►
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
Intro + Deep Dive: Serverless WG - Scott Nichols, Google & Doug Davis, IBM
In this session the CNCF Serverless Working Group will provide an update to the community on the state of Serverless since we produced our whitepaper and landscape document over a year ago. We'll overview what's changed in the community and what the new, and exciting, developments are - such as new open source projects and trends. Then we'll turn this into a BoF session where we'd like to hear from the community about where they would like the CNCF, and our WG, to go: e.g. what are the pain points people are seeing with respect to interoperability and portability of their Serverless workloads.
https://sched.co/MPlG
A
Welcome
everybody,
I'm
Doug
Davis,
co-chair
of
the
service
working
group,
Scott
and
Jude
to
be
doing
most
of
talking
at
least
initially.
This
is
a
little
bit
of
a
different
session.
We
are
gonna,
give
you
an
overview
of
the
service
working
group
itself.
So
don't
worry
about
that.
A
little
bit
talk
about
service,
its
uses
and
stuff
like
that,
but
we'll
actually
want
a
bigger
portion
of
the
section
or
this
session
to
be
an
interaction
with
you
guys
right
to
find
out
what
you
guys.
A
We
want
from
the
service
working
group,
because
our
job
here
is
really
to
do
what
you
guys
want
in
terms
of
make
your
life
easier
relative
to
using
service
right.
That's
why
the
cloud
of
inspect
that
Scott
will
talk
about.
Was
it
came
up?
What,
because
we
identified
a
pain
point
they
want
to
try
to
ease
for
you
guys.
So
we
want
to
hear
about
your
adoption
experiences
where
you
see
pain,
points
why
you
are
or
are
not
using
service
stuff
like
that.
A
So
that's
why
you
wanted
to
be
a
little
bit
of
interaction
thing,
okay,
so
with
that
here's
we're
gonna
do
obviously
welcome
right
now,
after
that,
Scott
will
talk
about
service
working
group
latest
stuff,
going
on
there
a
little
bit
of
cloud
events,
and
then
Jude
will
talk
about.
You
know
all
things
service,
white
wintry,
where
it's
appropriate
stuff
like
that,
then
we'll
go
to
the
birds
of
a
feather
type
of
session,
so
be
thinking
as
you
listen
to
these
two
guys
about
some
questions.
A
B
B
B
B
So
we
defined
a
common
envelope
for
the
the
cloud
event.
So
it's
it's,
this
common
metadata,
that's
meant
for
helping
routing
device
understand
where
to
deliver
it
to
and
we
we
didn't
really
want
to
worry
about
the
exact
semantics
of
the
payload,
because
that's
not
actually
very
important
to
the
the
middleware
pieces.
What
is
important
is
how
that
event
actually
gets
to
the
destination.
B
Oh
sorry,
last
year,
yeah
you're
right,
but
did
we
run
to
the
next
stage
or
it's
okay,
we're
we're
on
the
edge
of
incubation,
it's
like
not
ready
to
incubate,
but
almost
we
have
version
three
coming
out
very
soon
and
then
maybe
a
100
in
a
few
months,
several
several
months,
so
a
an
example
of
a
cloud
event.
This
one
happens
to
be
HTTP.
B
Normally,
your
application
would
do
something
like
this,
where
it's
a
you
do
a
post
and
you
have
some
data
in
the
body,
but
if
you
make
this
request
and
you
send
it
through
middleware,
how
is
that
middleware
supposed
to
understand
the
meaning
of
what
your
payload
is
without
some
common
understanding?
So
what
cloud
events
does
is
says:
here's
how
to
understand
this
and
I.
B
Don't
know
exactly
what
you're
going
to
do
with
it,
but
here's
some
data
that
I
think
is
pretty
standard,
that
you
can
make
choices
on
how
it's
going
to
be
routed
and
then
this
the
same
event
can
be
represented
in
both.
So
what
we
call
HTTP,
binary
mode
or
HTTP
structured
mode,
an
instruction
mode
is
the
whole
envelope
is
JSON
encoded
in
the
post
body.
B
So
we
have
a
specification.
We
also
have
serialization
in
JSON
and
Kewpie
binary
protobuf,
and
then
we
also
have
a
way
that
helps
you
understand
how
to
map
the
the
actual
envelope,
the
the
hole,
all
the
attributes
and
your
opaque
data
on
two
different
transports.
So
now
you
can
send
your
event
with
the
correctly
configured
transport.
You
can
send
an
event
that
will
be
understood
on
the
other
side
as
a
cloud
event,
so
you
can
write
your
application
to
be
fairly
transport
agnostic.
B
B
B
So,
what's
next
for
cloud
events
or
the
service
working
group
is
weird
we're
not
really
sure
we
that's.
So
that's
why
we're
here
today
is
there's
some
proposals
and
we
need
help
to
understand
what
should
be
a
priority
for
the
next
work
or
maybe
feedback
on.
What's
not
working
with
cloud
events,
if
you're
not
using
it.
B
C
B
C
It's
functions
and
what,
if
it
applies
to
solace,
functions
and
not
service
right,
so
I'm,
a
big
proponent
of
the
entire
service
experience,
but
when
it
comes
to
service
functions,
deploying
functions
in
the
cloud.
There
are
a
lot
of
different
things,
different
things
to
consider
right,
so
we
start
with.
When
do
you
use
them
and
we
can
draw
a
quick
comparison
between
object
or
it
programming
right?
How
many
of
you
do
Java
or
Python,
pretty
much
like
I
think
80%
right.
C
So
these
are
objective
object,
oriented
programming
languages
and
you
have
high
cohesion
between
a
class
and
an
interface
and
different
classes
are
loosely
coupled
right.
You
don't
use
them
directly,
so
what
I
mean
by
high
cohesion?
So
a
function
should
do
exactly
one
unit
of
work
and
sorry
but
yeah,
okay,
and
just
that
one
piece
of
work
it
shouldn't
do
too
many
things
right
functions
should
be
loosely
coupled.
That
means
each
function
shouldn't
have
like
a
dependency
on
another
function,
which
has
an
dependency
on
another
function.
C
We
will
see
why,
shortly
when
you
write
them,
this
should
be
easily
testable
right.
So
when
we
wrote
the
cloud,
the
the
cube
count
aloud,
events
demo,
Doug
Scott
and
floss
and
oven
right.
What
really
hampered
us
to
like
you
know,
get
it
up
and
running.
Really
quick
was
debug
ability,
because
all
of
us
were
all
isolated
as
functions
in
different
clouds.
We
can
I
found
it
hard
to
debug
right.
C
So
that
is
why
you
should
just
write
like
you
know,
one
that's
why
they
should
be
highly
cohesive
and
easy
to
test
now
coming
to
portability.
Today
you
have
various
providers
like
AWS,
you
have
lambda,
you
have
a
sure
you
have
cheek
Google
cloud
functions
and
so
on,
but
if
you
notice
something
when
you
deploy
to
these
guys,
okay,
so
let's
take
an
example:
I
want
to
write
a
function
that
just
does
Fibonacci
numbers
right.
C
It's
a
simple
function,
recursive
now
the
way
I
deploy
it
on
all
these
clouds
changes
from
cloud
provider
to
cloud
provider.
I'd
leave
my
functions
to
just
look
like
define
function,
a
in
one
of
our
Python
write,
a
def
function
name
and
do
something
right.
If
there
were,
if
there
was
a
specification
around
this
I
could
take
this
stream
function
and
deploy
it
on
every
single
cloud
function.
C
Provider,
but
as
of
today,
I
cannot,
for
example,
on
lambda
I
need
to
respect
import
the
lambda
the
packages
right
same
with
the
other
four
same
with
the
other
cloud
providers.
So
we
do
not
have
anything.
That's
that's
like
a
generalized
specification,
so
it's
not
easily
portable.
As
of
today
at
least
and
then
you
come
come
come
to
a
couple
of
examples
when
these
things
are
useful
right.
So
so,
let's
functions
are
really
useful.
Have
you
guys
ever
used
slack
yeah
all
right?
How
many
views
you
have
used
the
giffy
slack
integration?
C
Okay,
that's
it
just
3!
Oh
you
guys
are
missing
out
in
life.
Then,
okay!
So
let
me
tell
you
a
bit
of
what
gif
he
does
then
so
say:
you're
debugging,
a
bunch
of
code
and,
like
you
know,
stuff,
is
not
working
and
you
want
to
vent
out
your
frustration
in
a
graphical
form.
So
you'll
say,
like
you,
know:
/qv,
slash
home
or
strangling
bot
right.
C
If
you
guys
know
like
Simpsons
and
and
then
gif,
you
will
kind
of
send
you
like
a
slack
message,
saying:
okay,
this
is
the
video
or
or
like
a
small
chip
of
you
know
how
much
strangling
part
and
do
you
want
to
post
it?
Yes
or
no
right,
and
then
you
hit
yes
and
then
it
goes
on
to
the
slack
channel
and
everybody
has
a
good
laugh
and
whatever.
So
this
is
a
classic
example
of
automating
a
workflow
right.
You
have
a
particular
workflow.
C
You
want
to
publish
a
Jif
to
the
channel
and
this
entire
work
workflow
is
perfectly
suited
for
service
functions
right.
Another
example
is
MapReduce
how
many
familiar
with
MapReduce,
okay,
so
yeah
MapReduce,
because
it
just
spawns
multiple
functions.
The
instances
are
created,
destroyed,
just
works
right,
and
then
you
have
parallel
workloads.
So
a
really
nice
popular
use
case
for
this
is
thumbnail
generation.
You
run
like
you,
know,
grafter,
you
guys
know
what
crafter
automatics
the
you
associate
an
image
with
an
email
address
right
so
to
generate
thumbnails
for
this
kind
of
a
service.
C
You
can
use
so,
let's
functions,
it
works
really
well
and
finally,
we
have
a
real
life
example
other
than
giphy,
which
I
don't
know
how
it's
deployed.
We
haven't
a
real
life
example
of
the
check
to
do
app
right.
So
this
app.
Let
me
tell
you
a
bit
water.
Does
it's
basically
a
github
app
where
every
time
you
create
a
PR
on
github,
all
it
does?
Is
it
checks
if
you've
added
any
lines?
C
That's
that
have
the
word
to
do
in
them,
okay,
so
hence
so
the
app
is
called
check
to
do,
and
if
you
do
all
that
it
does.
Is
it
just
like
you
know,
list
down
and
it
creates
like
a
negative
github
check.
So
it's
like
okay,
hey,
you
guys,
have
added
a
couple
of
two
routes:
what
to
do
with
them
right.
So
what
do
we
see
from
this
app?
What,
rather,
what
do
we
learn
from
it?
C
C
Why?
Why
do
I
say
service
functions
you?
Why
do
I
say
that
you
need
to
be
careful
of
any
using
service
functions?
If
you
write
your
entire
back-end
in
service
functions
like
you
know,
you
have
functions
to
take
data
from
like
a
website,
push
it
to
a
database
and
then
say
later
on
trigger
an
email
from
an
email
provider
and
then
run
some
for
the
computing
processing
on
all
of
that,
there
is
no
clear
visibility
in
the
flow
of
this
data.
You.
What
do
you
have?
C
You
have
a
bunch
of
functions
deployed
on
a
cloud
that
basically
do
not
tell
you
how
your
data
is
moving.
So
that's
the
first
kind
of
the
first
consideration
that
you
need
to
make
now
with
all
of
these,
with,
like
fast,
like
you
know,
function
that
the
service
being
a
very
hot
topic
these
days
and
it's
a
really
cool.
C
Like
you
know,
wave
you
got
to
see
how
these
these
functions
are
being
run,
how
they
being
executed
in
the
background
by
these
major
providers
so,
for
example,
some
of
them
all
they
do
is
they
just
run?
They
run
a
binary
like
a
command
like
like
a
go
binary
or
whatever
and
they
executed,
and
they
actually
use
standard
input
and
standard
output
to
communicate
with
the
function.
Now
this
is
a
it's
not
highly
scalable.
Well,
it
is,
but
you
can
throw
a
lot
of
money
on
it
and
B.
C
These
are
just
pure
C's.
Now,
why
do
I
say?
Does
anybody
in
the
end
the
audience
have
an
idea
why
I
say
this
is
just
a
POC
and
not
scalable?
Oh
yeah,
let's
give
you
an
example:
let's
take
Python
interpreted
language,
okay
and
let's
say
that
your
cloud
provider
is
running,
sed,
I/o,
right,
standard
input
and
output.
Now
what
will
happen
is
for
every
single
call
to
this
function.
C
The
Python
interpreter
is
going
to
be
started
destroyed
and
if
you
have
n,
let's
say
a
thousand
concurrent
requests
and
let's
assume
that
the
Python
interpreter
takes
about
200,
to
add
two
megabytes
of
standard
memory
right,
so
you're
spending
about
two
gigabytes
which
is
and
mm
so
you're
spending
about
two
gigabytes
in
an
interpreter
overhead
versus.
If
you
just
ran
like
a
micro
service
or
a
stateful
service,
it
will
be
just
the
standard
two
megabytes
for
overhead
and
everything
else
would
just
be
memory
that
your
app
actually
uses.
C
So
that's
that's
what
I
mean
men
consider
how
these
runtimes
run
your
functions
and
then
comes
resource
management
right.
So
let's
take
database
connections
Postgres,
my
sequel
DB.
All
of
these
are
really
amazing
databases,
except
for
the
fact
that
if
you
spawn
100,000
connections
to
them,
they
will
well,
unless
you
would
like
you
know,
put
them
on
two
terabyte
boxes
and
throw
money
at
it.
Why?
Because
these
are
connect?
Connections
are
inherently
expensive
and
then
comes
like
my
last
point.
The
measurement
of
this
is
hydrated.
C
What
I
just
said,
would
you
rather
do
hundreds
of
function
replicas
or
rather
do
two
months
microservices
running
on,
like
you
know,
a
cloud
run
or
some
other
offering
right
think
about
why
so
these
days
are
all
of
therfore.
The
cloud
provider
functions
that
you
see
there.
Demos
are
like
this.
Okay,
they
see
like
okay,
stand
out.
C
Five
requests
per
second
I
am
able
to
it
like
the
run
times,
pawns
about
five
instances
of
a
function,
and
then
you
up
it
up
to
like
say
ten
or
twenty
requests
per
second,
and
then
you
see
Oh
God
twenty
instances
of
my
function
and
at
that
point,
you're,
like
okay,
what
happens
when
I
run
when
I
increase
my
concurrency
to
a
thousand
requests
per
second?
Do
you
I
will
I
get
like
a
hundred
function
in
instances
or
will
I
get
a
thousand
now?
C
This
is
insane
because
if
you,
if
your
function,
made
use
of
a
tree
base
connection-
and
you
were
creating
and
tearing
down
those
database
connections
so
frequently
you'd
be
wasting
a
lot
of
resources.
It's
at
this
point.
We
should
think
ok,
I
will
not
do
function
and
instead
I
will
do
two
instances
of
just
a
micro
service
right
because
just
works
too
long-lived
TCP
connections
to
each
rebus
and
you're
good
to
go
right,
say
yeah.
So
this
is
my
conservation
point
and
now
I'll
get
dug
for
the
BOF.
A
A
A
D
E
A
Okay
cool
now,
so
let
me
ask
you
before
everybody
else,
who's
actually
not
using
it.
Yet
is
there
anybody
willing
to
share
why
they're
not
using
it?
Is
it
simply?
You
just
have
enough
time
to
play
with
it.
It's
still
too
new
I
like
to
understand,
what's
been
the
any
hesitation
in
terms
of
why
you
guys
aren't
using
it,
even
if
it's
just
to
play
around
with
you
can
still
say:
okay
yeah
after
him,
let's
go
back
to
the
gentleman
back.
There
main.
D
It
appears
to
be
also
very
ugly
because
they're,
all
kind
of
sort
of,
like
you
have
a
very
nice
Python
code,
for
example,
or,
like
anything
else
and
then
inside
of
this
sort
of
Python
code,
you'd,
have
like
an
ugly
call
to
a
lambda
which
doesn't
look
like
anything
that
anything
will
view
of
your
code.
It's
just
a
reference
to
something
else,
just
ugly.
Instead,
you
could
have
used
a
regular.
You
know
kind
of
Java
like
object
with
reference
to
name
of
a
function,
sort
of
again.
A
D
A
D
F
Maybe
this
function,
or
maybe
this
other
function
and
I
just
I
just
ended
up
wanting
something
more
like
like
Apache
storm
or
heron,
or
beam,
or
something
like
that
where,
where
you
write
this,
you
know
complex
or
too
complex
a
cyclic
graph
with
the
different.
You
know
parts
and
pieces
being
independently
scalable
like
automatically,
but
the
issue
there,
of
course
is
that
is
that
that's
all
very
tightly
coupled
because
it's
a
complete
graph,
so
maybe
I
so
so.
For
that
reason,
I
ended
up.
F
A
Because
I,
that's
actually
one
things:
I've
I,
think
a
lot
of
people
probably
have
problems
with.
We
talked
about
going
from
monolith
to
microservices
and
we're
gonna
go
in
one
step
further
to
functions
and
people
were
I
think
are
struggling
just
going
to
micro
services.
Now
we're
gonna
ask
you
to
break
it
up
even
more.
A
B
B
A
B
Oh
yeah,
so
I
work
on
Canada
of
eventing,
like
I,
said,
and
we
we
solved
both
those
same
problems
using
cloud
events
with
the
known
metadata
and
then
using
the
that
our
our
new
model
is
the
broker
trigger,
which
is
basically
an
eventing
mesh
and
then
a
query
out
of
that
a
mesh.
So
we
would
make
these,
like
so
I've,
been
making
many
examples
of
these
a
cyclical
graphs,
using
kind
of
hacking,
the
event
type
to
understand
what
the
next
stage
is.
B
So
it
says,
like
I'm
I'm
done
with
stage
2
or
whatever
that
means,
and
then
it
goes
back
into
the
broker.
And
then
another
query
comes
out
and
maybe
selects
one
or
two
other
functions
like
go
and
do
some
work
and
then
push
that
event
back
into
the
broker.
And
you
get
this
interesting
thing
where
sorry
I
have
a
demo
that
I
could
do
but
or
we
could
do
later-
that
does
reverse
polish
notation
calculation,
and
so
there's
it
like
gets
a
GCS
event.
It
splits
the
file
it
pushes.
B
It
makes
a
new
cloud
event
pushes
it
to
the
broker.
A
trigger
is
listening
for
the
expression
event.
It
does
one
operation
and
pushes
it
back
until
it
gets
to
a
result,
and
you
kind
of
get
this
interesting
thing
where
it's
not
coupled
and
it's
testable
as
a
unit,
but
it's
also
serverless,
and
so
it
scales
to
as
many
files
as
you
pushing
it,
and
so
it's
possible
and
because
we're
using
cloud
events
where
those
functions
are
transport
agnostic.
So
they
can
be
invoked
by
many
different
queueing
technologies.
G
Hello,
so
we
were
always
like
concerned
about
as
a
fact
that
you
have
this
cold
startup
time
the
drastically
different
from
the
hot
execution
time,
and
especially
if
you
are
using
like
AWS
and
you
have
lambda
within
your
V
PC.
It
may
take
like
up
to
15
seconds
and
you
cannot
really
control
it.
You
cannot
say
please
always
keep
like
few
more
landis
ready
to
to
like
to
hold
the
Lord.
G
For
example
like
if,
if
regular
request
executed
for
like
I
know
like
500
milliseconds,
for
example,
I
would
say
more
than
second
in
health
would
be
an
acceptable
because
if
you
have
like
function
beyond
function
beyond
function,
it
all
like
Cascades
and
in
result
customer
me
if
it.
If
he
is
unlucky
enough
me
wait
for
like
10
15
seconds
and
a
man
shall
leave
the
site.
Yep.
A
C
C
Be
on
okay,
so
what
about
long-term
support
like
we
know
VMS
are
here
to
stay.
We
know.
Kubernetes
clusters
are
like
you
know
your
your
estate.
What
about
long-term
support
like,
for
example,
Google,
just
launched
cloud
run
right?
Will
this
support
it
for
the
next
five
years,
because
we
know
for
a
fact
that
virtual
instances
are
here
to
stay
right?
You
have
dissolution,
you
have
AWS,
you
have
all
of
these.
They
work
and
you
can
run
kubernetes
that
set.
Are
these
clustering
applications
on
top
of
them?
C
A
Me
ask
you
about
that,
though,
because
it
seems
to
me
that
you,
you
can't
guarantee
there.
Anybody,
even
the
big
guys,
are
gonna
keep
their
even
quote:
Google
run
it's
a
big
guy.
Obviously,
but
even
you
know
they
they
may
decide
to
kill
it
off
tomorrow,
for
whatever
reason,
but
it
seems
to
me
what
your
what
the
thing
against
that
would
be
portability
of
your
function
but
you're
on,
but
but
but
you
want
to
be
able
to
port
it
to
someplace
else
in
case
I
do
go
away.
C
C
B
So
in
the
particular
case
of
K
native,
the
this
isn't
a
real
concern,
because
the
the
thing
you're
deploying
to
Club
run
is
a
K
native
serving
manifest
that
has
a
reference
implementation
in
open
source.
So
you
could
go
and
choose
to
host
it
on
your
your
own
kate's
cluster,
that's
on
Prem
or
in
IBM
cloud
or
in
IKS,
or
pick
a
kubernetes
cluster
digitalocean
right
like
it's
all
installable.
B
A
You're,
right
and
but
the
bigger
point
is
what
he's
trying
to
say.
Talk
to
is
Kay
native
is
used
by
many
different
people
open
source.
They
have
no
implementation
of
it,
as
well
as
the
open
source
version,
but
you
want
portability
your
function
so
that
if
someone
does
decide
to
go
down,
you
can
move
it
someplace
else
how
to
assume
that
would
only
adjust
your
concern.
Yeah.
Okay,
it's.
H
Portability
of
functions
really
that
important,
like
okay
I,
can
move
my
lambda
code
because
we
live
in
a
perfect
world
where
they
all
have
the
same
signature
but
mild
serverless,
app,
my
D
is
in
dynamo,
might
be
using
AWS
I
am
might
be
using.
Okay,
Aurora
and
databases
are
generally
easy
to
move,
but
as
soon
as
you
go
into
the
whole
service
space,
you
end
up
using
a
lot
of
very
cloud
specific
features.
You're
in
AWS
you
will
likely
use
dynamo,
you
will
use
I
you're
in
Google
Cloud.
H
You
will
use
their
AI
services
because
they're,
awesome
and
so
on
and
so
forth.
Your
in
Azure
you're
gonna
use
their
media
services,
and
is
there
really
that
much
value
in
being
able
to
move
your
function?
If
you
it's
like
using
like
the
increment
is
across
cloud,
it
doesn't
make
sense
because
you're
limiting
yourself
so
much
to
what's
common
between
all
the
clouds.
So
does
that
make
sense
and
I
guess
this
is
a
question
for
everybody
turn.
H
Like
there's
also
a
really
cool
series
of
blog
posts
about
hexagonal
lambdas,
where
every
dependencies
hidden
behind
an
implementation
and
say
if
you
have
dynamo
you
you,
you
can
use
dynamo
local,
you
can
do
stuff
in
memory
to
mock
and
develop
locally,
but
in
my
opinion,
that
comes
with
a
lot
a
lot.
A
lot
of
mocking
like
I
spent
half
an
hour.
Writing
my
code
and
three
hours.
Writing
mocking
code
to
be
able
to
test
my
function
locally.
H
A
Let
me
ask
a
question
along
those
lines,
one
of
the
things
I
hear
about.
We
talk
about.
You
know
what
the
service
working
groups
to
do
next
function.
Portobello
these,
it
seems
to
pop
up
a
lot
function,
signatures
you
want
some
commonality,
but
then
I
also
here's
one
of
the
people
that
Porter
your
function.
Other
platforms,
yes,
there's
a
cost
because
the
signature
may
change,
but
it's
not
that
big
a
deal.
It
may
be
a
little
bit
of
effort.
A
You
know
30
minutes
or
so
to
change
it
over
because
most
of
them
do
the
exact
same
or
have
it
similar
concepts
involved.
So
it's
the
portability
of
function,
isn't
as
big
as
some
people
think
it
is
that
kind
of
what
you're
asking.
So
what
are
you
guys
are
actually
using
servlets
today.
Is
portability
really
that
big
concern,
or
is
one
of
those
things
that
super
officially
it
sounds
like
it
should
be,
but
it's
not
really
that
big
a
deal
because
you're
not
changing
platforms,
every
five
minutes.
What
doesn't
matter?
What?
D
A
Good
to
know
because,
because
as
Scott
was
showing
on
previous
slide,
you
know
we
had
that
list
of
things
we're
going
to
talk
about
in
terms
of
what
to
work
on
next
and
phones
of
signatures
was
one
of
the
ones
that
that
came
up
there
on
the
vote
and
they
didn't
win
and
are
in
our
votes.
Actually
work.
Workflow
popped
up
was
the
first
thing
they
wanted
to
work
on
next
and-
and
so
we
started
working
on
that
a
little,
but
a
lot
of
people
keep
saying.
No,
that
was
a
mistake.
A
D
A
J
Maybe
not
the
exact
portability,
but
personally
I
want
to
consume
someone's
house
function.
If
it's
a
really
coffee,
coke,
hiko,
fusion
and
I
feel
always
that
reinventing
myself.
So
there's
a
very
smart
engineers
that
work
for
some
functions
so
why
Cannot
I
just
consumed
either
with
someone
who
worked
at
the
base
using
their
database?
I
am
Laura
and
just
use
that,
for
my
business
logic,.
A
C
Think
about,
oh,
he
assigned
function
right
at
times.
You
know
what
it
does.
It
calculates
distance
between
two
places
or
not
right.
If
there's
a
function
like
this,
why?
Why
would
I
need
to
keep
rewriting
this?
For
my
service,
I
could
just
as
Easter
a
function
store
could
be
really.
Work
could
really
work
today
that
doesn't
exist
because
there
is
no
common
spec,
so
you
can't
deploy
this
function
wherever
you
got
it
from
on
lambda
or
whatever
product
your
company
is
using
today.
A
J
Yeah
first
as
a
developer,
I
want
to
load
it
and
test
it
locally,
of
course,
to
see
how
exactly
behave.
I
want
to
feel
safe
that
even
if
user
someone
else
code
I
want
to
know
how
they'd
behave
so
I
want
to
capture
all
of
the
traffic
day.
It's
going
outside,
so
probably
that's
their
stuff,
but
yeah,
and
then
I
want
to
decide
that
the
code
provider,
if
there's
a
dependency,
I'm
happy
to
use
that
cloud
provider
like
it
over
there.
So,
okay.
H
Sharing
functions
and
code
is
still
a
big
issue.
I
AWS
has
the
AWS
servers
application.
Several
a
certification
repository
some
agua,
where
you
should
start
functions,
but
it's
only
filled
with
hello
world
and
that's
terrible.
Even
at
the
talk,
we're
like
hey
how
we
were
using
servlets
to
implement
our
own
services
like
I
guess
runs
a
Baker's
control
plane.
The
ETS
control
level
runs
fully
civilized.
They
were
speaking
about
how
they're
doing
like
how
they
package
their
functions.
H
To
do
automatic
grid
tries
with
the
laser
queues
and
all
that
and
sharing
that
it's
still
very,
very
difficult.
I
didn't
want
to
parse
github
events
anymore,
I
wrote,
github
events
parsing
hundreds
of
times
I,
don't
want
to
do
that
anymore.
I
just
want
to
be
able
to
click
about
and
we're
getting
towards
that
like
duty
idle.
A
service
report
story
is
helping
this
story.
Script
is
helping
with
the
coding
part,
but
it
seems
like
we're
still
so
far
away
and
we
keep
implementing
every
single
thing
every
single
time
for
every
single
service
right.
Okay,.
A
So
there's
a
different
question,
especially
for
people
who
did
not
raise
their
hand
about
using
service.
Are
you
guys
actually
using
kubernetes
so
who's
in
here
using
kubernetes,
okay,
I'm,
actually
surprised
that
every
bet
everybody's
hand
did
not
go
up.
That's
that's
interesting!
So
for
you
guys
who
are
using
kubernetes
but
not
serverless,
why
haven't
you
made
the
switchover?
Have
you
are
now
I'm,
not
gonna,
pick
on
you
anymore.
So,
gentlemen,
right
here
so
I
want
to
understand.
Is
it
just
because
it's
still
too
new?
A
K
Yet
to
find
a
use
case,
for
it
sorry
interesting
most
of
our
use
cases
we're
actually
working
on
a
platform
for
exactly
this.
We
are
a
single
Service
Catalog,
where
developers
can
actually
upload
at
the
moment
its
docker
services,
but
we've
looked
at
allowing
developers
to
add
their
own
functions
as
well.
We
provide
unified
billing
and
everything
to
go
along
with
that
open
ID,
Connect
authentication,
but
we
internally
have
no
use
for
server
lists.
I
can
see
it
being
used
as
a
almost
a
reactive
stage.
K
For
example,
I
have
a
colleague
in
a
different
startup
who
does
it
for
reading
from
Kafka
events
and
doing
stream
processing
from
website
tracking,
because
it's
sporadic
it
helps
scale
and
they
don't
they've,
never
lost
an
event
through
it,
but
we
internally.
We
want
to
be
reactive,
so
we
don't
want
the
call
startup
time
there
reaches
a
stage
where
the
server
list,
as
was
mentioned
before
server
list
kind
of,
reaches
its
limit
with
database
connections,
and
you
are
better
just
using
a
micro
service
anyway.
A
L
A
community
for
you,
hey
so
I
work
for
juniper
and,
like
you
can
imagine
we
make
hardware
services
like,
but
we
use
kubernetes
extensively
and
where
I
see
the
problem
is
like
how?
How
do
we,
my
biggest
problem
with
service?
Is
it's
a
nice
HelloWorld
application,
but
then
all
the
problems
that
you
see
them
just
beginning
to
solve
in
kubernetes
with
Prometheus
and
key,
and
all
that
all
the
problems
we
already
had
solved
and
micro-services
came
along
and
broke
the
world
apart
now
we
have
to
resolve
them
right.
L
Remember
that
this
is
now
another
case
of
that.
How
are
we
gonna
do
all
those
things,
because
our
customers,
all
gonna,
go
yeah,
that's
very
nice!
How
do
we
know
what
the
hell
happened?
That's
the
question.
They
asked
us
right,
so
you
know
firewalls
and
security,
and
all
that
kind
of
stuff
and
big
banks,
and
things
like
that
service
is
what
are
they
gonna
use
that
for
so
it's
it's
a
it's
a
it's
a
question
that
I'm
asking
I
can
see.
Maybe
in
provisioning,
that's
something!
We
have
a
lot
of
events
like
infrastructures
code.
L
That's
my
baby!
It's
like
trying
to
make
people
use
that
for
their
networks
and
that's
where
service
might
have
like
a
real
niche,
because
we
already
use
things
like
salt
and
those
salt
stack
and
stuff
like
that,
which
is
all
event
based.
So
maybe
that's
the
application
where
we
could
use
server
lists
and
we
could
use
your
service
thing
because
it
wouldn't
be
something
you're
not
gonna.
L
Put
on
the
cloud
necessarily
would
be
something
where
you'd
have
your
local
service
implementation,
which
is
what
you're
talking
about,
which
is
kind
of
why
we
came
to
your
talk
as
there
are
local
service
implementation.
This
is
something
we
can
be
interested
in
right,
so
yeah.
So
from
my
point
of
view,
while
we
don't
really
use
service
is,
is
that
tracing
capability
of
like
what
happened?
Yep,
you
know
what
just
happened.
We
spoke
to
20
different
things
and
there's
no
way
of
had.
How
do
you
know?
A
L
Thank
you,
that's
what
we
found
with
our
aventure
than
stuff
right,
it's
really
hard
to
keep
track
of
what
all
the
things
that
happen
and
that
matters
to
our
customers
right
when
you
say
the
fire
rule,
rules
didn't
get
created
in
the
right
order.
You
end
up
in
the
newspaper
right,
so
that's
bad!
So
that's
why
it's
kind
of
like
I'm,
not
sure
if
it's
useful
for
well,
it
may
be
useful.
We
just
haven't
fixed
that
problem
right.
A
Actually,
his
hand
might
have
been
up
first,
then
we'll
go
over
here
now.
There
are
no
wrong
of
paints
on
this
or
great
that
severe.
The
whole
point
of
this
is
a
gather
input,
whether
they're,
right
or
wrong.
That
way,
we
can
take
it
back
and
think
about
it
and
think
about
what
we
need
to
do
next
cuz,
some
of
the
stuff
I
haven't
heard
anything
that
I
did
have
not
heard
before
so
there's
I'd.
So
far,
everything
sounds
valid
to
me.
There
was
a
potential
problems,
yeah.
M
So,
maybe
to
speak
more
for
our
customers
in
the
past,
so
coming
from
old
systems
like
SMP
systems,
the
customers,
if
they
do
extensions-
that
is
not
a
huge
application.
Programming
right.
It's
usually
some
few
lights
of
courts
to
extend
the
business
process,
for
example,
and
in
the
past
we
saw
many
customers
using
this
kind,
very
easy
and
lightweight
functions,
and
that
was
not
called
functions.
What
cost
differently,
but
that
is
widely
adopted.
Actually
in
these
old
application
systems.
M
A
Interesting
yeah
actually
what's
interesting
is
the
subscription
API
line
is
something
I
added
just
recently,
simply
because
it
has
come
up
several
times
as
we
start
working
on
Mackay
native
project.
In
particular,
we
want
to
help
people
automate
the
process
of
getting
those
or
subscribing
to
the
event
producers
exactly.
M
A
We
wanted
to
know
whether
having
a
common
API
to
do
that,
it's
something
that's
actually
needed,
or
it's
not
that
big
a
deal,
because
it's
not
that
hard
for
you
actually
make
the
connection
yourself
so
you're
saying
a
common
API
subscription
API
actually
may
be
useful
for
the
scenarios.
You're
saying
definitely.
M
A
I
So
just
a
very
quick
but
like
forward-looking
question.
So
there
are
a
lot
of
talk
about
like
cloud
fair
fastly.
They
all
use
like
wisdom,
webassembly
for
their
server
lists,
so
it
sort
of
different
from
what
we
are
talking
about
in
kubernetes
world,
like
a
native,
so
I'm
wondering
what's
your
opinion.
So
do
you
think
there
will
be
a
like
platform
in
service
similar
to
community
stack,
who
support
like
either
micro,
VMs
or
container
based
or
even
web
assembly
based
service
offerings?
I.
A
B
N
Thank
you,
I
wanted
to
get
back
at
something
the
gentleman
behind
me
was
hinting
at.
Do
you
already
see
some
kinds
of
well
best
practices
or
good
practices
emerging
in
this
realm
of
observability
in
service,
because
at
least
in
my
mind,
this
whole
full
based
thing
where
you
have
this
prometheus
instance,
which
pulls
all
your
metrics
from
your
micro
services
works
very
well
there,
and-
and
it's
really
nice,
if
you
can
instrument
all
your
libraries
and
so
on,
and
also
a
infrastructure.
H
Have
many
opinions
on
this
so,
as
the
gentleman
in
the
back
has
said,
observability
for
server
s
it's
harder
than
for
micro
services,
but
that's
not
necessarily
because
it's
a
new
technology
or
because
it's
different
it's
because
you
have
a
lot
a
lot
more
tiny
pieces.
You,
if
you
have
three
micro
service
with
three
replicas.
Okay,
you
can
kind
of
stream
the
logs.
If
they're,
not
too
many,
you
can
still
SSH
into
more
chips,
ETL
exact,
which
is
the
new
SSH
into
each
container
and
see
what's
happening
in
there.
H
If
you
have
a
the
same
app
in
serverless
you're,
not
gonna
have
three
replicas
you're
gonna
have
like
hundreds
or
thousands
of
functioning
vacations
and
what
I
really
like
about
server
less
is
the
fact
that
it
is
forcing
you
to
add
observability.
If
you
don't
follow
best
practices,
you
cannot
do
server
less
and
you
are
going
to
have
a
ton
of
issues
for
micro
services
for
kubernetes.
H
It's
forcing
you
to
instrument
your
function,
because
you
don't
have
the
visibility
into
it
and
that's
more
of
an
issue
of
something
that
I
believe
we
all
have
been
ignoring,
which
is
an
issue
like
we
aren't
focusing
as
much
as
we
should
on
observability
and
I'm,
like
I'm
cooking.
A
lot
from
charity
majors,
who
is
was
the
CEO
funny
combina
is
the
city
of
honeycomb.
They
are
very,
very
focused
on
observability.
They
have
servers
observability
in
their
examples
and
it
helps
a
lot
light.
H
Step
also
is
another
vendor
who
helps
a
lot
with
this,
but
you
cannot
just
have
a
function
and
copy/paste
your
code
from
a
micro
service
where
you
have
like
printf
logs
and
you
go
into
cloud
watch.
You
see
the
pun
of
vlog
streams.
You
know
like,
oh,
my
god,
what
did
I
get
myself
into
service
is
really
forcing
these
best
practices
and
software
development
and
software
delivery,
and
actually
writing
software
and
I.
Don't
really
believe.
That's
a
issue
of
the
issue
of
how
we
need
to
develop
software
for
lands.
It
needs
to
be
focused.
A
D
Deal
first
for
it
frankly,
I'm
totally
siding
with
you
about
practices
because
well
Cyril
kind
of
leaning
towards
functional
style
and
functional
kind
of
pure
functional
style.
So
you
probably
should
sometimes
ask
yourself
if
that
doesn't
work
for
you
that
well
maybe
you're
doing
something
wrong
in
the
first
place
and
I'm
not
saying
that
you're
you're
coding
you
it's
a
bad
carrot,
maybe
just
not
using
serverless
as
it's
meant
to
be
used,
and
it's
just
a
bad
application
say
pure
functional
style.
D
Just
would
imply
that
you
can
test
functions
without
side
effects
and
then
your
observability
by
itself
is
kind.
It's
been
questionable,
because
if
you
can
run
it
here,
then
you
can
run
it
here
and
it's
fully
reproducible
and
it's
you
know
that
basically
it's
about
functional
purity,
so
I
know
it's
bit
more
like
abstract
and
real
world
is
very
different,
but
at
the
same
time
there
is
a
piece
of
like
there's
something
in
this.
D
You
know
like
some
things
you
can
do
in
a
different
way
and
then
all
the
problem
may
just
go
away
and
that
actually
happened
a
few
times
when
we
just
started
thinking
differently.
A
little
bit
about
different
problems
and
suddenly
observability
by
itself,
stop
being
such
a
pain
in
the
butt
interesting.
O
O
I
have
a
question:
I
mean.
Let
me
think
about
a
function
perspective
we
can
think
like.
Is
it
pure
stateless
function
or
stateful
function,
I
mean?
Is
it
idempotent
function
or
not?
So
in
real
world?
We
don't
have
too
many
case
like
just
like
calculate
Fibonacci,
but
we
do.
We
must
need
to
create
some
states
and
we
aggregate
some
data
and
then
purchase
through
somewhere
and
we
change
something
and
it
is
most
not
idempotent.
O
My
question
is
with
this
approach
is:
if
we
already
triggered
an
event
and
if
it's,
if
it
starts
to
work
somewhere
in
the
community
native,
is
it
possible
to
kill
it?
Stop
it
somehow
to
find
it
and
second
question
and
if
there's
an
event,
it
means
there's
an
also
queue
and
there
and
our
resource
are
already
limited.
So
we
may
not
just
as
fire
the
same
function,
maybe
not
ten
times.
Maybe
we
will
give
a
slot
like
just
a
at
the
same
time
on
only
just
a
five
five
function
can
run
parallel.
A
I
think
you're
asking
very
specific
implementation
questions
in
particular
about
key
native,
which
are
great
questions,
but
can
we
save
that
for
after?
Because
we
because
a
lot
of
us
work
on
K
native-
and
we
can
answer
those
questions
in
particular,
but
I'd
rather
try
to
keep
the
conversation
more
generic
for
service
in
general?
Otherwise,
it's
gonna
turn
into
a
K
native
session.
Is
if
that's,
okay
with
you
I
I,
don't
I
do
want
to
get
to
it
says
I,
don't
turn
to
a
K
native
session,
so.
B
So,
okay
think
think
about
this.
What
the
state
of
computing
was
about
10
years
ago
versus
what
it
could
be
in
ten
years
and
the
function
in
half
state,
it's
in
some
sort
of
blob,
store
database
thing
and
we're
worried
today
about
connections
to
the
database
and
scale
of
functions
like
starving
the
service.
A
So
well,
you
guys
have
been
talking
about
events
and
I'm
curious.
Is
that
really,
where
you
see
services,
sweet
point
in
terms
of
processing
events,
or
is
it
just
generic
triggers?
You
know
any
kind
of
message
coming
in
or
is
it
really
just
events
in
your
mind?
That's
really
the
is
the
serverless
sweet
point,
because
I've
seen
people
use
service
frameworks
for
things
that
have
nothing
to
do
with
events
and
I'm
wondering
whether
that's
just
weird
or
whether
that's
just
something
most
who
aren't
thinking
of
yet
you
got.
A
P
The
my
main
concern
was
I:
don't
want
to
occupy
the
resource
on
the
edge
for
like
the
implement
the
micro
service
or
VM
or
something
in
the
edge.
You
cannot
get
rid
of
it.
Even
you
don't
have
the
jobs
running
right,
so
my
main
concern
was
putting
servers
there,
so
you
can
only
occupy
the
resources
when
you
have
actual
jobs,
so
that
was
my
thought
to.
P
Right
because
I
have
something
important
running
on
the
edge
right,
something
else.
Let's
say:
nfe
and
I
have
some
available
resources
for
general
purpose,
but
those
resources
not
available
well
for
anytime.
This,
for
example,
the
other
application
can
scale
out.
Then
these
available
resources
will
be
go
in
this
in
milliseconds,
okay,
so
I
don't
want
to
keep
it
up
for
the
any
other.
General
purpose
tasks.
Okay,
so
that
was
my
use
case.
Okay,.
A
J
A
J
Okay,
yeah,
we
built
our
platform
because
there
it
was
not
there
at
the
market,
so
we
will
do
part
from
using
the
communities
which
I
gave
a
talk
yesterday.
But
the
point
is
that
we
treat
there
is
a
trigger
for
that
death.
So,
like
a
session
open
event
or
something
there,
we
go
start
a
video
conference.
That
kind
of
event
to
the
fire.
I
mean
spawn
up
the
function,
but
actually
even
actor.
Input
for
that
function
is
streaming
data.
So
that's
something
that
probably
it's
a
little
bit
different
to
the
typical
function.
H
A
lot
sorry
and
me
yeah,
the
client
I
did
my
demo
with
yesterday
is
Amazon.
They
are
also
doing
media
workflows
and
those
are
super
super
super
event-driven,
like
you,
have
a
file
coming
coming
in
a
movie
frames.
Whatever
you
want
Transco
them,
it's
very
much
even
driven,
and
it's
a
perfect
use
case
for
serverless
and
getting
back
to
the
earlier
question.
Why
are
people
why
we?
Why
are
some
people
using
kubernetes
but
not
server?
H
Less,
like
we
looked
into
running
several
s
frameworks
on
kubernetes
and,
in
my
opinion,
on
our
opinion
and
I,
speak
for
myself.
In
my
opinion,
those
who
are
not
yet
ready,
like
you,
get
more
benefits
and
a
high
reliability
from
kubernetes.
A
lot
of
frameworks
are
getting
they're,
not
clear,
open,
fast
came,
a
tiff
is
getting
their
big.
Rogers
framework
is
getting
their
cublas
was
also
doing
a
lot
of
stuff
for
kubernetes
because
we
had
issues
with
containers
being
killed.
H
A
That's
interesting,
so
let
me
ask
a
different
question:
I
this
mainly
may
be
more,
for
the
people
are
actually
using
servlets.
What
is
it
about
serverless
that
appeals
to
you,
because
I've
heard
some
people
talk
about
it's
difficult
to
figure
out
how
to
break
things
up
into
functions
and
making
them
small
and
stuff?
That's
obviously
a
problem,
but
what
it?
What
features
of
server
lists
actually
drew
you
to
it?
Is
it
the
fact
that
it's
functions
and
small
little
pieces,
or
is
it
the
auto
scaling?
Is
it
being
able
to
scale
down
to
zero?
A
D
To
you
to
say
it's
server,
low
overheads,
overhead
of
not
having
servers,
and
it's
just
well.
You
still
have
to
do
some
kind
of
provisioning
and
put
those
lambdas
or,
like
anything,
you're
dealing
with
somewhere
right,
but
we
don't
know
we
don't
have
to
do
any
servers
and
you
kind
of
like
the
well.
The
normal
continues.
The
can
was
it's
like
infrastructure.
Basically,
the
infrastructure
related
overhead
that
we
have
to
otherwise
would
have
to
deal
with.
We
don't
so.
A
D
Of
the
arguments,
okay,
it's
it's
one
of
the
arguments.
The
other
argument
is,
of
course,
that
you
could
well
there's
you
well
well,
one
of
the
arguments.
Let's
see
there
are
other
others
are
less
important,
they
say,
but
the
other
one
is
well
I.
Guess
it's
just
conveniently
placed
somewhere.
You
know
it's
all
the
flexibility
that
comes
with
working
in
the
world
sort
of
in
the
cloud
or
it's
sort
of
like
a
platform,
but.
N
What's
a
huge
thing
for
us
is
just
how
well
it
integrates
with
other
proprietary
services
of
cloud
providers.
I
mean,
maybe
us
on
upside,
for
for
that's
the
table
on
Amazon
Web
Services.
You
have
data
in
Keeney's
streams
and
then
you
can
just
put
in
the
function
behind
that
and
they
are
triggered
on
events.
Something
like
this,
and
maybe
this
is
also
something
one
could
could
go
into
the
direction
of
how
how
to
glue
together,
maybe
open
source
services
or
other
standards.
N
A
So
that
we're
almost
imply
you
can
make
from
them.
That
almost
implies
that,
if,
if
lambda
was
there
by
itself
and
isolated
I'm
not
meant
to
manage
to
pick
on
Amazon,
but
if
you
had
something
like
lambda
in
isolation
without
the
back-end
services
available,
it
may
not
be
as
attractive
to
you.
That's
interesting,
yeah,
okay,
yeah.
M
For
business
applications,
the
extensibility
is
very
important
so
that
customers
can
evolve
in
their
own
speed
as
their
own
innovation
needs
and
agility
and
yeah.
This
kind
of
business
application
to
the
users
on
top
are
not
pretty
often
not
even
experienced
developer
right,
so
the
skill
set
is
low.
It's
a
human
factor.
Okay,
we
have
a
broader
reach.
We
can
reach
more
people
who
can
extend
business
applications,
and
that
makes
it
so
so
valuable,
interesting.
Okay,.
J
So,
from
my
perspective,
besides,
the
typical
benefit
I
feel
that
it's
a
opinionated
wait
I
mean
how
to
write
that
software.
So,
as
he
said
it,
we
need
to
implement
in
the
best
practice
matter.
If
we
do
the
service
and
cornutus
is
a
opinionated
platform
to
manage
infrastructure
application
and
I
think
server.
This
could
be
opinionated
way
to
write
a
software
so
I'm,
because
I
need
to
write
in
the
event
or
even
nothing
else
is
I
mean
allow
and
it's
allowed
by
it
will
mess
up.
O
Just
correct
me,
if
I'm
wrong,
so
I
think
you
we
can
think
a
case
like
increments,
to
run
a
batch
kind
of
jobs.
What
we
have
just
you
can
create
a
kind
job
or
you
can
run
a
cron
job
or
in
UNIX
philosophy.
You
can
have
a
cron
job
and
when
we
look
to
the
cron
jobs,
what's
the
main
things
in
period
of
time,
you
want
to
do
some
sorts
of
things,
but
just
a
also
there
might
be
some
requirement
that
all
the
function
provide.
O
This
on-demand
request
like
to
run
something,
and
if
we
don't
have
a
some
some
certain
off
period
to
run
those
on
demand
things.
You
cannot
create
cron
jobs
and
when
I
will
just
say,
you
need
to
trigger
a
job
to
do
something
like,
for
instance,
some
files
already
started
to
get
bigger
and
you
want
to
shift
them
somewhere
and
periodically,
and
you
already
have
some
criteria
to
actually
run
that
function
and
for
me
it
is
not
nice
to
create
job
metadata
job
pot
in
the
community
every
time
and
it
in
that
way.
O
A
Good
no
yeah
I
agree.
Thank
you.
Anybody
else
want
to
comment
on
that.
Okay
hold
on.
Let's
do
it.
Carolyn's
question
was
where
we
talk
a
little
bit
with
that.
Okay,
so
actually
relative
to
the
workloads
that
you
guys
are
seeing
for
service.
Are
there
certain
workloads
that
you
think
are
not
good
fits,
but
should
be
because
I
think
your
use
case
was
interesting,
the
streaming
one
because
I
think,
depending
on
who
you
talk
to
they,
some
people
would
say.
Yes,
we
have
the
exact
same
use
case.
A
You
want
functions
that
live
for
a
long
time,
whether
it's
because
the
input
data
or
the
trigger
is
a
stream
or
just
because
it
takes
a
long
time
to
get
his
job
done
either
way,
it's
going
for
a
long
time.
Other
people
look
at
that
and
say:
that's
not.
Service
service
is
short-lived
right
and
it
depends
on
your
mindset.
A
A
A
Right,
like
you
said
it
is
possible
to
do
it,
but
it
makes
it
much
harder
to
do
it
and
shy
away
from
it.
Okay
and
I
just
want
to
come
in
from
that,
okay,
but
that
does
lead
into
another
question
that
I
have,
which
is
a
night.
Well,
it
is
a
que
native
question:
it's
not
it's
kind
of
late.
It's
not
directly
creative!
One
of
the
things
that
I
personally
have
been
thinking
about
and
trying
to
promote,
mainly
because
of
key
native.
A
Is
this
idea
of
trying
to
bring
server
lists
and
containers
of
service
platform
the
servers
all
merged
together,
because
at
the
base
level,
it's
all
just
a
container
right
and
it's
sort
of
the
stuff
around
that
that
may
make
it
a
little
different
serve
lists.
You
have
auto
scaling
scale
down
to
zero
stuff
like
that.
But
do
you
guys
agree
that
this
merging
of
the
world
that
we're
seeing
with
things
like
Cana
is
a
good
thing,
or
do
you
actually
think
that
there
really
are
separate
pillars
like
no
Cloud?
Foundry
is
distinctly
different
from
vanilla?
L
L
L
Open-Source
right,
then,
then,
you
can
just
put
your
service
function
out,
there
can
be
popular
and
people
can
use
it,
and
you
can
tell
how
many
people
have
used
it
yeah
right
just
the
same
as
you
would
for
a
docker
container,
so
you
don't
need
to
have
dock
up,
just
use,
github
and
and
then
the
thing
is
then
the
response
I
think
solar
service
is
different
from
containers.
I
think!
L
Yes,
you
can
just
in
the
same
way
that
you
can
run
Cuban
Iggy's
on
top
of
a
VM
and
the
same
way
that
you
run
this
que
native
thing
on
top
of
Cuban
et's.
It's
all
a
response
to
try
and
keep
in
a
way
I
kind
of
check
in
a
balance
right
so
that
at
the
other
side,
there's
always
the
option
of
going
with
this
open
source
open
standard
and
that's
really
useful
and
I
think
that's.
They
are
different.
Server,
listened
and
communities
and
containers
and
whatever,
but
they
represent
the
open.
M
A
A
L
Some
people
will
but
we'll
get
it
right,
like
federal
customers,
they
are
going
to
have
their
own
kit,
but
they
also
might
like
VMs,
and
then
they
will
also
like
to
put
Cuba
niches
on
there
and
then
they'd
also
like
to
have
service
themselves
all
implemented
in
their
own
black
ops
data
center,
which
I
can't
talk
about.
Otherwise
I'll
have
to
kill
you
all,
but
you
get
the
idea
right.
Yes,
sir.
Yes,
of
course,
these
things
me
it
will
all
be
separate
as
technologies
they're,
separate
and
as
like
operate.
You
use
them
differently.
L
They're
operationally
different.
You
have
different
tea.
You
might
not
have
different
teams.
You
have
different
skill,
sets
right,
you're,
never
gonna,
so
that
the
point
that
he
made
about
the
observability,
which
I
take
by
the
way
that
one
about
you
should
have
in
every
function
like
the
necessary
stuff
to
log,
not
log,
you
know,
create
the
right
matrix
or
whatever,
so
that
you
know
that
makes
perfect
sense.
That
is
the
answer
to
the
question
of,
as
observability
that
you
actually
have
to
every
function
must
tell
something
else.
L
Of
course,
that's
gonna
generate
a
heap
of
traffic,
but
we'll
talk
about
that
a
different
time,
yeah,
so
I,
don't
I,
do
see
them
as
different.
Absolutely
but
I
think
that,
to
my
mind,
that's
how
I
think
commercially
it's
just
like
there
has
to
be
a
way
of
creating
a
standard.
We
had
the
same
problem
like
20
years
ago.
We
were
all
writing
RFC's
right
and
all
the
kids
that
were
there
writing
the
RFC's
I
was
back
then
I
was
a
kid.
L
Then
we
were
like
doing
the
same
thing.
We
were
saying:
oh,
this
is
how
we're
going
to
do
it
and
we
were
agreeing
and
we
agreed
standards
and
then
the
internet
and
all
the
different
standards.
You
see
today,
like
all
the
different
broadcast
multicast,
all
these
kind
of
things.
It's
the
same.
Now
we're
basically
doing
the
same
Pro
and
accept
the
RFC's
or
github
repos
right.
I
L
If
you
have
a
github
repo
you're,
just
it's
exactly
that
it
feels
the
same
as
when
we
were
writing
these
RFC's
and
that's
what
you're
you
write
a
github
repo
with
a
bunch
of
code
in
and
then
you
get
to
present
at
the
conference
and
then
and
then
it
becomes
a
standard
for
the
industry.
Yeah
I
think
this.
Is
that
really
exciting,
because
it's
that
part
where
we
go
from
yeah,
we
always
had
a
proprietary
way
of
transmitting
packets
across
the
internet.
L
O
Limitation
of
service
I
mean
the
exact
meaning
of
service,
and
the
guy
in
here
already
mentioned
about
a
patches
term,
and
there
are
some
concepts
and
actually
I,
don't
know
where
services
already
fits
in.
For
instance,
Apache
storm
is
the
really
good
implementation
of
flow-based
programming,
which
is
which
can
came
from
1970s
what
it
tells
it
all
to
have
execution
input
and
output,
and
also
which
solace-
and
there
are
too
many
function-
words
that
already
relies
on
it.
So
do
we
mean
with
service
as
a
methods
or
pure
functions,
so
actually
there
are
three
concepts.
O
So,
as
is
the
difference
of
the
methods
and
the
functions
in
a
simple
way,
what
we
have
in
golang,
for
instance,
we
have
injector
right
for
the
functions
if
you
already
keep
the
object
inside
the
function,
so
it
is
not
a
functional.
It's
just
a
method
already
gary
retrieve
the
object
and
state
it
can
change
it
or
it
might
not
change
it
and
then
create
some.
O
You
do
some
state
do
some
things
and
in
that
so
for
methods
actually
solders
I
would
like
to
use
it
that
matters
I
mean,
and
my
events
will
be
the
main
object
and
with
that
event,
I
would
like
to
run
my
methods
in
the
self-service
domain,
or
should
it
be
on
the
go
through
pure
functional
like
Haskell
function,
implementation.
So
maybe
it
might
be
good
to
really
clarify
the
borders
of
what
we
mean
by
service
they're,
going
to
use
like
flow
base
like
methods
or
like
pure
function.
A
Anybody
have
any
I,
don't
person
have
an
answer
to
that.
Anybody
else
want
to
comment
like
I.
Think
it's
a
great
question.
I
think
it
goes
back
to
where
we're
saying
earlier
is:
where
do
you
draw
the
boundary
between
microservice
function?
How
do
you
split
it
up?
Where
do
you
split
it
up?
I,
don't
think
it
was
a
clear
answer
for
that.
Anybody,
wanna
yeah.
O
E
I
had
a
similar
comment
that
maybe
also
this
relates
to
is
that
one
of
the
things
that
we
really
find
that
especially
at
our
customers,
that
using
serverless
in
a
very
productive
way
for
the
customers
is
still
some
of
the
on
mysteries.
That
is
kind
of
black
knowledge
kept
inside
some
people.
Heads
who
already
successfully
did
that
inside
some
some
I
don't
know
companies
or
something
exactly
as
it
was
I,
don't
know
tens
of
years
ago
for
how
to
do
proper
development,
software
development
and
before
the
open
source
world.
So
kinda
me
I'm
thinking.
E
A
Cuz
we
definitely
heard
people
want
the
best
practices.
I
think
it
goes
hand
in
hand
that,
but
it's
not
the
phrase
best
practices
doesn't
describe
it.
It's
more
like
document
your
experience
with
it,
your
your
findings,
your
learnings,
so
other
people
can
learn
from
that
as
well.
I,
don't
know
what
the
word
is
for
that,
but
yeah.
E
Okay,
one
other
comment:
it's
related
to
that
question,
which
there
was
a
previous
talk
by
yonas,
the
City
of
Light
Bend
in
this
conference,
about
fateful
functions,
which
is
pretty
interesting
and
also
made
me
personally,
at
least
think
that
the
model
of
the
actor
model
that
was
at
some
time
the
peak
of
the
interest
in
the
in
the
industry,
probably
fits
in
the
service
world
and
also
probably
on
the
cover
in
this
environment
some
sort.
So
this
is
also
something
that
currently
is
not
under
heavy
focus
of
service,
but
probably
fits
there.
Yeah.
J
Hello,
actually,
a
comment
to
your
question
about
if
Cerberus
should
nourish
with
I
mean
tie
with
container
or
that
I
don't
have
answer,
but
already
left
said
that
wave
assembly
Syrian
wave
assembly
was
is,
is
emerging,
yeah
I,
don't
know
if
whether
assembly
is
was
marry
well
with
docker
containers
at
the
first
place.
So
just
a
curious,
but
well.
F
A
Native
is
advertised
as
a
survivalist,
depending
on
who
you
talk
to
uh-huh
I
personally,
use
it
as
kubernetes
I,
don't
use
it
as
service
I
have
use
it
a
service,
but
I
don't
usually,
and
so
to
my
mind,
they
are
kind
of
merged
together.
So
you
asking
I
think
you
were
asking.
You
know
what
other
people
who
actually
are
trying
to
merge.
It
I
think
key
native
I
mean
yeah
I
mean.
J
A
J
H
Since
we're
discussing
serverless
like
we
are
focusing
on
server
list
functions,
but
there
are
a
lot
of
other
server
less
services.
I
guess
but
sounds
weird
like
s3
could
be
considered
server
less
we
have
stuff
like
or
or
a
server
less
like.
There
are
other
things
that
could
be
considered
server
less.
Do
you
believe
the
service
working
group
should
do
something
about
that
and
it's
so
what
wait.
A
H
A
Well,
personally,
I
think
the
work
I
think
anything's
fair
game
as
long
as
it
doesn't
stray
too
far
from
that.
But
what
other
do?
What
do
people
think
I
mean
is,
is
extending
it
beyond
service
functions,
something
that
you
guys
think
we
should
consider
so
just
amplify
you
just
yell
into
it.
It's
not
it's
on
okay
and
then
we're
almost
out
of
time.
I'll
go
to
his
commenting
them.
We
can
stop
so
I.
Just.
C
Think
that
surveillance
is
inclusive
of
everything,
and
so
this
functions
to
be
completely
different.
Fromm's
Ovilus,
yes,
pods
can
be
deployed.
Yes,
I
think
personally,
if,
if
you
have
a
managed,
kuben,
avenge
and
I,
think
it's
freaking
so
ready,
because
I'm
not
really
managed
nodes
right,
but
solace
functions
is
just
one
small
unit
like
a
function.
That's
precisely
what
I,
what
at
least,
what
I
think
and
I'm
frankly
surprised
that
everybody
seems
to
confuse
the
two
together.
A
P
I
think
this
could
be
get
more
complex,
I
mean
just
a
simple
function,
not
equal
service
right.
It
can
be
calling
functions
through
the
functions
and
something
like
Franzen
chaining
that
could
be
equal
to
the
method.
That's
generally
saying,
early
and
I
think
we
should
have
more
prohibition
on
the
server
less,
because
it's
not
about
the
functions.
There
are
some
other
technologies
like,
for
example,
like
a
media
streaming
it
it's
not
the
like
simple
function
right,
so
it
could
fetch
data
from
the
outside
and
you
have
another
load
of
functions
working
inside
the
function.
P
A
Right
cool
just
have
opinion.
Oh
that's
good!
That's
good!
I'm,
glad
to
hear
it
and
technically
we
are
out
of
time.
I
think
you
guys
very
much
for
joining,
hopefully
of
those
who
are
not
using
a
server
list.
Yet
here
in
the
discussion
me
give
you
some
insight
into
some
of
the
issues
we
have
to
think
about
when
to
use
it
when
not
to
use
this.
A
Hopefully
it
was
useful
to
even
see
you
guys
are
not
using
it
yet,
but
from
our
point
of
view
from
a
service
point
of
view,
thank
you
for
all
your
input,
and
hopefully
this
was
a
little
bit
entertaining
if
nothing
else
to
hear
that
is
back
and
forth.
Discussions
do
have
some
information
here
about
the
service
working
group.
The
cloud
events
information
if
you're
interested,
but
thank
you
guys
for
making
it
an
interactive
discussion.
I
was
a
little
frayed.
You
guys
gonna
be
quiet,
but
you
did
step
up.
I
appreciate
it.