►
From YouTube: CNCF Serverless WG Meeting - 2019-02-28
Description
Join us for KubeCon + CloudNativeCon in San Diego November 18 - 21. Learn more at https://bit.ly/2XTN3ho. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy and all of the other CNCF-hosted projects.
A
B
B
B
So
if
you
guys
have
an
SDK
or
if
you're
involved
in
one
of
the
SDKs,
please
keep
an
eye
out
for
that,
because
we
don't
want
someone
to
to
miss
out
opportunity
to
participate
in
that
and
then
related
to.
That
was
the
whole
discussion,
mainly
by
the
work
that's
going
on
and
go
SDK
around
and
versioning
in
particular,
what
version
of
the
spec
should
people
support
going
forward
so,
for
example,
the
go
SDK
it's
currently
trying
to
support
all
three
versions
of
the
spec
and
also
trying
to
keep
up
with
master
as
it
goes
along.
B
So
we
kind
of
start
having
discussions
on
this
SDK
site.
You
know
about
what
are
the?
What
are
the
expectations
from
the
other
SDKs
as
well,
and
obviously
that's
probably
going
to
be
very
much
related
to
what
we
choose
to
do
like
ooh
Connie,
you
around
its
events,
so
just
FYI
for
these
discussions
that
are
going
to
happen
on
these
two
keysight
channel
markers,
scott
or
anything
else.
You
guys
wanted
to
add
to
that.
They
may
forgot
it.
C
B
B
D
Remember
yeah,
so
there's
there's
interest
in
this
real
world
example
in
general,
it's
a
some
sort
of
pipeline
where
there's
producers
and
then
there's
like
some
sort
of
inner
routing
piece
and
then
there's
consumers
of
events,
and
my
original
demo
proposal
was
the
simulation
of
an
e-commerce
pipeline
and
then
Doug
brought
up
there's
this
partnership
that
he's
been
working
with
and
through
other
working
groups.
Doing
an
example.
D
What
is
called
a
smart
city
for
an
airport,
so
potentially
the
demo
could
migrate
to
be
same
sort
of
situation
where
there's
event,
producers
and
consumers.
But
the
focus
is
around
how
an
airport
views
itself
and
maybe
can
react
to
two
events
that
are
happening.
You
could
still
get
that
same
ecommerce
pipeline
because
there's
merchants
and
shipments
and
other
people
buying
stuff
in
an
airport
and
then
the
the
the
thing
could
grow
to
simulate
what
an
airport
is
eventing
at
any
given
time
and
then
I
think.
D
The
overall
goal
is
that
each
consumer
and
producer
can
use
the
events
to
do
something.
Interesting
and
the
airport
itself
could
actually
view
all
of
its
components
as
and
then
make
some
sort
of
determination
about
the
health
of
the
airport,
and
so
that
would
be
like
the
long-term.
If
you
continue
working
on
the
demo,
this
is
what
it
does,
and
so
each
vendor
would
like
potentially
act
as
a
entity
inside
the
airport,
doing
events
and
consuming
things.
So
if
there's,
we
think
it's
interesting,
and
hopefully
everyone
else
thinks
it's
interesting
enough
to
participate.
I.
B
A
Evolved
an
information
system
that
extends
you
know
from
flight
systems
down
to
transport
patients
claim
retail
components
it's
all
under
the
they
call
it
accros
ACR
is
a
is
for
airport
and
a
few
I
don't
know
a
dozen
prominent
international
airports
have
implemented
that
model,
Heathrow
being
kind
of
the
lead
on
that
and
then
in
the
u.s.
I
think
San,
Francisco
and
Orlando
or
are
behind
it.
So
it's
it's
gaining
momentum
and
so
I
see.
This
is
something
that
would
be
mutually
beneficial
to
connect.
B
A
B
E
E
Everything
from
what
happens
when
a
flight
is
delayed,
orchestrating
all
actors,
elements
within
the
airport
across
the
airline
to
the
airport
itself
in
an
event-driven
way
using
our
event
Gateway
project
and
we
spent
a
lot
of
work,
did
a
lot
of
research
on
this,
and
you
know
we
had
a
whole
story
and
we
can
start
it
with
just
this,
this
linear
sequence
of
events
and
who
would
respond
and
how
they
response
we
haven't.
We
have
a
whole
bunch
of
materials
from
that.
If
any,
if
it's
of
interest
to
anyone.
B
B
Punko
I,
don't
know
what
the
next
phone
call
is,
but
I
don't
think
usually
much
change
there.
Other
than
I
did
find
out
that
the
surrealist
practitioners
summit
is
still
moving
forward.
I
believe
they're
planning
on
having
a
CFP
type
of
setup
so
expect
to
see
some
notes
about
that
bill
to
be
soon
so
I
think
they're,
basically
looking
at
it
being
a
basically
a
mini
summit
kind
of
a
thing
with
the
keynotes.
F
B
Then
breakout
sessions
and
stuff
like
that
from
people
looking
some
but
proposals
for
talks
and
stuff,
assuming
they
do
that
we
probably
need
to
figure
out
whether
we
should
have
our
surplus
working
group
meeting
as
part
of
that
or
still
under
the
troll
coupe
common
thing.
I
think
it
might
be
a
little
still
premature
to
try
to
make
that
kind
of
decision.
Guess,
I,
don't
know
exactly
what
the
format
is
for
this
other
summit,
but
I
just
thought:
I'd
bring
you
up
to
speed
that
they
are
still
going
forward
like
that.
B
It's
not
the
silence
does
not
mean
it's
gone
away.
All
right.
Let's
see!
Okay,
so
I
did
ask
for
35
minutes
intro
and
deep
dive.
Two
separate
ones
there
for
coop
Connie
China
and
asked
for
an
80
minute
long
session
for
service
to
match
up
what
we're
doing
a
coupon
EU
just
lets.
You
guys
know
that
I
did
request
that
I
believe
the
call
papers
is
open.
B
So
if
you
guys
want
to
submit
papers
just
in
general
around
any
topic
at
all,
whether
a
service
or
not
I
just
reminded
you
that
the
CFP
period
is
open
now
and
with
that
I
believe
we
can
start
talking
about
PR.
So
I'm
just
do
a
quick.
Let's
go
check
you
about
one
thing:
okay,
no,
more
votes;
okay,
so
Rachel's
PR!
So
last
week,
mr.
no
votes
to
it.
Oh.
B
A
B
So
we
took
votes
these
four
choices.
Nothing
just
a
list
of
the
external
specs
include
the
specs
and
then
include
the
specs
with
the
TCK.
Now,
hopefully,
I
did
thought
the
voting
rights
I
had
to
convert
the
numbers.
You
guys
gave
me
into
the
format
that
the
tool
actually
wanted,
meaning
for
four
columns
and
with
a
number
in
the
column
representing
your
preferred
choice.
So,
for
example,
for
Google
their
first
choice
was
number
three,
which
is
the
specs,
their
second
choice,
the
specs
with
SDK
or
tzk,
and
then
third
choice
was
just
lists.
B
So
look
at
this
lift
this
isn't
any
Tim
in
stock.
You
guys
wanna,
take
a
look
at
and
verify
my
nice
stuff,
but
I.
Think
it's
right
later
on.
I
will
take
this
and
run
it
through
the
official
tool,
but
I.
Think
if
you
just
look
at
it,
you'll
see
number
one
does
pop
up
in
column
number
two
most
often
and
I
will
double
check
that
through
the
tool,
but
I
think
it's
pretty
obvious.
That
number
two,
which
is
just
the
list.
It
is
gonna
win.
So
that
would
be
the
step
point
forward.
B
B
H
Yeah,
let's
talk
about
the
new
one,
first,
okay,
so
the
old
one
we
discussed
a
couple
of
times
so
basically
main
goal
is
that
we
have
a
event
a
size
of
an
event,
and
we
know
that
it
will.
Everyone
should
accept
it.
That
means,
if
I'm
as
a
event,
producer
I'm,
creating
event
that
is
below
the
size.
I
am
can
be
certain
that
everyone
that
follows
me
accepts
it
unless
they
have
really
big
reasons
not
to
because
there
are
some
supers
constrain
device,
but
like
in
the
cloud.
H
Let's
say
everyone
should
accept
it
after
me,
so
the
proposal
I
gave
last
time
was
that
just
take
the
events
utilize
it
as
JSON,
even
if
you
don't
send
it
with
Jason
on
the
wire
and
I,
was
criticized
because
yeah
well,
if
you're
not
sending
Jays
nobody
wire,
you
have
to
do
one
serialization
in
Jason
that
sort
of
pointless
what
I'd
like
to
point
out
that
the
good
part
or
we
are
requiring
everyone
to
support,
raise
in
any
way.
So
everybody
should
have
this
implemented.
H
But
after
that,
critique
that
we
maybe
shouldn't
force
the
serialization
in
Jason
I
tried
a
different
way
or
I
here,
I
described
a
different
way,
how
to
measure
the
size
of
any
went,
independent
of
any
are
encoding
and
formatting
and
so
on.
So
this
is
what
the
star.
So
these
are
a
few
more
rules.
So
there's
limit
on
the
number
of
attributes
on
the
attribute
name
length.
H
We
basically
already
have
that
in
the
edge
with
spec
itself
and
then
individual
limits
on
binaries
on
strings,
which
are
basically
those
that
are
unbound
and
then
the
last
bullet
point
is
a
well
as
size
limitation
for
all
attributes
together
and
then
we
can
discuss
what
the
individual
size
would
be.
But
the
point
here
is
that
we
measure
that
independently
of
the
encoding,
so,
for
example,
an
integer
would
always
be
4
byte.
H
H
H
So
to
summarize,
in
comparison
to
the
proposal,
I
made
last
time,
you
don't
have
to
go
through
to
Jason's
utilization.
That's
the
good
part.
The
other
good
part
is
that
the
limits
are
much
fine
granular
and
you
know
you
also
have
then
an
limit
on
the
number
of
attributes.
For
example,
the
downside
is
that
everyone
who
wants
to
well
follow
these
rules.
They
have
to
implement
something
new.
They
should
already
have
deejays
and
civilization,
implemented.
A
I
H
I
think
there
will
be.
Let's
say
there
will
be
some
event:
producers
who
will
simply
always
be
below
these
limits,
because
they
only
have
a
few
attributes
and
their
data
will
not
be
so
big.
So
for
them
it's
really
an
unconcerned,
and
then
there
are
others
who
may
want
to
push
it
and
I
think
for
them.
H
So
I
think
we
should
settle
on
something
so
that
every
middleware
knows
what
it
has
to
support
at
a
minimum.
So
I
think
if
I
as
a
middleware
now
would
say:
okay
Jason
until
256
kilobytes
I
just
accept
that
then
you're
fine,
because
that's
already
that
will
definitely
hold
anything
according
to
these
rules.
So.
F
I
H
I'm
with
you,
then
okay,
thank
you.
So
the
idea
is
to
limit
the
other
attributes.
Is
that
as
a
middleware,
you
want
to
cars
all
them,
or
at
least
potentially
parse
and
look
at
them.
So
if
you're
limited,
we
also,
basically
what
all
HTTP
service
do.
Did
they
limit
the
size
of
the
headers?
So
this
is
sort
of
similar
so
that,
as
a
middleware,
you
kind
of
have
a
stopping
ground
of
how
much
stuff
you
have
to
process
and.
I
H
The
last
bullet
point
includes
the
data
attribute.
So
basically,
if
you
have
no
well,
you
will
have
a
few
other
mandatory
attributes,
but
if
they're
really
small,
then
you
can
have
a
larger
data
attribute.
But
if
you
put
more
stuff
into
the
other
attributes,
you
will
have
less
size
for
your
data.
Okay,.
H
B
Nothing
interesting,
okay,
it's
the
mask
woman.
So
this
is
not
my
area
of
expertise
in
the
slightest.
So
let
me
ask
this
question
because
I
definitely
understand
the
desire
for
some
sort
of
minimum
size.
I
can
understand
that
and
from
our
overall
perspective,
but
relative
to
spelling
out
individual
things
like
you
know,
in
what
were
the
size
of
each
individual
property.
How
often
is
that
a
concern?
People's
experience
relative
versus
the
entire
size
being
a
problem.
I
B
But
I
guess
I'm
kind
of
wondering
great.
Obviously,
people
are
concerned
about
this
size
issue
for
a
reason
and
I'm
trying
to
figure
out
where
those
size
limitations
come
into
play
in
practice.
Is
it
because
the
transport
can
one
handle
64k
or
is
it
because
the
processing
engine
of
these
of
these
payloads
only
supports
things
like
you
know,
after
your
names
that
are
only
good,
like
20
characters
max
right,
I'm
trying
to
figure
out
where
people's
experiences
are
in
this
space
because
I
just
don't
have
it?
That's
why
I'm
curious
I.
I
Believe
it
was
an
interrupt
issue
yeah,
so
you
know
if
they
don't
I
know
if
they're
ones
from
AWS
is
on
the
line.
But
if
AWS
can
support
300k
events,
but
Azul
can
only
do
64,
then
there's
the
media
impedance
mismatch
there.
So
I
think
it
was
I.
Think
was
just
to
get
at
the
lowest
common
denominator
crystal
vine.
It
was
that
where
this
all
originated
from
basically.
H
Yes,
so
I
think
for
me
that
the
point
is
that
I
I'm,
an
event
producer,
I
produce
events
and
at
some
point
right,
someone
can
decide
to
send
them
to
an
editor
is
a
every
service
whatever,
and
my
events
should
go
through
all
of
these
services,
but
for
that
as
a
event,
producer,
I
kind
of
have
to
anticipate
what
the
size
limitations
of
each
service
will
be
and
the
only
place
I
can
look
for
guidance
on
that
is
the
spec
I
think
so.
Yes,
that
is
exactly
the
concern.
C
B
G
G
G
G
What
is
it
message?
Queue
event
grid
event
cream
because
you're
not
measuring
the
size
on
the
wire
you're
measuring
on
intermediary
format
right.
So,
if
the
problem
is
events
fitting
onto
specific
transports,
I
don't
understand
how
measuring
enormous
informant
that
doesn't
actually
measure
the
end
size
helps.
H
H
H
Big?
Why?
Why
do
I
want
the
name
lengthier,
because
if
they
become
much
longer
than
the
whole
computation
for
the
overall
size
becomes
more
complicated,
because
if,
if
they
can
be
like
thousands
of
characters,
long
then
I
also
have
to
measure
the
attribute
name
length.
I
can
also
do
that.
If
that's
before
I
see
thanks.
B
B
It
might
be
your
I,
don't
know
what
they
want
to
call
your
choice
or
your
responsibility,
one
of
the
two
to
decide
which
one
you'd
like
to
put
forward
for
the
group
to
consider.
If
you
want
to
do
either
one
I
thought
I
say
you
could
choose
to
close
both
of
them
if
you
want,
but
at
some
point,
I
think
since
they're,
both
your
PRS
usually
decide
the
next
step
forward
in
terms
of
what
you
want,
the
group
to
decide
yes
or
no
one.
Is
that
sound,
fair.
All.
H
D
D
G
J
B
So
that
raises
an
interesting
question
in
my
mind
anyway.
Is
it
clear
from
the
text
here
that
all
these
rules
apply
to
an
individual
events?
I
think
it
does,
because
it
says
right
here,
except
events
as
opposed
to
transport
thingy,
but
ever
going
to
call
it
transport
payload.
Is
it
clear
that
this
is
just
about
individual
events
and
that
this
answer
to
Scott's
question
should
should
already
be
in
the
text
or
do
we
need
to
add
the
additional
text
or
make
pretty
good
blur.
B
Okay,
I'm
not
anyways,
because
I
guess
it's
okay,
the
way
it
is
okay,
so
hopefully
people
please
really
leave
comments
on
they're,
not
hearing
anything
else.
This
may
be
the
one
performer
next
week
for
people
to
say
yes
or
no
on
so
be
prepared.
For
that
you
don't
like
it
add
comments
all
right,
Thank,
You,
Kristen,
very
much
for
your
patience
as
Jim
said.
B
Okay,
excellent,
let's
see
Clement
is
not
here
so
I,
don't
I
think
we
talked
about
this
one
last
time
the
data
encoding
thing,
but
I,
don't
think
I
made
any
progress.
I
think
he
may
need
to
go
back
and
rest
comments
on
that.
But
since
he's
not
here,
we
can't
talk
about
it.
However,
he's
next
to
I
don't
want
to
talk
about
them
per
se,
since
he's
not
here
ever
I
do
want
to
draw
people's
attention
to
them.
The
first
one
is
just
that
are
adding
an
architectural
section
to
the
primer.
B
It's
because
the
primer
is
non-normative,
it
doesn't
technically
impact
the
spec,
but
it
does
give
insight
into
what
people
are
thinking
about,
how
they
should
use
the
spec
or
what
our
design
decisions
were.
So
please
take
a
look
at
that
when
you
get
a
chance,
I
want
to
make
sure
that
it
accurately
represents
the
consensus
of
the
group.
Likewise,
with
the
SDK
object
model
PR
that
he
opened
up,
this
is
going
to
be
making
changes
to
a
document
in
our
in
our
repo
itself.
B
There's
a
SDK
dot,
NV
file,
I
believe
he
in
this
document
put
together.
Some
pretty
strong
recommendations
for
what
SDK
should
and
should
not
do
relative
to
how
things
get
exposed
to
user
and
stuff
like
that
and
I,
really
think
the
SDK
authors
need
to
think
look
at
that
make
sure
they're
okay
with
it
on
the
surface,
some
of
them
sound
reasonable,
but
at
the,
if
you
think
about
it,
it
actually
does
put
quite
a
big
requirement
on
implementation
details.
So
I
think
people
need
to
think
very
close.
Look
at
that.
B
But
I
wanted
to
get
a
general
sense
from
this
group,
whether
using
source
and
the
idea
together
is
heading
and
headed
in
the
right
direction,
or
whether
people
have
some
concerns
about
that,
because
I
know
the
co
people.
Only
a
few
people
make
comments
on
the
PR,
but
I
wanted
open
it
up
to
the
broader
audience.
So
let
me
just
sort
of
pause.
There
is
source
plus
ID
sound
right
in
terms
of
uniqueness.
What
is
that
reason.
B
J
A
H
J
G
G
If
an
open
source
project
uses
cloud
events
as
their
format,
they
will
have
to
somehow
make
sure
that
the
source
is
globally
unique
if
they
want
to
be
conformant
if
we
require
it
and
that
either
means
that
they
generate
random
strings
or
put
the
burden
on
their
user
to
make
it
global,
unique
and
I,
don't
think
either
scenario
makes
sense.
I
gave
an
example
along
these
lines
wherever
I
put
that
comment
where
you
can
see
more,
but
that's
my
thinking
here
so
are.
G
In
your
context,
it's
I
do
not
believe
that
if,
if
a
cloud
provider
has
a
thousand
deployments
using
that
same
open
source
project,
it
would
not
be
make
more
sense
for
them
to
prevent
the
customer
or
workspace
or
something
idly
before
the
source
and
event
ID
before
thinking
that
they
are
unique.
There's
no
way
they
could
actually
rely
on
our
spec
saying
that
the
events
are
global
in
unique
in
source
for
society.
It's
just
not
practically
possible
without
some
kind
of
global
registry.
B
L
Yeah
I
guess
my
point
was
you
know
this
is
a
problem.
That's
we've
been
no-hit
again
and
again
and
again
and
there's
lots
of
different
ways
of
solving
this
and
in
the
context
of
cloud
you'd.
Normally
he's
got
some
kind
of
security
context
and
the
security
context
is
something
that
you
do
have
some
kind
of
awareness
over
and
you
do
have
you
know
an
operational
model
for
defining
things
within
your
context
that
are
potentially
that
should
be
globally
unique
to
you
relative
to
everyone
else.
L
If
those
higher-level
security
context
are
also
unique
amongst
themselves
and
you
naturally
inherit
the
globally
unique
behavior,
but
at
the
same
time,
I
don't
think
miss
speck
is
far
enough
along
to
to
make
any
kind
of
grand
claims
about
that,
because
until
we
get
to
security
and
context,
and
things
like
that,
then
I
think
what
we've
got
for
now
is
something
that
could
evolve
naturally
and
I.
Think
this
problem
will
naturally
be
solved.
L
L
L
H
Agree
with
what
has
been
said
and
I
tried
to
write
that
as
comments
I
think
there
is
still
a
need
to
clarify
it,
but
I
guess
I.
Think
one
thing
we
can
do
is
really
specify
and
say
that
a
consumer
is
allowed
to
ignore
events
that
I
have
both
source
nerdy
and
maybe
when
type
or
source
nerdy.
Maybe
is
fine,
so
they're
allowed
to
use
that
for
the
application
added
application
so
that,
as
a
clarification,
is
good
and
then
maybe
in
the
primer
say
one
also,
while
he
rode
on
hey.
H
B
Other
people
think
about
that,
and
let
me
rephrase
it
let's
make
sure
I
understood
what
you're
saying
there
don't
necessarily
add
text
to
clarify
or
to
to
be
as
prescriptive
about
the
values
of
these
fields,
but
rather
focus
on
the
receiver
being
able
to
do
some
D
tube
logic
based
upon
source,
ID
type
or
something
like
that
and
then
leave
it
at
that.
Basically,
is
that
for
your
face
name,
fine
yeah,.
H
So
it
is,
you
can
do
DD
application
on
this
field.
That's
okay,
to
do
your
stolen
wallet,
con
consumer
of
events,
if
you
need
applique
to
events
based
on
these,
but
they're
actually
two
different
events,
that
is
not
your
sort
of
fog,
but
at
the
same
time
it's
not
the
event
producer,
the
code
that
has
that
must
ensure
it.
Basically,
that's
an
unfortunate
circumstance
in
over
all
of
your
application
and
negative.
G
I
think
that
sounds
sounds
good,
except
that
it
is
actually
in
a
single
deployment
or
actually
one
producer.
However,
you
want
to
define
that
is.
They
are
unique
in
in
the
context
of
the
producer.
So
when
you
say
that
the
producer
doesn't
need
to
care
about
it,
you
actually
does
according
to
the
spec.
B
Might
be
good
to
do
just
to
get
that
thought
process
out
there
think
about
it
from
from
my
perspective,
I,
since
no
one
else
is
jumping
up.
I
I
often
tend
to
wonder
about
things
like
this,
whether
these
things
I
think
as
both
some
people
have
already
said,
whether
these
things
kind
of
solve
themselves
right.
B
If
it's
going
to
cause
problems,
because
they
don't
do
that,
so
that's
why
I
tend
to
sometimes
think
that
you
don't
need
specs
to
be
too
prescriptive
here,
because
people
will
just
do
the
right
thing
anyway
and
by
being
super
strict
too
prescriptive,
we
actually
may
limit
the
usage
of
the
spec,
because
there
may
be
some
situations
where
a
person
either
can't
or
just
doesn't
want
to
be
that
unique
about
these
things
and
I'm
perfectly.
Okay
with
that,
so
I
thought
that's
where
I
might
had
kind
of
the
Llano.
B
B
All
right
moving
forward,
then
that's
it
for
the
PRS.
What
I
want
to
do
now
is
quickly
talk
about
some
of
the
security
issues,
so
I
think
one
of
the
big
milestones
or
one
of
the
big
items
we
have
in
this
milestone
is
to
address
all
known
security
issues.
I
believe
that
a
fair
number
of
them
fall
into
the
same
category
as
this
one
that
we
have
highlighted
here,
which
is
doing
things
like
encrypting.
The
data
determining
things
like
event,
confidentiality
and
stuff
like
that.
So
far,
most
of
those
have
either
gone.
B
Uncommented
or
someone
like
Clement
speaks
up
and
says:
don't
go
there.
It's
scary
and
we
end
up
saying
we're
gonna
deal
with
it
after
1.0
or
we'll
deal
with
it
as
a
follow-on,
suspect,
they're
on
top
of
a
cloud
event
or
something
like
that.
But
when
I
guess,
when
I
aren't
getting
to
here
is
I
wanted
to
know
whether
you
guys
on
the
call
are
okay
with
that
general
direction,
because
I
believe
at
least
to
the
PRS
or
issues
out
there
have
had
Clements
comment
on
it.
B
Basically
saying,
let's
not
go
there
and
I
haven't
heard
any
pushback
from
that,
but
I
wanted
to
get
it
I
want
to
bring
it
up
here
so
one
for
you
guys
to
look
at
those
issues,
but
to
gave
you
opportunity
to
voice
your
opinion
on
this
call
because
I'm
going
to
assume
at
this
point
that
silence
means
no
one's
really
that
interested
in
addressing
it
and
I.
Don't
want
to
make
the
incorrect
assumption
anyway
to
want
to
speak
up
on
these
things.
B
I
B
Please
doing
I
guess:
I
am
misspoke
a
little
there
I
think
at
least
in
one
of
the
comments.
Clemens
did
say
potentially
add
something
to
the
primer
or
maybe
it
was
meat
or
suggesting
it.
I
do
think
we
need
something
in
the
crimer
to
explain
why
we're
doing
nothing
if,
as
the
choice
we
do
make
just
so
people
know
when
he's
not
about
it,
I.
B
M
B
B
B
J
That
seems
a
little
bit.
You
know
that
seems
like
something
that
might
throw
off
some
JSON
parsers
or
some
libraries
I
think
type
is
actually
allowed
to
contain
Unicode,
so
you
could
actually
have
a
vent
type.
That
is,
you
know,
a
smiley
face
or
a
wink,
or
something
like
that
and
again.
That
seems
like
something
that
people
might
be
thrown
off
by
actually
seeing
on
the
wire.
B
Come
on
guys,
okay,
so
from
my
perspective,
I
thought
it
was
useful,
mainly
because
this
could
expose
how's
your
business
saying
you
know,
problems
in
the
spec
from
an
era
of
building
perspective
and
I.
Think.
The
last
thing
we
want
to
get
into
is
a
fight
between
two
people
where
the
producer
says:
I
am
producing
valid,
spec
compliance
stuff
and
the
receiver
gets
this
smiley
face
in
there
and
it's
completely
unexpected
and
everybody
else
says
no.
No,
that's
just
too
weird.
We
shouldn't
allow
that
so
I
was
thinking.
B
The
spec
might
need
to
be
a
little
bit
more
precise
on
these
things
as
you're,
suggesting
there
Evan,
but
in
terms
of
next
steps,
is
it
a
matter
of
just
looking
at
each
property
to
figure
out
what
rules
we
may
want
to
put
in
place,
or
is
it
more
of
a
TCK
kind
of
thing
that
we
need
to
created?
You
were
suggesting.
J
B
L
L
But
sometime
me,
my
background
is
in
events
and
streaming
and
so
when
I
think
about
events,
I
think
about
them
in
the
context
of
a
stream
and
that
stream
is
defined
by
the
key
and
that's
basically,
this
all
goes
back
today
to
modeling
as
far
as
I'm
concerned
and
I.
Guess
that's
the
lens
that
I
see
this
through,
because
I
work
for
confluent
and
you
know
with
a
company
that
effectively
behind
Kefka.
L
So
it's
hard
for
me
to
see
how
it
could
be
the
responsibility
of
the
consumer
to
determine
what
the
stream
key
is
when
you
think
about
a
relational
table.
When
you
do
your
data
modelling,
you
define
what
the
key
is
for
that
row
and
that,
for
me,
is
like
exactly
what
this
scenario
is
doing.
It's
it's
largely
a
date
and
one
during
exercise
and.
B
So
the
reason
I
hired
to
bring
this
one
up
here,
aside
from
Neal
being
on
the
call,
is
because
Neal
that
what
sentence
that
I've
highlighted
in
there
I
thought
was
probably
the
key
one
for
me.
Reading
the
latest
set
of
comments
and
I
wanted
to
get
a
sense
from
there.
Other
people
on
the
call
when
it
comes
to
a
receiver
being
able
to
put
events
into
particular
buckets
way.
I
guess
Kafka
does
who
typically
defines
what
bucket
it
goes
into.
I
B
K
G
It's
the
original
producer
has
no
way
of
knowing
how
some
intermediary
want
to
partition
or
split
the
event,
and
that's
the
point
that
Clemens
is
making
not
that
it's
the
consumers
responsibility,
but
that
it's
the
responsibility
of
the
part
or
the
the
whatever
is
putting
it
in
the
Kafka,
for
example,
it's
their
responsibility
to
set
the
key
here
and
if
it's
the
semantic
key
as
you're
talking
about
modeling,
that's
the
source.
We
already
have
a
feel
for
that.
L
L
Yeah
that
one
of
the
comments
on
this
is
proposing
that
we
pull
it
out
of
the
data,
payload
I,
don't
think
that's
very
secure,
providing
anyone
access
to
the
data
payload
unless
they
have
the
correct
credentials.
The
other
aspect
on
top
of
that
is
allowing
partitions
and
topics
to
be
created.
That's
something
that's
done
at
design
time.
L
You
can't
just
go
and
say
I'm
going
to
create
a
topic
with
two
partitions
or
5,000
partitions
or
even
200,000
partitions,
because
they
all
are
all
very
workload
in
whose
case-specific
it's
not
something
that
you
can
automate
like
a
database
table.
Will
you
define
it?
You
model
it?
You
shape
it
for
the
workload
characteristics.
It
can
never
be
done
in
an
automated
way,
because
the
implications
are
that
massive
sure.
G
Sure
it's
not
about
creating
the
topics,
it's
let's
say,
there's
a
producer,
hey
that
creates
the
event.
It
goes
through
cloud
aid
to
cloud
be
where
there's
another
producing
component
that
gets
the
like
that
access,
both
the
consumer
and
producer.
It
takes
the
event
and
wants
to
put
it
forward,
but
split
in
a
different
way,
partitioned
in
a
different
way.
Now
you're
saying
that
we
would
have
to
create
a
new
event.
Basically,
that
has
the
event
key
set
differently
instead
of
just
passing
on
the
event
but
party
different
with
so.
B
I
Gold
I
am
I,
think
I'm,
echoing
what
Neil
was
saying.
Maybe
I
should
clarify
my
first
comment:
it's
the
person
that
on
the
wire
who
gets
to
you,
decide
how
to
partition
it
yeah.
The
original
producer
will
do
that
at
the
time
he
originally
sends
it,
and
then
some
intermediary
will
do
that.
Each
half
along
the
way
and
and
who
knows
or
dare
I
say
who
cares
how
they
decide
to
do
that?
L
F
L
Said
come
add
one
final
comment:
yeah
I
prefer
to
go
with
the
simplest
solution:
I'm,
not
in
favor
of
introspection
or
having
adding
more
administration.
Overhead
I
mean
if
this
effect
can
evolve,
which
is
what
it
should
do
through
our
learnings,
and
the
simplest
thing
for
now
would
be
to
put
it
in
as
an
extension,
and
if
we
then
do
discover
that
it
doesn't
work,
then
we've
learned
and
we
can
make
a
decision
based
upon
that
because
at
the
moment
there's
a
lot
of
different
lenses
that
people
against
problem.
L
C
L
From
adopting
this
Claire
event,
because
we
see
a
lot
of
our
customers
wanting
to
use
cloud
events
but
at
the
same
time
I'm
not
saying
well,
let's
just
see
how
off
you
see
the
homes,
because
before
we
can
use
this
properly
within
capita,
we
have
a
few
kind
of
things
that
we
need
to
figure
out
right.
I,.
B
Know
IBM
actually
wants
to
get
the
Kafka
binding
done
to
so
we're
anxious
as
well.
So
let
me
think
the
selatan
then
take
this
action
item
to
go
off
and
talk
to
Clemens,
hopefully
tomorrow
offline
and
see
what
his
objection
is
to
making
it
a
Kafka
transport
binding
specific
extension
for
right
now
to
at
least
get
us
over
the
current
bump
in
the
road.
B
B
E
B
B
F
D
P
C
B
F
Another
thing
is
that
okay,
we're
an
open
source,
workflow,
automation,
vendor
so
I'm,
quite
active
in
the
mood
for
automation,
space
and
I'm,
currently
giving
a
lot
of
talks
talking
about
orchestration,
choreography,
event-driven
things
and
what's
currently
coming
up
quite
often,
is
that
even
in
serverless
worlds,
but
also
like
in
the
whole
cloud
native
space,
a
lot
of
people
are
yeah
missing
kind
of
work
for
functionalities.
All
its
all
vendors
are
building
something
like
AWS
step
functions
or
Ezzor,
durable
functions
or
Google,
Cloud
composer
and
all
these
kind
of
tools.
F
B
So
we
do
have
a
topic
on
the
agenda
for
the
the
bigger
service
working
group,
where
I
believe
Scott
and
then
we're
going
to
take
a
lead
on
seeing
what
we
needed
to
update
on
the
service
Docs,
and
that
does
include
the
landscape,
so
I
do
that
would
definitely
fit
into
there
obviously
very
nicely
I
have
what
I'd
like
to
do
is
put
your
name
there.
It's
not
right.
B
B
Sorry,
that's
like
so
I
was
thinking
burned.
This
is
obviously
a
topic
that
you
care
strongly
about,
may
be
worthwhile
for
you
to
take
a
look
at
the
that
the
terminus
dock
that
I
put
together
and
figure
out
what
types
of
changes
you'd
like
to
see
in
there
or
you
didn't,
go
as
far
as
to
write
an
additional
section,
if
you
think
that'd
be
worthwhile
or
even
what
sort
of
editorial
changes
you'd
like
to
see
basically
I
think.
F
B
B
B
Alright,
so
then,
going
back
to
the
bigger
issue
or
bigger
topic,
we
have
this
general
layout
here.
Have
you
guys,
given
any
thought
to
this?
You
guys
still
okay
with
the
general
layout
here,
otherwise
I
think
the
next
steps
might
be
just
put
together.
Some
outline,
you
know
PowerPoint,
slides
kind
of
a
thing.
How
do
you
guys
want
to
move
forward
out
here.
O
B
O
The
base
is,
yes,
now
supports
deployments
using
lambda
because
to
have
access
to
an
EPS
cluster,
you
can
use
an
AWS
I'm
raw
and
I'm
doing
a
proof
concept.
I
did
approve
concept
with
lambda
that
deploys
using
that
I
roll.
The
thing
is
you
do
your
deployment?
You
do
need
to
transmit
like
what
act
you'd
like
to
deploy,
what
version,
what
options
and
stuff
like
that,
and
that's
where
cloud
events
comes
in
I,
don't
know
what
code
I
could
run,
though.
O
A
B
Why
don't
I
reach
out
to
Chris
and
check
just
to
see
if
we
can
get
the
the
templates
if
they
want
us
to
use
for
this
stuff?
And
if
that's
available
I'll
just
create
some
some,
you
know
placeholder
files
and
then,
as
you
guys
get
time,
you
can
start
adding
more
to
it.
At
least
that
way,
it
feels
like
we're
at
least
make
it
some
sort
of
forward
progress,
and
that
way
you
feel
like
there's
a
document
out
there.
You
can
make
slow
and
incremental
progress.
One
slide
at
a
time:
hey
Doug,.
D
B
D
D
H
I,
just
one
a
little
bit
of
clarification
on
these.
Well,
basically,
these
third
thing,
the
service
rock
group
session.
My
name
is
under,
but
as
of
now
it's
kind
of
unclear
to
me
what
the
service
summit
is
exactly
going
to
be
and
if
it
will
make
sense
to
put
to
stop
I
want
to
talk
about
like
it's
part
of
the
subtlest
workgroup
or
if
that
would
rather
be
a
separate
talk
or
how
that
would
be
handled
best.
B
B
What
I'd
almost
rather
do
is
see
if
we
can
get
Chris
Anna
check
to
agree
that
maybe
we
should
do,
is
take
this
session
and
turn
it
into
a
session
at
the
service
summit,
but
I
guess
lower.
So,
okay,
so
I,
guess
that
literally
answer
your
question
about
whether
you
should
talk
about
your
stuff
under
here,
regardless
of
where
it's,
where
which
summit
it's
in
or
whether
you
should
do
your
own
CFP
for
the
sermo
summit
right.
What.
D
H
I
can
also
like
the
thing
I
would
talk
about.
I
could
also
make
a
cool
request
on
the
white
paper
and
then
maybe
we
can
have
a
session
for
multiple
multiple
people
talk
about
what
has
changed
on
the
landscape
and
yeah,
and
then
I
would
have
one
part
there.
So
everybody
talks
about
about
what
they
have
contributed.
Basically,
that
would
be
one
idea,
then
it's
sort
of
clear
that
it
comes
out
of
the
service
Raghu.
If
that
is
what
we
want
to
do.
B
H
Think
what
I
want
to
talk
about
is
there
are
vendors
that
bring
out
their
own
functional
service
to
use
with
their
service.
So,
for
example,
well
pay
Paula,
Braintree
release
the
functional
service
of
Xero
has
their
own
functional
service,
Twilio
has
their
own,
Adobe
has
their
own
and
so
on.
So
we
have
basically
function
as
service
providers,
who
are
not
a
general-purpose
cloud
but
are
really
specific
to
be
used
with
another
service.
H
H
On
personally
I
do
because
I
think
it
it
will
end
up
if
you
don't
have
a
standard
and
it
or
anything
physically
I
use
five
services
and
each
service
comes
with
their
own
function.
As
a
service,
I
have
to
have
I
have
five
different
interfaces
of
what
is
a
function.
How
do
I
deploy
it?
How
do
I
get
metrics
or
logs
out
of
there,
so
it
basically
becomes
a
mess.
H
That
can
depend
I
mean
I
can
do
like
life,
demos
and
and
talking
details
about
all
these
products.
Then
I
can
stretch
it
out,
but
I
guess
at
minimum
I
need
I
could
also
do
like
a
lightning
talk
or
talk
your
really
quickly.
Here's
what
they're
doing
a
few
examples
and
then
talk
about
the
problems,
I
see
and
then
I'm
done
in
like
10
minutes
or
so.
Okay,.
B
B
What
I
was
wondering
was
whether,
aside
from
the
session
of
throw
a
summit,
is
whether
you
could
condense
it
down
into
like
a
very
brief
5-minute,
but
you
said
lightning
talk
for
the
for
the
service
working
group
as
a
way
to
jumpstart
the
conversation
with
the
community.
This
to
say:
do
you
guys
agree
with
Christophe
that
these
are
some
areas
where
we
should
look
at
possible
standardization?
And
then
you
could,
you
know,
use
the
five
minutes
to
talk
very
quickly
about
the
use
cases
or
the
scenarios
that
you've
run
into
does
any
that
make
sense.
B
Dad
does
make
sense.
Yes,
that's
a
good
I
do
think
it's
a
very
interesting
topic
and
I
wouldn't
want
you
to
to
not
submit
a
service
summit.
You
think
it
sound
like
a
work
when
it's
worthy
enough
enough
to
get
its
own
timeslot.
Pc
buddy,
but
I
do
I,
do
want
to
use
the
information
in
our
birds
of
a
feather
session
or
with
the
interaction
with
the
community.
So
I
don't
want
to
be
both.
B
B
In
terms
of
next
steps,
though,
I
know
Bern,
you're
gonna,
look
at
making
a
PR
and
actually
Christophe
you
mentioned
possibly
making
a
PR
to
the
white
play
paper
as
well.
Gonna
be
very
welcome,
but
in
terms
of
next
steps,
do
you
guys
want
me
to
schedule
another
call
now,
for
you
know
well
known
time
or
do
you
want
to
wait
until
we
get
closer,
then
my
panic
sets
in
we'll
set
up
a
call.
How
do
you
guys
want
to
work
this
I.