►
From YouTube: Serverless on Kubernetes: What We Learned From Customers - Eduardo Laureano, MicrosoftSDCC
Description
Serverless on Kubernetes: What We Learned From Customers - Eduardo Laureano, MicrosoftSDCC
Serverless platforms managed by cloud providers are well established and growing steadily in production usage. In contrast portable serverless platforms in Kubernetes is a much newer space and components are still evolving. As a vendor providing both options, we have been comparing and contrasting customer use cases, features and patterns. This talk presents what we learned from Azure customers that are processing billions of events monthly on mission critical scenarios.
A
So
I'm
here
to
talk
about
service
on
kubernetes
what
we
learned
from
customers.
In
this
case
we
is
Microsoft
Microsoft
Azure,
that's
where
we
work.
I
was
actually
Barcelona.
The
first
edition
of
this
event
and
I
thought
it
was
great
to
see
all
the
technologies,
the
community
contributing
to
the
kubernetes
system,
but
I
felt
like,
as
we've
been
in
this
space
for
a
while
now
and
we
didn't
have
a
chance
to
kind
of
contribute
to
really
say
what
are
we
building,
what
we're
doing
and
what
we
learned.
So
that's.
A
Why
I'm
here
to
talk
about
this?
How
many
of
you
word
cube
con
attended
server.
Last
prediction:
early:
let's
see
it's
about
less
than
half
here,
but
great
event,
and
even
today,
you've
seen
some
of
the
text
know.
Some
of
the
questions
are
really
interesting
and
I
hope
to
actually
tackle
some
of
them
here
before
I
go
too
far,
so
I
lead
some
of
the
Microsoft
service
initiatives
in
particular,
I
lead.
The
product
team
fresher
functions
because
you're
not
familiar
with
it.
I
got
a
shirt
with
the
logo.
A
So
now
you
know
what
it
is:
you've
never
seen
it
and
for
next
30
minutes.
What
I
would
like
to
do
is
tell
what
we
learned
from
customers,
but
go
a
little
bit
a
little
further
back
our
journey
where
we
are
in
the
organization.
We
saw
the
application
paradigms
evolving
over
time
and
customers
enabling
more
and
more
scenarios
so
covering
each
paradigm.
What
customers
liked,
what
they
didn't
like,
what
the
gaps
were
and
then
our
move
hang
with
me.
A
I'll,
take
you
guys
on
a
journey
and
I'll
get
the
kubernetes
towards
the
middle
of
the
presentation.
What
are
we
doing
in
kubernetes
and
what
we
think
we
can
even
do
further
and
how
we
can
all
collaborate,
we're
here,
to
chat
and
to
learn
as
well,
but
but
there's
some
stuff
that
we
bring
to
the
tables
Microsoft.
That
I
think
is
really
interesting.
A
A
A
We
launched
a
product
called
as
your
website's
later
become,
as
your
web
space
sites
later
become,
as
your
app
service
I'm
now
going
to
talk
about
our
branding
apartment,
but
Stewart
is
there
today
very
popular,
but
because
we're
talking
service
I
want
to
talk
about
how
we
delivered
how
he
activated
applications.
The
first
thing
in
that
service
for
journey
we
did
was
HTTP
serving.
A
Essentially,
you
could
bring
your
applications
to
essential
to
our
our
cloud
web
servers
and
we'll
do
the
load
balancing
for
you
and
I
explained
how
we
work
in
the
cloud
in
a
very
high
level,
diagram
of
the
architecture
we
have
on
the
blue
box.
We
have
our
infrastructure,
has
front
ends
and
then
green
are
workers
where
we
run
your
code.
We
run
the
customers
code,
there
are
applications.
A
All
of
those
are
in
typically
I
didn't
make
it
to
scale,
but
there
are
thousands
of
machines,
literally
over
50
data
centers
around
the
world
and
they're
in
scale
units
so
allows
for
us
to
give
sort
of
that
infinite
elastic
scale.
Back
then,
when
service
foe,
you
could
say,
hey
I
want
to
have
a
plan.
I
want
to
put
my
application
in,
let's
say,
1
to
10
20
nodes.
You
would
tell
us
where
you
how
many
nodes
you
want
an
application
to
run
what
we
took
care
for.
You
was
how
to
sort
of
route.
A
A
We
could
so
that's
how
we
started,
but
we
also
built
a
bunch
of
shared
services
and
then
we
could
provide
additional
functionality
like
makeout
authentication,
much
easier
talking
to
other
services
without
having
to
learn
tons
of
api's,
provide
security,
the
ability
of
adding
certificates
and
so
on
so
forth,
because
our
mission
was
abstract.
You
from
things
that
you
shouldn't
need
to
worry
about
with
that
and
now
I'll
sprinkle,
some
of
what
the
customers
did
at
each
stage
that
we
developed
those
products.
A
Wonder
I
want
to
talk
about
this,
one
of
the
largest
TV
networks
in
Canada,
they're,
running
the
elections
and
as
the
candidates
were
talking
on
the
screen,
they
want
to
feel
a
sentiment
whether
the
population
watching
were
favorable
or
not
favorable
to
some
of
those.
So
they
would
put
a
link
a
short
link
there
right
there
as
the
broadcast
was
going
on,
that
you
could
click
and
you
could
you
could
answer
a
poll.
I,
don't
know
a
whole
lot
about
Canadian
election.
A
So
spare
you
of
those
details
but
I
know
the
technical
part,
which
is
they
were
able
to
and
might
be
a
little
small
but
they're
able
to
get
even
back
then
close
to
1
million
requests
per
second
on
that
type
of
architecture,
which
is
all
that
the
company
had
to
do
was
deployed
your
code.
All
this
scaling
all
the
nice
functional
certificates
and
all
that
was
all
taken
care
of
for
them.
A
So
we
kept
going
on
service
for
when
we
did,
we
did
a
lot
more.
We
did
out
of
scale
and
looking
back.
That
was
amazing
that
we
could
detect
on
your
cpu,
was
at
80%
and
add
another
machine
or
skill
based
on
memory
or
HTTP
key
went.
We
started
multi-language
from
the
get-go
back,
then
there
are
tons
of
PHP
developers,
I
hope
not
as
many
today,
but
joking
aside,
we
started.
A
We
had
Java,
we
had
Python,
etc
C
ICD
before
DevOps
as
such
as
big
of
a
thing
as
it
is
now,
zero,
downtime
deployments
and
so
on
and
so
forth.
So
service
full
super
popular
to
this
day
as
rep
serve
is
the
third
most
popular
service
in
Azure.
That's
how
most
of
the
people
build
sort
of
applications
in
Azure,
then
in
2014,
which
turned
on
the
event-driven
journey
is
when
we
notice
not
everything's
HTTP
in
HTTP
in
is
not
the
best
protocol
for
you
to
implement
all
applications
out
there.
A
A
You
cannot
change
their
entire
ass.
You
cannot
access
the
registry
you'd
stay
at
that
high-level
abstraction,
but
given
that
you
can
have
this
concept
of
web
jobs
and
would
be
integrated
with
your
source,
you
could
do
your
CI
CDs,
just
like
you
normally.
Would
you
get
all
the
other
service
full
properties
of
it?
And
then
the
pattern
that
we
saw
emerging
was
a
lot
of
these
websites.
Now
they
will
try
to
respond
as
quickly
as
possible
to
their
customers,
so
what
they
will
do
is
they
will
do
the
synchronous
workload?
Would
the
website
would
respond?
A
Anything
could
be
done
later
would
be
entered
in
a
queue
to
be
later
processed.
Thank
you.
The
web
jobs
would
pick
up
an
item
from
that
queue
and
process.
It
asynchronously
and
then
do.
Let's
say:
AI
is
a
very
common
example.
The
web
site
would
log
data
a
web
job
would
pick
some
say.
Haters
a
new
entry
in
the
log
would
pick
that
log
for
processing
structure,
a
data
processor
may
I.
A
Now
you
have
some
intelligence
that
you
didn't
have
before
for
that
to
happen,
for
us
to
be
able
to
read
from
the
key
we
implemented.
Also,
we
released
also
the
web
jobs
SDK,
essentially
something
that
can
knows
how
to
talk
to
all
these
different
events,
source
types
and
know-how
activate
your
workload.
So
that's
when
we
start
on
this
event-driven,
and
that
was
almost
all
the
elements
we
need
to
have
a
full
something.
We
could
call
a
function
as
a
service
offering
as
well
in
service
and
that's
when
we
released
as
your
functions
came
around.
A
We
increase
the
amount
of
triggers
that
we
that
we
had
on
the
platform
I'll
show
some
of
them
later.
This
was
a
that
is
a
massive
difference.
Is
the
event-driven
scale
so,
instead
of
scaling
now,
because
your
machine
is
constrained
on
CPU,
we
would
look
at
those
events
predict
what
you
need
as
a
workload
and
then
allocate
your
code
as
many
machines
as
you
need
it
once
you
don't
need
them
anymore
scale
them
back
down
to
zero
again
folks,
who
are
using
trees
to
glue
applications
together.
A
That's
probably
still
to
this
day
the
main
use
of
service
that
I
see
in
extend
your
applications.
A
lot
of
customers
already
have
either
their
monolith
or
something
already
deployed
when
they
ask
us
how
they
get
started
with
service.
We
tell
them,
extend
your
app,
we
don't
typically
go.
Tell
them,
go
rewrite
everything
that
you
already
have
deployed.
A
I
think
the
best
entering
to
serve
Allah
says
find
something
that's
event
based
and
your
code
there
and
extend
that
way
and
before
I
talk
about
the
functions
you
might,
how
many
of
you
tried
functions
are
played
with
functions
before
the
entire
room
raised
their
hand.
No
I'm
kidding
just
about
half
actually
so
so
show
real
quick
for
those
of
you
never
seen
as
your
function
so
because
our
journey
was
always
developer
productivity.
You
have
a
whole
experience
on
visual
studio,
but
some
folks
don't
even
have
visual
studio
code
installed.
A
A
So
once
you
are
in
Visual
Studio
and
you
have
functions,
you
get
all
these
nice
commands
where
you
can
start
a
project
and
then
you
can
pick
one
of
the
languages.
Those
are
the
languages
support
and,
as
your
functions,
I'm
gonna
pick
JavaScript
and
you
can
pick
any
of
those
trigger
types
here.
So
I
mentioned
the
support
large
amount
of
triggers,
so
you
can
see
that
list
there
and
I'll
pick
you
first
to
see
you.
This
is
the
name
of
the
function,
just
Mike
for
my
own
code.
A
So,
if
I
go
look
at
the
files
here,
there's
one
setting
then
I
have
to
set
here,
which
is
this
one
is
to
talk
to
an
actual
queue
for
the
demo
to
work.
I'm
gonna
use
the
best
clipboard
in
the
world,
which
is
notepad.
It's
also
Microsoft
Project.
Okay,
that's
so
I
configured
I
configure
this
to
talk
to
actual
the
storage
account
that
I
have
the
only
other
thing
to
to
call
out
your
attention.
Is
a
functions
also
configuration
so
that
Keeler
we're
going
to
talk
to
you
is
in
this
JSON
file.
A
Here
it
says
the
queue
name.
That's
variable
that
I'm
going
to
read
from
no
word
connect,
so
once
I
do
that
you
could
debug
these
functions
if
I
click
debug
here
and
I'm
gonna,
once
you
saw
it
was
really
quick,
but
it
wants
just
so
the
ASCII
art,
just
like
my
shirt,
the
functions
logo
means
that
the
runtime
is
running.
Let's
run
that
version
of
runtime
that
runs
on
Visual
Studio
code
online
or
Visual
Studio
code
installed
on
my
machine
or
Visual
Studio
or
the
clouds.
A
The
exact
same
version
is
dotnet
course
or
behaves
the
exact
same
way
wherever
you
go
so
I'm
gonna
set
a
breakpoint
here,
so
we
can
debug.
This
function,
I
have
on
my
edge
browser
I.
Have
my
cue
that
I
wanna
write
you
so
once
there
is
a
message
on
that
cue
and
it
doesn't
matter
what
I
write
here
but
I'm
just
so
you
know
it's
demoing.
It
live
once
I.
Add
an
item
to
the
cue
because
that
function,
that
function
is
monitoring
that
kiwi
to
debug
in
and
just
stop
the
debugger
right
on
that
line.
A
So
you
can
see
that
exactly
what
I
typed
is
on
on
that
variable,
so
I
could
debug
see
if
something
is
wrong.
Continue
my
development
here
and
I'll
let
this
function
runs
once
it's
run.
If
I
go
back
to
the
key
you
that
item
got
processed
and
the
function
is
done
once
I'm
ready
with
this
code
after
I'm
done,
developing
I
can
go
to
the
Asscher
extension
and
deploy
this
to
the
cloud
and
again
you
can.
You
know
get
to
that
later.
A
But
how
does
scaling
work?
I
showed
a
single
function
that
that
works
out
of
a
queue
and
the
function
has
a
JavaScript
a
JSON
file,
but
the
function
typically-
and
you
see
the
functions
like
Oh
a
little
blurry.
It's
not
really
running
until
events
start
happening.
So
in
that
part
of
the
first
architecture,
I
show
that
you
have
the
shared
services
on
Azure.
You
have
one
component
called
the
scale
controller.
That
component
is
the
one
that
knows
when
to
wake
up
your
function,
I
went
to
go
from
0
to
1
the
scale
controller.
A
As
soon
as
you
deploy
a
function
is
going
to
start
looking
to
the
event
source
and
once
it
notes
is,
there
are
items
to
be
processed
in
this
case
in
the
queue
it
will
say,
ok
I
need
to
add
my
function
and
instantiate
that,
in
an
actual
machine,
the
function
itself
will
know
how
to
pull
those
events,
the
queue
and
start
processing
them.
So
the
scope
controller
is
really
the
brain
of
the
server
scaling
that
we
have
an
inner
quality
infrastructure
once
the
event,
the
scale
controller
sees
that
tons
of
events
are
happening.
A
So
that's
what
we
built
the
event-driven
scale
and
I
want
to
show
again
what
customers
started
doing
with
that.
So
probably
all
of
you
know
Fujifilm
as
a
brain.
What
I
didn't
know
when
I
came
came
to
talk
to
them?
Is
they
developed
this
product
called
image
works
that
they
sell
to
corporations
that
want
to
manage
all
their
digital
photography
assets
and
they
provide
functionality
such
as
browse
and
search
and
post
post
to
post
to
online
channels
as
well,
so
they
sold
that
to
the
Japanese
baseball
league.
A
Baseball
is
huge
in
Japan
and
what
they
want
to
do
is
as
soon
as
pictures
are
being
taken
by
multiple
photographers
during
a
baseball
game.
They
will
be
made
as
quickly
as
possible,
available
to
be
tagged
and
posted
to
social
media.
The
process
initially
was
literally
the
photographer's
have
their
Nikon
cameras
with
an
SD
card.
They
go
to
a
machine.
A
computer
put
your
SD
card
picture
by
picture.
Add
the
metadata
play
your
name
Jersey
position
that
they
play,
which
any
of
the
game
that
was
happening.
A
This
would
take
hours
for
the
picture
to
be
available
for
social
media
tagging
and
in
this
day
and
age,
that's
not
acceptable.
You
want
to
do
your
tweet
right
away
right,
so
they
brought
the
architecture
and
they
wrote
straight
into
server
less
they
they
get.
The
pictures
get
important
to
a
system.
As
soon
as
the
pictures
get
added
a
batch
of
pictures
from
a
photographer
gets
added.
They
start
the
service
workload
in
functions.
We
have
this
concept
called
durable
functions
that
allow
you
to
orchestrate
a
set
of
functions,
independent
functions.
A
It
will
keep
state,
so
it
knows
where
the
executions
are
and
knows
when
to
wake
up,
to
continue
executing
or
when
to
resume
some
execution
within
that
durable
function.
The
first
thing
it
does
is
pre
process
the
image
so
row
pick
the
image
resize
make
sure
it's
ready
to
be
processed
the
model
and,
in
parallel,
runs
a
facial
recognition
model
to
know
which
player
that
picture
was
from
and
who's
the
player
in
the
picture,
and
then
all
the
other
properties
and
custom
machine
learning
models
to
know
jersey
number
position
that
they
play
any
etc.
A
They
would
use
as
input
at
the
very
bottom.
Some
data
that's
available
to
everyone
about
the
professional
baseball
records
and
durable
functions.
This
great
concept.
That
knows
you
can
literally
add
to
your
code,
a
clause
that
says
when
all
which
means
when
all
these
functions
are
done,
because
if
you've
done
MapReduce,
the
map,
part
of
MapReduce
is
easy.
The
reduced
parts
much
harder,
but
you
can
do
when
oh
and
the
durable
function,
know
that
all
of
them
are
done
it
to
put
all
these
results:
good
together
extra
pallate
and
to
an
index
database.
A
So
these
pictures
could
be
searched
and
be
available
for
for
any
of
the
Japanese
baseball
league
customers
to
tag
and
post
online.
So
they
did
this
and
they
reduced
the
time.
In
less
than
two
minutes,
they
had
pictures
from
the
point
that
it
was
taken
to
be
available
which
was
taken
hours
before,
but
what
Fujifilm
really
liked
was
the
developer
productivity.
They
built
this
whole
solution
in
months
instead
of
having
to
provision
infrastructure
and
take
much
longer
for
them
to
it
actually
less
than
two
months.
That's
what
they
took.
A
There's
a
whole
case
study
on
that
online.
So
that's
what
customers
did
then,
but
we
kept
evolving
as
a
platform
and
I
attained
a
lot
of
it's
only.
My
second
Cube
Khan
I
attend
a
lot
of
service
conferences
in
either
pass
containers.
The
coronaries
were
the
the
art
nemesis
of
service,
like
they
didn't
talk,
containers
at
all
in
those
conferences,
but
hey,
we
became
we
start
becoming
best
friends
over
time.
A
lot
of
it
was
this
ability
of
you
taking
the
same
container
and
be
able
to
deploy
a
new
air.
A
The
amount
of
targets
you
can
put
now
your
functions
was
was
much
higher.
You
could
also
bring
your
own
dependencies.
We
see
a
lot
of,
especially
once
we
release
Python
functions,
a
lot
of
machine
learning
dependencies
that
are
pretty
unique
and
you're
going
to
package
that
with
your
application,
so
so
short
tons
of
use
of
containers.
Obviously
cobranet
is
going
to
be
a
big
use
of
it
and
all
we
did.
That
was
a
little
different
was
we
didn't
want?
Although
people
wanted
containers
a
lot
of
developers,
never
added
to
the
docker
fire
in
their
lives?
A
So
so
we
is
in
our
tooling,
which
is
this
func
comment
that
we
see
it's
part
of
our
command-line
tooling.
We
did
a
command
that
function.
It
doctor
only
it
provisioned
the
docker
file
for
you
that
has
the
base
image
that
contains
the
function.
Runtime,
the
one
that's
portable,
that
one
right
demo
to
you
here
and
then
you
can
add
your
additional
dependencies
there.
But
even
if
you
don't
even
understand
what
this
is,
this
would
just
work
for
it
to
deploy
it
or
container.
All
the
rest
would
work
the
same
as
you
would
expect.
A
I
need
my
code
to
run
right
next
to
my
data,
and
my
date
is
on
pram
I
can't
just
move
that
baby
to
the
cloud
no
way
compute
needs
to
be
close
to
it.
Have
really
low,
latency
or
I
want
to
audit
every
step
of
the
way
or
I
want
custom.
Os
modules,
I,
don't
want
your
version
of
a
Windows
Server
I
want
the
version
of
your
server.
That
I
have
that.
Has
my
modules.
A
Multi-Cloud
became
much
more
instead
of
buzzword,
much
more
of
a
thing
when
customer
I
don't
have
an
example.
Here
talk
about
Smurfs
the
shipping
container
company;
they
do
deploy
both
asier
in
and
Google
at
the
same
time.
So
so
this
is
real
need
from
rio,
enterprise,
customer
and
then
hardware
options
also
like
some
customers
wanted
to
run
all
of
that
in
FPGA
hardware
or
GPU,
and
it
takes
a
little.
A
It
takes
some
time
for
those
options
to
be
available
in
our
service,
but
you
could
deploy
those
on
Prem,
so
there's
a
huge
need
for
kubernetes,
so
that
was
the
good
news.
Yeah
kubernetes
community
excited,
but
then
we
start
interviewing
application
developers
and
they
start
saying
this
is
this
is
way
too
complicated
for
us?
We
don't
want
to
learn
all
these
new
concepts
to
deploy
my
piece
of
Python
or
JavaScript
code
or
Java
code.
A
I
want
just
the
productivity
gains
that
you
showed
me
before,
but
on
the
kubernetes
infrastructure,
and
these
are
actual
quotes
for
some
of
the
some
of
the
folks.
We
talked
to
you
and
the
issues
that
they
had
with
kubernetes,
so
we
went
into
this
journey
of
let's
bring
some
of
what
we
learned
through
all
these
years
to
the
kubernetes
ecosystem.
So
so
now
this
is
the
same
slide.
These
functions
run
in
a
container,
but
in
imagine
that
now
you
have
your
kubernetes
cluster.
A
You
have
tons
of
machines
in
there
they
are
all
obviously
top
docker
compatible,
and
all
the
developer
wants
to
do
is
write
that
really
simple
javascript
function
that
reads
from
a
queue,
so
so
you're
still
what
you
would
you
do?
Is
you
do
the
funky
neat
docker?
Only
that
you
create
your
docker
file
and
here's
the
sort
of
the
magical
element
here.
A
Is
you
do
func
kubernetes
install
kada
and
caleb
replaced
this
scale
controller
that
component
that
I
showed
before
we
had
in
our
proprietary
infrastructure
in
Azure
we
developed
at
open
source
in
partnership
with
Red
Hat
and
others
to
provide
that
same
benefit
in
kubernetes
ecosystem
actually
native
to
the
concepts
that
are
there.
So
we
stole
this
year,
these
it
communicates
with
HPA
and
all
that,
and
actually
this
component
is
much
simpler
than
the
one
that
we
have
in
the
cloud.
A
The
the
types
of
event
sources
also
that
we
got
much
richer
because
it
whatever
the
folks
using
kubernetes,
wanted
to
have
I,
know
the
fonts.
Pretty
small,
but
you
can
see,
there's
12
or
13
of
them,
and
those
scalars
keep
getting
added.
Once
you
do
that,
you
can
run
another
command
that
will
take
that
function
and
deploy
that
to
kubernetes
again
application
developer
might
not
even
understand
kubernetes,
know
the
difference
between
a
pod
or
container.
It
doesn't
matter.
This
will
deploy
it
for
them.
Now.
A
A
There's
a
lot
more
details
on
kada,
actually
Jeff
wizard
has
a
session
later
in
the
week
on
Thursday
to
talk
more
about
it,
but
that
allowed
us
for
us
to
start
going
on
the
kubernetes
journey
and
I
love
application
developers
to
take
the
most
and
make
the
most
out
of
the
platform
and
just
to
show
how
real
this
is.
This
was
presented.
A
So
they
love
this
model
of
having
Asha
functions
that
they
can
put
in
the
cloud.
Some
of
the
workload
gets
process
on
pram
sends
event
to
the
cloud
for
additional
processing,
but
for
the
developers
is
just
a
spring
cloud
function,
its
event
base
architecture,
but
this
is
a
real
like
on
pram
and
cloud
application.
A
Now,
here's
where
we're
I
think
we
should
go
next.
I
talked
up
a
lot
about
the
technologies
that
are
already
in
place.
I
think,
once
we
combine
all
these
things,
I
think
I
showed
a
lot
of
serviceable
bullets
and
I'd
even
show
them.
How?
If
we
bring
some
more
of
that
functionality
to
kubernetes
is
a
win
for
everyone.
A
Now
in
again,
my
job
is
talking
to
customers
most
of
the
time,
and
what
we
see
is
service
is
still
not
being
using
highly
regulated
industries.
You
go
to
financial
industries.
You
go
to
health
care,
they
don't
use
it
because
they
don't
have
all
the
controls
that
they
need
so
doing.
The
first
two
bullets
there
I
think
is
when
servers
can
be
more
mainstream
instead
of
be
just
glue
or
accessibility.
A
So
it's
a
huge
opportunity
for
all
of
us
at
work
in
a
servlet
space
and
and
as
we
learn
and
kay
that
was
huge
first
in
terms
of
learning
and
the
success
we
we
achieved
there
that
we
can
only
get
that
through
open
source
to
partner
with
folks
that
have
been
in
that
space
longer
than
even
we
have
here's,
where
we
are
some
of
the
projects
that
my
team
has
been
involved
directly
are
there
there
are
open
source.
There
are
some
companies
contributing
with
us
and
we
want
to
do
much
more.
A
So
this
slides
essentially
a
call
for
contribution.
We
have
you
know
over
a
hundred
contributors,
but
I
think
we're
welcome
many.
Many
more
contributors
in
there
will
be
here
throughout
the
week.
I
put
some
links
there
if
you're
on
and
learn
more
about
the
technology.
There's
a
micro,
Microsoft
booth,
where,
where
the
team
is
gonna,
be
hanging
out
Jeff's
session
to
go
more
into
cater
and
learn
more
about
that
and
some
announcements
is
gonna,
make
and
and
I'm
available
for
questions
on
social
media
or
any
other
way.
That's
all
I
have.
Thank
you.
B
A
A
So
Debra
and
I
was
considering
adding
it,
but
because
was
Neils
like
I
need
to
get
my
ducks
in
order,
but
dapper
extends
the
programming
model
so
in
functions.
You
can
do
a
bunch
of
things
with
the
program
a
lot
of
functions,
some
of
it
I
showed
here,
but
there
are
some
things
that
once
you
plug
in
the
a
pair,
you
can
do
more,
like
pub/sub
being
a
great
example
thing
that
you
can
do
with
just
functions
natively.
A
A
See
more
an
extension
of
the
code
in
the
application
that
you're
running
not
on
the
service
part,
so
you
can
get
an
application
that
you
have
by
implementing
a
simple
protocol
and
running
that
in
a
sidecar
container.
Next
to
you
now,
your
application
is
much
more
powerful
without
you
having
to
change
your
code
much
just
by
implementing
the
protocol
thanks.