
when
we
started
I
started
to
think
on
on
offering
lambdas,
we
realized
that
probably
we
need
to
make
an
extension
of
the
control
plane
that
we
call
collar
managers
that
the
end
was
okay.
Maybe
when
you
are
calling
a
lambda,
we
want
to
offer
an
improved
experience
as
an
example.
Dealing
with
okay,
okay,
I
can
a
slip
to
lambda,
but
probably
when
you
call
that
land
we
need
to
wake
up
again.
So
in
that
case,
probably
everything
should
be
managed
by
by
some
peers
that
interact
with
the
control
plane.
B
It's
an
example
to
wake
up,
or
on
your
case,
to
in
the
in
the
way
we
decide
to
to
build
of
learners.
When
you
code
that
that
code,
we
are
not
dealing
with
with
any
protocol.
It's
just
data,
it's
pure
data,
so
it
doesn't
matter
if
it's.
If
it's
GRP
see
if
it's
it's
an
HTTP
request
or
just
something
call
from
a
queue.
So
you
are
completely
agnostic
so
that
color
manager
allowed
asked
to
offer
that
that
experience
to
our
developers,
so
it
was
one
of
the
of
the
biggest
changes.
B
A
So
we
have
simplified
all
this
and
we
put
all
the
pieces
together
to
make
the
system
okay.
This
is
this
code
is
going
to
be
available
later
on
github.
If
you
want
to
play
with
it,
so
we
have
define
a
type.
This
type
is
going
to
receive
the
code
and
then
to
point.
The
entry
point.
Remember
is
the
class
on
the
method
and
because
it's
all
all
your
function
internally,
we
are
storing
it
with
an
ID
because
to
call
the
function,
you
need
an
ID.
This
is
all
in
memory.
A
Okay,
we
don't
even
have
a
storage.
Is
a
map
map
a
string
function
or
something
like
that
and
we
have
a
URL.
The
URL
is
something
that
our
users
aren't
going
to
see.
The
URL
is
something
that
we
save
to
call
the
function
from
this
caller
manager,
so
the
the
functionality,
let's
say,
is
there,
but
they
are
not
two
different
services,
all
the
same
service.
A
So
if
you
see
the
matters
that
we
create
it
can
you
see?
Can
you
see
well
at
the
back
yeah?
Okay
awesome,
so
we
have
a
handler
for
create
function
which
is
going
to
receive
this
model.
We
have
a
handler
to
get
the
function
which
is
going
to
show
you
that
model
except
the
URL,
and
we
have
a
handler
for
calling
the
function.
The
down
important
bit
now
that
we
are
talking
about
the
control
plane
is
the
the
creation
of
the
function.
A
The
creation
of
the
function
is
just
going
to
do
this
on
cornet
we
are,
we
are
using
kubernetes
behind,
but
our
usage
doesn't
know
that
we
are
using
cornet
in
the
bank.
We
are
using
OpenShift,
because
I
know
that
there
are
some
people
from
record
there
and
the
thing
is
we
created
deployment
after
we
create
the
deployment
we
create
a
service
and
then
later
on,
we
are
going
to
store
the
URL
of
that
service
for
calling
the
function
and
that's
it
from
wander
playing.
B
Chord,
this
is
just
something
parametric
to
your
crowd.
We
still
need
to
execute
the
that
lambda
function
and
the
next
thing
to
think
about
it's
about
executing
a
stack.
We
are
started
with
with
Java.
It
was
quite
using
in
the
bank,
so
it
looks.
It
looked
like
the
best
option
and
at
that
point
after
that,
dealing
with
with
developers,
they
told
us
that
they
prefer
Python
or
maybe
not
yes,
but
ok.
B
We
needed
to
start
with
with
something
so
more
or
less.
The
executing
stuff
looks
like
this
interesting
part
is
that
we
try
to
be
as
immutable
as
possible.
Of
course,
if
we
will
need
to
inject
something-
and
we
don't
have-
we
don't
want
to
have
the
image
revealed
that
it,
it
was
something
that
we
don't
want.
It
didn't
want.
Sorry,
we
need
something
that
we
call
here.
The
init
container
was
trouble
at
the
end.
It
was
just
a
binary.
B
It's
a
really
small
piece
of
code
that
was
in
charge
of
dealing
with
some
credential
get
the
the
binaries.
In
this
case,
DB
was
jars,
but
can
be
Python
code,
not
just
an
NPM
package
or
whatever,
to
download
it
and
to
place
it
in
the
in
the
right
place.
So
the
lambda
server
can
that
code
and
I
started
security
in
that
mainly
more
or
less.
The
flow
was
okay
download.
The
thing
in
this
case
an
example
that
we
are
going
to
show
you
it
it's
even
simpler,
because
the
bootstrapper
is
just
compiling
the
code.
B
B
You
cannot
execute
anything
that
is
at
that
function,
make
some
verification
about
the
signature
of
the
function
and
if
it's
there,
if
not
it
crosses-
and
you
cannot
do
anything
and
at
the
end
internally
offers
two
endpoints
that
it's
a
post
and
I
get
they
get
HTTP
endpoint
is
just
to
be
sure
that
it's
up
and
running
and
the
post
is
to
execute
the
function
as
you.
As
you
remember,
we
had
something
called
the
color
manager,
the
protocol
between
the
color
manager
and
this
server
is
internal
JSON,
RPC
protocol.
B
It's
just
to
kind
of
messages
that
we
support,
so
one
the
color
manager
calls
the
function
that
it's
the
NS
received
by
this
server.
There
is
a
listener,
get
the
message:
parse
it
try
to
try
to
DES
realize
the
everything
into
an
object,
try
to
execute
the
function
if
it
took
if
it's
trans,
okay,
prophet
all
the
information,
is
returned
to
the
user.
If
not,
it
returns
error.
It's
quite
simple!
Why
we
did
in
this
way
it
was
because
okay,
we
started
with
Java,
but
we
wanted
an
easy
way
to
to
add
new
languages.
B
Python
ODS
were
probably
the
the
most
demanded
one
or
maybe
go.
That
was
our
preferred
language
in
our
team.
So
we
decided
to
invest
a
lot
of
time
in
the
experience
endo
in
those
interfaces,
so
we
were
not
coupled
with
Java
or
the
way
would
be
able
to
build
the
things.
So,
let's
show
you
a
little
bit
on
how
this
was
implemented
for
this
demo,
yeah.
A
So
again,
for,
for
the
sake
of
the
demo,
we
simplify
this
a
little
bit.
We
are
not
going
to
have
the
neede
container
with
the
bootstrapping.
We
are
not
even
going
to
download
anything.
A
sector
was
saying
we
mainly
get
the
code.
So
it's
this
one
here,
so
we
we
get
the
code
that
the
control
plan
and
playing
was
receiving
and
we're
going
to
compile
it
after
we
compile
it
with
reflection,
we
are
going
to
create
two
instances:
an
object
of
the
class
on
the
entry
point.
A
This
is
what
the
finishing
HD
toward
us-
and
this
is
the
important
bit
so
important
things
here.
This
request,
patient
reader
is
instantiated
with
the
type
that
the
that
was
your
your
input
param.
So
if
your
input
parameters
a
string-
and
here
you
are
trying
to
read
an
integral,
this
is
going
to
file
the
racing
exception
and
we
are
going
to
return
an
error,
but
int
or
string
is
just
a
simple
type,
but
we
can't
have
there
whatever
you
want.
A
If
we
can
deserialize,
it
is
going
to
be
okay
and
we
are
going
to
call
the
function
if
we
kind
of
the
serialize
it
we
are
not
going
to
call
the
function,
then
we
invoke
the
the
method.
The
method
is
the
full
path
of
the
entry
point
with
the
parameters
on
the
context,
and
then
we
write
the
value
when
we
write
the
value,
we
are
just
say,
realizing
the
output
type
of
the
function,
and
that's
mean
it
here.
You
want
me
to
mention
anything
else.
Maybe.
A
D
D
A
A
It's
a
starter
function.
Server.
This
again
is
the
control
playing
on
the
execution
playing
all
together.
Okay
and
we
are
going
to
deploy
a
function
first
with
the
player
function
and
then
I'm
going
to
show
you
what
we're
doing
so
is
the
example.
The
example
lambda
function
that
we
create
before,
but
what
we
are
going
to
be
doing
is
setting
foo
and
setting
bar
and
then
returning
phone
bar
all
together.
C
A
You
hear
that,
but
basically
what
it
says
is
the
result:
I
love
it
with
my
demo
comes
together,
which
is
a
concatenate
in
the
two
strings
and
that's
it.
We
had
a
plan
B
because
when
you're
working
a
bang,
you
know
that
you
always
need
to
have
a
plan
B,
but
it
wasn't.
It
wasn't
needed
the
demo
work
at
this
time
and
thank
you.
If
you
have,
if
you
want
the
code,
is
there.
D
B
A
summary,
as
a
summary,
maybe
two
or
three
main
topics,
focus
on
the
user
experience
it's
one
of
the
lessons
that
we
learn
and
we
it
took
two
years
until
we
understood
that
and,
as
you
can
see,
it's
quite
easy
to
implement
something
that
looks
like
a
lambda.
Probably
the
most
difficult
part
is
to
integrate
all
these
things
in
your
ecosystem.
Probably
you
have,
if
you
can
choose
open
files
or
Canada
or
whatever,
and
you
can
extend
or
make
the
minimal
extension
to
to
interact
with
with
your
system
is
a
very
solution.
B
B
E
Okay,
so
my
question
is
as
far
as
I
can
see
or
I
know
the
main
problem
with
this
swamp
that
kink
in
Amazon
is
the
cold
starts.
What
I
mean
is
that
when
I
start
a
function,
it
may
take
200
milliseconds
to
boot
or
maybe
5
milliseconds
to
boot.
It
depends
on
on
Amazon.
So
do
you
have
such
problem
in
your
application?
I'm,
not
pretty
sure,
because
you
have
a
Java
app
and
if
you
have
to
boot
up
the
JVM
each
time
when
a
function
is
called
I.
Think
it's
just
too
much.
Yeah.
E
A
B
F
B
The
the
spin-up
time
was
not
the
really
important
thing,
but
basically,
if
you
see
the
original
implementation,
the
one
that
is
deployed
at
the
EBA
as
an
example,
the
Java
function
deployed
from
scratch
clean
new
brandy,
kubernetes
cluster.
It
was
like
one
minute
because
of
the
cache
of
the
images,
after
that
it
was
like
less
than
a
second.
That
was
why,
because
we
placed
a
lot
of
CPU
there
once
we
started
thinking
because
more
or
less
lambda
was
built
like
bronzes
service,
is
you
if
you
remember
our
our
first
or
second
slide?
B
We
have
continuous
service
appengine
lambda
lambda
was
like
the
bronze
one,
so
we
were
placing
like
10%
of
20%
of
CPU
and
it
took
like
30
to
40
seconds
to
spin
up
are
really
simple,
lambda
function,
so
it
depends
on
the
resources
we
were
turning.
We
decided
to
okay,
it
will
be
a
slow,
but
it
was
what
we
needed
a
in
our
case
as
an
example,
we
can
define
at
some
point
some
kind
of
goal
function
with
a
lot
of
CPU,
and
in
that
case
it
will
be
faster.
It.
A
B
B
We
started
off
into
the
user,
so
roll
it
there
at
some
point.
Part
of
the
team
is
here,
so
you
can
ask,
maybe
about
the
latest
version,
but
probably
will
be
improved
in
the
future.
Based
on
real
matrix
would
be
a
has
a
really
good,
matrix
aggregation
system,
so
at
some
point,
I
think
they
will
use
all
the
same.
All
that
information
to
make
a
unimproved
a
skill.
Now
it's
really
simple.
D
D
B
A
Okay,
so
the
thing
is
that,
for
us,
this
project
was
not
a
greenfield
project.
You
cannot
get
like
kubernetes
and
offer
it
to
people
from
the
bank
to
use,
or
you
cannot
get
some
private
like
open
path,
which
is
pretty
cool
and
getting
the
people
to
use
it.
It
needs
to
integrate
with
a
lot
of
things
and
in
the
bank's
way
as
well.
We
are
never
going
to
show
you
what
is
behind,
so
we
always
try
to
avoid
locking
by
offering
an
API
on
top
of
it.
A
B
And,
and
also
take
into
account
that
it's
not
the
same,
you
need
to
build
something
from
scratch.
Without
experience,
that
was
not
our
case.
You
know
you
have
30
developers
and
people
that
is
building
these
kind
of
services
and
this
kind
of
tools
and
have
something
that
it's
quite
similar
at
the
end,
it's
yes
to
to
make
any
service
something
that
is
quite
simple.
That's.