►
From YouTube: OpenShift Commons Briefing #122: State of FaaS on Kubernetes - Michael Hausenblas (Red Hat)
Description
FaaS (Function-as-a-Service) or serverless as some call it is a promising compute paradigm suitable for event-driven scenarios. In this briefing, Red Hat's Michael Hausenblas and Brian Gracely review the current open source offerings for FaaS on Kubernetes (Apache Open Whisk, kubeless, OpenFaaS, etc.) and address pros/cons both on an architectural level as well as from a UX point of view. They will also discuss the topic FaaS vs. containers from a developers as well as an operators perspective.
A
A
You
have
like
five
minutes
and
maybe
five
slides
to
do
it,
but
it's
a
topic
that
deserves
a
little
bit
more
time
and
I
thought
that
as
a
prequel,
we
could
get
Michael
and
Bryan
to
sort
of
scientific,
eight
on
the
state
of
fahza
function
as
a
service
on
kubernetes
and
flush
out.
Some
of
the
salient
points
for
his
lightning
box.
So
without
any
further
ado,
I'm
going
to
let
Michael
and
Brian
don't
take
this
away
and
introduce
themselves
we'll
have
Q&A
in
chat
and
Q&A
at
the
end.
So
please.
B
Okay,
so
hello,
and
thanks
for
having
me
that's
a
great
idea,
I
think
spending
a
little
quality
time
here
and
interactive
fashion
really
helps
the
whole
topic.
My
name
is
Michael
hausenblas
I'm,
a
developer
advocate
in
the
Obi
15
and
been
doing
this
functions
of
service
and
service
things
roughly
for
two
years,
a
little
bit
more
than
two
years
now.
C
B
So,
let's
try
jump
right
into
it.
If
you
look
at
what
the
whole
thing
comes
from,
then
people
would
argue
we
had.
You
know
traditional
paths
and
services,
service-oriented
architecture
and
so
on.
For
you
know,
15
20
years
already,
I'm
not
gonna,
go
up
and
defend
things
like
Korver
and
other
things,
but
it
was
around
right.
B
This
idea
of
botanist
services
communicating
in
a
kind
of
loosely
coupled
manner
and
then
some
10
years
ago,
or
so
the
whole
thing
moved
to
it's
smaller
services
that
have
a
certain
purpose
and
they're
even
more
loosely
coupled
and
they
can't
be
independently
scaled
and
to
be
clear,
backed
up
by
their
own
data,
store
and
automation,
definitely
played
a
big
role
and
as
a
kind
of
natural
next
step.
If
you
break
down
this
function
to
even
further,
you
arrive
at
functions
and
functions,
essentially
the
main
characteristics
there.
B
It's
a
it's
stateless,
so
you
don't
have
any
kind
of
real
state
in
the
function.
There's
some
scripts
credit
race,
typically
a
few
megabytes,
whatever
Iowa
contemporary
store
or
something,
but
typically
they
are
totally
stateless
and
they're
aventurine.
So
it
really
makes
most
sense
when
you
have
some
kind
of
events
triggering
that
function.
Execution
typically
we're
talking
about
ephemeral
and
elastic
elasticity
a
lot,
and
that
also
depends
quite
a
lot
on
execution
environment
and
we
can
make
it
to
that
later
in
terms
of
public
versus
on-premises
deployment.
B
It's
on
a
conceptual
level.
You
have
essentially
three
parts
to
any
kind
of
functions
to
service.
You
have
two
triggers
and
the
trigger
can
be
something
very
concrete,
for
example,
uploading
a
file
to
an
s3
bucket,
for
example,
and
that
event
triggers
then
or
you
could
use
this
event
triggering
than
a
certain
functional
execution.
You
could
also
think
in
a
more
an
abstract
way,
for
example,
time
right.
If
you
have
every,
let's
say
our
you
say:
I
would
execute
a
function.
That's
also
a
trigger,
and
some
of
us
remember
good
old,
crunch
ops
right.
B
That's
the
century
also
trigger
you
have
then
this
Corbett
titled
interest
management,
which
typically
includes
a
CLI,
a
HTTP
or
whatever
API
UI.
In
that
a
developer
can
upload
their
functions,
manage
and
define
the
triggers
and
another
things,
logging,
monitoring
and
so
on,
and
then,
which
is
sometimes
a
little
bit
overlooked.
C
Gonna
ask
a
sort
of
a
dumb
question,
simple
question,
so
you've
used
the
the
name.
Functions
is
a
service
a
couple
times
sort
of
fast
right
and-
and
you
know
at
the
beginning,
we
talked
about
serverless
like
are
those
just
interchangeable,
or
is
there
a
point
when,
when
it
shifts
from
being
one
to
the
other
great
question,.
B
Good
question
I
was
originally
when
I
started
off.
I
was
like
a
server
list.
That's
so
dumb.
It
reminded
me
of
no
sequel
right.
It's
like
defining
something
by
what
it
is
not
or
what
it
doesn't
have,
and
I
actually
blogged
about
it
and
select
come
on.
Let's
just
call
it
functions
of
service
over
time.
I've
changed
my
view
in
it,
because
a
few
people
pointed
out
that
there
is
more
that
there
are.
B
There
are
potentially
more
surveillance
offerings
than
purely
function
as
a
service,
so
anything
that
essentially
allows
you
to
not
focus
on
provisioning
and
by
that
I
also
mean
dealing
with
containers
really
on
it
on
a
very
high
level.
So
anything
they
are
not
only
functions.
The
service
could
be
called
server,
that's
a
few
or
for
now,
I
would
understand
service
as
being
like
the
bigger
concept
and
functions
and
service
being
one
particular
manifestation
of
service.
If
that
makes
sense,
okay.
C
B
Yeah,
that's
that's.
Essentially
the
concept
triggers
some
types
of
management,
and
there
is
where
you
see
quite
a
lot
of
differences
between
the
different
frameworks
or
solutions
that
you
have
in
terms
of
what
kind
of
language
is
to
do
support.
What
attitude
do
they
bring?
You
know,
do
they
use?
For
example,
most
of
them
do
use,
containers
or
not.
How
powerful
is
the
CLI?
Is
there
a
UI
or
not,
and
so
on
and
so
forth
and
integration?
B
B
Although
the
whole
thing
is,
you
know,
maybe
three
years
old,
AWS
lamda
started
in
2014,
so
three
three
and
a
half
whatever
years
old,
you
still,
you
know
you
can
already
see
that
there
are
quite
some
offerings,
so
it's
early
days,
but
still
there
are
offerings
on
each
of
these
layers.
The
topic
of
our
discussion
today,
however,
is
committed
and
how
these
functions,
our
service
frameworks
or
offerings,
can
be
used
on
top
of
committees.
I
obviously
have
a
special
interest
in
the
enterprise
distribution.
B
Opposite,
that's
what
I'm
working
on,
but
in
general,
all
of
those
run
in
little
communities,
and
you
see,
if
you
we
start
from
a
very
right
column
there,
that
most
of
them
are
not
really
that
all
right
there
are
some
of
them
that
have,
you
know,
started
out
like
open,
Wisc,
initially
by
IBM
heart
of
bluemix
right
in
2015,
but
many
many
of
them.
You
know
they
are
maybe
half
a
year
three
year
old,
late,
2017,
mid
2017
and
some
of
them.
B
If
you
look
at
function,
I
intentionally
put
them
into
italic,
because
we
essentially
at
some
point
in
time
refocused
when
I
say
we
I
mean
redhead
from
doing
function
and
putting
our
engineering
focus
and
and
muscles
behind
Apache
of
risk,
and
two
things
are
notable
here.
We
can
discuss
about
the
community
science
right,
I'm,
I'm,
pretty
sure
there
will
be
someone
out
there
who
says
like.
Oh,
but
you
know
here,
you
got
that
totally
wrong.
This
is
way
bigger
or
a
small
or
whatever.
This
is
in
general.
B
If
you
look
at
overall
activity,
contributors
use
cases
around
that,
what's
going
on
on
mailing
lists
outside
of
the
core
backers,
offer
certain
solution
and
what
what
stands
out
here
is
really
Apache
of
risk
right.
It's
really
backed
by
a
number.
Obviously,
it's
a
petra
Software
Foundation
project
now
nowadays,
but
they're.
Quite
some
habits
behind
that
and
I'm,
not
saying
that
you
necessarily
like
everyone
has
to
go
all-in
with
a
whisk.
There
are
really
good
reasons
why
you
want
to
use
cublas
or
yes
or
dad.
C
I
think
it's
a
good
yeah,
it's
a
good
summary
and
again
you
know
people
can
can
debate
any
any
sort
of
qualitative
measure
of
it.
I
I
think
that
the
takeaway
from
you
know
the
to
me
that
a
couple
of
big
takeaways
from
this
are.
You
know
it's
always
good
to
see
a
number
of
projects,
because
it
shows
that
there's
there's
a
lot
of
interest
in
this
space.
We're
gonna
see
some
some
very
interesting
things
come
out
of
it.
You
know
we
probably
could
add
a
column
here.
That
is,
you
know
it.
C
Is
there
a
funk?
Is
there
a
focus
of
each
of
these
projects
and
I?
Think
in
some
of
them
you
know
you,
might
you
might
see
more
distinct
focus
in
others?
You
know
subjectively.
You
could
sort
of
say
well
that
you
know
they're
sort
of
the
same
thing
so
I
the
other
big
takeaway
for
me,
so
so
good
to
see
a
lot
of
interest
in
it.
Obviously,
there's
people
trying
to
figure
it
out.
You
know
the
other
one
and
you've
been
working
on
this.
C
A
little
bit
is
you
know,
while
in
the
top
column,
you
know,
you've
got
Red
Hat
who's
going
to
be
backing
open,
whisk
as
sort
of
the
de-facto
one
that'll
be
an
open
shift.
I
mean
we've
already
seen
people
you
know
making
other
other
frameworks
work
on
top
of
this.
Alright,
so
you
know
we've
seen
cube
list.
You
know
you've
shown
that
we've
seen
open
faz
we've
had
blogs
about
that.
So
I
think
the
nice
thing
about
this
is
if,
as
a
group
within
a
company,
you
decide.
Well,
we
really
like
one
of
them.
C
That's
fine.
You
know
pick
that
and
and
go
with
it
if
you
like
something
else
great
and
they're
and
they'll,
be
reasons
why
you
do
it
you,
like
a
back-end
data
integration,
you
just
like
the
languages,
they
support
you,
you
like
the
community
whatever
it
might
be.
So
to
me,
that's
a
it's
a
good
sign
at
this
point.
We
may
see
this
consolidate
over
the
years,
but
you
know
good
healthy
community.
At
this
point.
A
If
you
go
back
a
slide
to
the
landscape
slide,
what
we
interesting
for
me
is
another
angle
on
this
is
of
all
these.
The
proprietary
offerings,
which
ones
of
the
open
source,
are
under
the
underpinning
some
of
these
proprietary
offerings,
like
so
who's
using
what,
under
the
hood
to
add,
add
a
flavor
to
the
thing
is:
where
are
there
and
everything
enterprises
the
open
source
screen
here,
but
would
but
I'm
I'm
very
curious
with
what
are
these
enterprise
offerings
are
using
the
different
open
source
projects
as
well
right.
B
Time
I'm
only
aware
of
one
or
two
examples
here
at
the
end
cloud
and-
and
there
will
be
obviously
a
risk-
I
am
not
sure
if
any
of
the
other
commercial
offerings
is
actually
using
anything
or
you
know
something
on
the
next
slide
in
terms
of
of
fare.
Some
communities
I
would
love
to
learn
about
it,
but
that's
the
only
literally
only
example
that
I
am
aware
of
and
I'm
very
happy.
If
anyone
is
on
the
call
educating
me
here,
I'm,
very,
very
happy
and
to
learn
about
that.
I
don't
know.
C
B
B
Cool,
so
we,
the
takeaway
from
this
is
there
is
there's
a
lot
and
maybe
that's
a
horrible
comparison,
but
it's
a
bit
like
if
I
look
at
a
operating
system,
like
you
know,
rel
or
whatever
there
are.
You
know
different
I,
don't
know
editors
or
whatever
you
might
prefer
VI,
you
might
be
for
Emacs
or
whatever,
and
there
might
even
be
a
default.
B
B
Unfortunately,
don't
really
have
time
to
go
through
all
of
these
different
options
in
greater
detail,
and
it's
really
a
matter
of
quickly
go
back,
because
you
asked
it
earlier
or
pointed
at
that
direction
in
terms
of
focus
or
whatever.
There
are
some
that
focused
more
on
usability.
Let's
put
it
that
way,
which
are
more
developer-friendly
to
Blair's.
Openness
comes
to
mind.
There
are
others
that
target
more
like
the
the
enterprise
space
that
more.
They
have
more
in
mind
this.
B
Okay,
how
are
people
gonna
use
that
in
an
enterprise
set
up,
but
my
recommendation
here
would
be
that
you
start
off
by
the
languages
you
want
to
see
supported
there
and
each
of
them
has.
There
are
some
that
are
kind
of
usual
suspects
like
no
chairs,
for
example,
pretty
much
everyone
supports
that
and
start
there
and
and
have
a
look
at,
for
example,
if
you
are
in
in
a
certain
environment-
and
you
work
closely
together
with
one
of
the
the
providers
there,
you
might
have,
you
might
start
with
every
lighting
stuff.
B
B
Let's
move
a
little
bit
further
and
talk
a
little
bit
about
challenges
and
opportunities
that
I
really
intentionally
picked.
This
phrasing
to
say,
I'm,
a
huge
believer
in
functional
service,
I
I
think
that
it
is
a
herb
tool
in
the
toolbox,
but
it
is
one
tool
and
not
a
silver
bullet.
So
the
first
question
that
I
hear
use
it
very
often
I'm
having
these
discussions
with
books
out
there
is:
does
it
actually
make
sense
to
use
anything
other
than
you
know?
B
But
that's
just
one
aspect
there.
So
you
know.
Sometimes
you
see
that
and
hear
people
arguing
it's
containers
on
one
hand
and
surveillance
or
function
service,
on
the
other
hand,
and
I
would
say
that
is
a
little
bit
limiting.
That
is
it's.
It
doesn't
really
help
anyone
to
to
paint
a
picture
like
that,
because
at
the
end
of
the
day,
most,
if
not
all,
of
the
functions
of
service
offerings
communities.
B
Remember
your
table
earlier
on.
Do
use
containers.
You
know
to
package
and
ship
your
functions,
so
maybe
it's
just
a
matter
of
the
focal
shift
see
the
containers
are
less
visible
to
developers.
There
are
some
that,
where
you
can
explicitly,
you
know
use
containers
as
well,
but
typically
you
you
don't
really
get
to
see
them.
So
it's
essentially
not
only
surveillance,
but
you
know
these
these
bits
or
this
infrastructure
parts
are
not
miscible
to
the
end
to
developer
gradient
and
don't
play
any
role
there.
B
At
the
end
of
the
day,
that
means
I
as
a
developer.
Don't
need
to
worry
about.
Oh,
is
it
this
image
for
that
scanning?
Anything
security-related,
really
I
just
say
this
is
my
code.
Go
will
execute
my
code.
There
are
a
couple
of
examples
where
the
public
cloud
offerings
are
sometimes
too
restrictive.
Just
to
give
you
an
example,
the
default
that
I
checked
earlier
in
a
mm
that
is
still
300
seconds.
So
that's
five
minutes,
that's
not
bad,
but
if
you
have
anything
longer
running,
then
you
split
it
up
and
you
need
to
coordinate.
B
You
need
to.
You
know,
put
intermediate
results
somewhere
else,
and
you
need
to
do
it
on
your
own.
So
I'm,
not
I'm,
not
saying
either
is
lambda.
It's
bad.
All
I'm
saying
is
if
you're
running
that
on
your
own,
when
it
is
cluster
that
you
might
already
have
in
place
anyways,
then
you
get
to
choose,
you
can
say:
okay,
I
need
I,
don't
know
five
seconds
or
maybe
10,000
seconds
whatever
you
need
for
your
use
case,
and
that
is
just
one
example.
B
There
are
many
many
others
where,
because
you're
in
control
or
someone
in
your
organization
is
in
control,
you
have
the
choice
to
make
it
work
for
your
use
case
and
often
we
don't
bend
over
backwards
to
make
it
work.
The
other
aspect
there
is
elasticity,
and
that
is
something
where
I
would
say.
Yes,
it's
true
that
in
the
case
of
an
on-premises
deployment,
so
if
you
have
your
data
center
and
you
have
ten
racks
there
well,
you
know
you
can't
rack
up
for
something
there
within
a
few
seconds.
B
That's
where
public
cloud
deployments
shine
and
no
one
is
or
it's
not
the
case
that
you
cannot
positive
way.
You
can,
of
course,
deploy
qualities
and
in
the
public
cloud
and
benefit
from
from
auto
scaling
there
as
well,
so
all
in
all,
no
matter
if
your
on-premises
or
in
the
public
cloud
using
these
tests
on
communities
service
deployment
options
early
on
mentioned,
gives
you
portability
and
freedom
to
do
what
what
you
need
to
do
for
your
use
case.
C
Yeah
I
was
gonna,
we're
gonna
have
one,
maybe
like
one
one
other
one.
Other
sort
of
aspect
of
this
I
think
to
me.
I
always
look
at
this
as
there's
there's
two
other
elements
that
are
kind
of
worth.
Considering
the
first
one
is,
and
this
doesn't
have
to
be
AWS
specific,
but
it's
really
sort
of
any
cloud
specific
is
you
know
which
which
data
services
do
you
want
to
be
specifically
tied
to?
Or
do
you
necessarily
want
to
immediately
leverage-
and
you
know
there
this
is.
C
This
is
where
there
is
some
significant
differences
between
even
what
say,
AWS
does
versus
as
your
functions
versus
Google
functions,
but
also
you
know
if
you
were
using
Open
whisk
or
you
were
using
cube
lists.
You're
gonna
have
different
sort
of
native
connectors
that
are
going
to
be
available
to
you.
That
may
very
well
come
into
play
as
to
what
you
what's
important
to
you.
C
We've
heard
we've
heard
different
companies
who
will
tell
us
you
know
I
want
to
use
serverless,
but
maybe
I,
don't
necessarily
want
to
be
tied
just
to
one
of
them.
You
know.
Is
there
going
to
be
a
way
to
to
run
the
same
service
same
operations
in
an
azure?
So
so
that's
one
of
them,
I
think
the
other
one
I
think
that
comes
into
play,
and
you
mentioned
like
pricing
and
so
forth.
C
I
think
you
need
to
consider
this
is
this
is
always
the
conversation
that
comes
up
of
like
do
you
deal
with
containers
or
you
deal
with
serverless
I
think
it
really
comes
down
to
who's
who's
going
to
to
manage
this
sort
of
development
of
the
functions
and
then
who's
going
to
manage
the
underlying
platform.
You
know
if
you
have
no
capabilities
or
no
interest
in
managing
the
underlying
platform.
You
know
using
one
of
the
managed
clouds
is
probably
the
way
to
go
if
you're
you're
capable
of
running
a
kubernetes
environment.
C
At
that
point,
you
you
more
or
less
have
the
skills
necessary
to
be
able
to
spin
up
immediate
resources,
scale,
resources
and
then
you've
got
some
more
flexibility.
So
you
know
definitely
more
than
just
saying
bit
of
lock-in.
I
think
it
comes
into
you
know:
what
do
you
wanna?
What
do
you
want
to
take
into
consideration
your
data?
How
well
you
know
how
you
operate
it
and
then
how
portable
you
want
the
system
to
be
not
just
an
individual
function.
Functions
func,
is
pretty
portable
platform.
Has
some
other
considerations
great.
B
Great
point:
that's
that's
exactly
what
I
meant
I
might
not
have
used
the
best
words
for
it,
but
what
the
thing
that
comes
to
mind
is
really
trade
offs
right.
You
are
either
on
the
one
extreme
willing
to
say
well,
I'm
fine
with
with
you
know,
feature
or
vendor
whatever
login,
and
for
that
I
get.
You
know
everything
managed
it's
very
convenient.
I,
don't
need
to
take
care
of
anything
and
I
know
that
it
will
be
hard
or
difficult
to
then
move
somewhere
else.
B
The
other
hand,
if
you
are
you
know
into
that,
where
you
say
okay
to
me,
portability
is
importance
and
I
want
to
have
certain
level
of
decoupling
in
terms
of
data
stores
or
whatever.
Then,
then,
you
know
this
to
see
the
actual
trade-off
there
too,
at
some
point
I'm.
You
have
to
decide
and
yeah
great
great
point
now
moving
on
to
a
it
can
call
it
controversial
topic,
but
to
me
it's
so
central
and
I
was
surprised
when
I
learned
yeah
some
some
reactions
to
that
topic.
B
In
your
data
center,
the
people
who
run
the
actual
platform,
it
is
amiss,
for
example,
it
typically
don't
really
as
a
developer,
have
access
to
them
in
AWS
case.
It's
it's
very
obvious
right.
You
can't,
you
simply
cannot
reach
out
to
admin
there
and
say
hey.
You
know,
I
have
that
problem,
and
and
even
if
it's
within
the
any
organization.
A
B
B
You
simply
cannot
assume
that
someone
in
the
role
will
be
around,
and
that
means
that
at
the
end
of
the
day,
you
have
to
ask
yourself
if
something
goes
wrong
in
that
function.
If
you
know
in
terms
of
logging,
monitoring
or
whatever,
who
will
carry
the
patron
since
or
if
you,
if
you
subscribe
to
that
idea
that
you
do
not
have
access
to
ops
folks
that
take
care
of
stuff,
because
at
the
end
of
the
day
there
will
be
service
somewhere
who's
on
call
right.
B
Who
will
be
the
one
who
looks
after
a
function
or
a
set
of
function,
typically
working
together
and
there
my
conclusion
was
essentially
well,
it
probably
will
be
the
developer.
So
if
you're
developing
a
function
and
uploading
it
to
one
of
the
frameworks
here,
then
at
the
end
of
the
day,
probably
you
who
will
be
on
comm,
you
will
be
responsible
for
that
and
I
ask
the
question,
and
there
are
some
out
there
who
say
yeah.
Of
course,
that's
that's
our
attitude.
B
C
Well,
it's
I
I
think
it
feels
like
it
moves
the
conversation
a
little
bit
so
so
in
the
past.
If,
let's
say
you
were
a
developer
in
your
application,
interacting
with
the
database,
let's
keep
it
simple
right.
You,
you
probably
had
some
organizational
way
of
saying.
Okay,
if,
like
let's
say,
the
application
doesn't
work
from
getting
certain
error
messages.
That's
a
that's
an
application
problem,
but
if
the
response
time
of
the
application
is
slow,
you
know
when
do
I
engage,
say
a
DBA,
for
example,
who
wouldn't
necessarily
be
a
developer,
would
be
sort
of
considered.
C
I
think
this
is.
This
is
in
the
same
vein
of
things,
because
if
we
take
a
take
a
serverless
application,
let's
let's
take
up,
and
let's
take
your
example
where
something
comes
in
through
an
api
gateway,
a
picture.
For
example,
a
file
gets
stored
in
s3,
a
function,
changes
that
into
a
thumbnail,
and
then
it
goes
out
to
a
web
page
to
be
displayed
right,
simple
concept.
You've
got
you've.
Really
you
really
have
the
same
idea
there
right.
C
So
if
the,
if
the
picture
didn't
get
transformed-
well,
that's
probably
an
application
problem,
but
if
an
API
call
didn't
happen,
you
get
a
404
who's.
Who
do
you
talk
to
at
that
point.
Right
is
that
is
that
an
OPS
problem,
because
it's
an
api
gateway
thing
the
developer
know
about
the
though
I
think
that's
really
what
it
is
is
you
you?
You
want
to
take
probably
similar
constructs
as
you
had
before
right.
Where
does
the
application
stop
and
start
and
then
beyond
that?
C
What
are
you
willing
to
do
and
it
that
all
might
be
the
same
person
or
you
might
have
sort
of
a
set
of
operations,
people
that
are
now
dealing
with
all
those
sort
of
like
you
know,
inputs
and
outputs,
right
sources
and
sinks?
If
you
will
but
yeah
I,
don't
think,
there's
clear
I,
don't
think,
there's
an
answer.
I
think
it
depends
on
what
your
your
organization
wants
to
be
able
to
do,
but
I,
don't
it
doesn't
it
doesn't
go
away?
That's
think,
that's
the
key.
Is
it
doesn't
go
away?
Yeah,
yes,
I!
Think.
A
Also
changes
the
nature
of
the
developer
role
as
well,
because
a
lot
of
times
developers
are
brought
in
just
to
do
the
code
and
not
to
maintain
the
application
for
the
life
of
it.
So
it
maintain
maintaining
the
relationship
with
the
developer,
for
the
life
of
the
application
becomes
something
that
you're
on
the
hook
for
as
a
northern
state.
B
Yeah
I
guess
my
initial
reaction,
especially
back
then,
whenever
the
service
table
spoke
there
was
that,
like
I
kinda
know
who
did
it
and
where
it
came
from,
but
this
term
no
ops
that
really
rocked
me
off
the
wrong
way.
It
was
kind
of
like
suggesting.
Oh,
there
is
no
offset
that's
pecked.
Of
course
there
is
it's
just
you
know
it's
even
further
like
it
is
already
there
with
containers.
You
have.
B
You
know
certain
aspects
in
terms
of
logging
and
metrics
think
about
3/3
and
other
things,
so
that
awareness
is
already
there
and
the
more
developers
care
about
that
and
are
aware
of
that.
The
easier
it
is
for
deaf
ops
into
players
and
with
service
with
function
serves.
This
was
even
further
is
more
extreme,
so
calling
the
whole
thing
no
arts
to
me
is
really
you
know
suggests.
That's
probably
the
main
motivation,
yeah.
C
It
is
what
do
you
do
if
the
thing
you're
dealing
with
is
a
to
pizza
API,
like
what's
your
contract,
with
an
API
from
from
an
office
perspective,
because
that's
ultimately
what
you're
gonna
be
dealing
with
right
and
if
you-
and
if
you
don't
know
the
answer
to
that
or
you
haven't,
explored
it,
then
then
you're
gonna
have
then
you're
gonna
have
a
challenge,
but
but
that's
what
you're
dealing
with
is
I'm
now
dealing
with
an
API
and
I.
Don't
know
what
the
team
is
behind.
B
B
No
ops,
saying,
oh,
we
don't
have
any
obvious
anymore,
like
someone
somewhere
will
have
to
take
care
of
that
logging
monitoring,
optimizing
runtime,
whatever
troubleshooting
our
profile
cook
last
at
least
my
favorite
picture
and
that's
I-
think
a
wild
public
mean
going
on
up
and
down
on
all
the
social
media.
The
you
know,
half
half
drawn
horse
there.
So
if
you
have
a
handful
of
function
and
many
use
cases
actually
started
like
that,
if
you
look
through,
you
know
the
the
success
stories,
they
don't
slander
or
anywhere
else.
B
There
are
countless
service
conferences,
chef,
Khan
and
many
many
others
in
many
cities.
Nowadays,
it's
a
handful
of
function
right,
it's
maybe
three,
maybe
five,
maybe
ten!
It's
in
general,
not
a
problem.
B
B
So
that's
that
some
somehow
familiar
to
the
problem
we
had
with
containers
right
as
long
as
you
did,
you
know,
running
a
container
and
on
your
laptop
you
don't
really
the
container
of
the
street
as
soon
as
you
start
spreading
out
hundreds
of
containers
over
hundreds
of
nodes
or
whatever
you'd,
probably
want
something
in
Bacchus,
treats
the
bits
and
as
far
as
I
have
seen
always
in
the
open-source
context.
This
is
a
rather
underdeveloped
space.
So
far,
the
only
thing
I'm
aware
of
and
again
I'm
very,
very
open
to
it.
B
If
someone
can
point
out
some
other
examples
is
IBM's
composer
that
tries
to
achieve
that
and
I
think
it's
even
worse.
It's
actually
not
even
in
a
position
yet
that
we
can
actually
point
out
like
what
is
such
an
orchestra
they're
supposed
to
do.
Is
it
things
like
dependency
management
between
functions
as
Kaeding
or
functions,
which
could
be
clean?
Public
cloud
environments
are
not
really
encouraged
or
a
lot.
Is
it
how
functions
communicate?
B
You
have
something
like
Kafka
or
whatever
they're
for
interplanetary
communication,
so
it
is
really
early
days
in
terms
of
orchestration,
but
at
some
point
in
time
it's
that
whole
thing
and
it
will
take
off.
You
will
end
up
with
a
lot
of
functions
and
even
and
you
might
be
laughing
about
it,
but
even
thinking
about
having
a
kind
of
registry
of
available
functions.
If
you
have
a
certain
base
function,
that
you'd
say
well,
why
should
I
over
and
over
implement
the
same
functionality?
B
I
might
put
it
there,
and
people
can
reuse
that
function,
who's
gonna,
do
that.
Is
it
your
spreadsheet
or
maybe
you
have
something
else
there?
Maybe
a
registry
or
whatever
there
for
functions.
So
wrapping
up
this
one
here.
It
might
not
yet
be
a
problem
because
we're
dealing
with
not
that
many
functions
so
far,
but
at
some
point
in
time
we
will
probably
very
likely
meet
some
orchestration
bits
here
as
well.
B
One
example,
as
I
said,
I'm
aware
of
this
a
composer
thing
from
IBM
a
couple
of
resources
if
you're
interested
in
any
of
these
things,
I
want
to
dive
deeper
if
you
want
to
be
active,
come
and
join
us
in
CNC
at
the
cloud
native
compute
foundation,
service
working
group
on
Schlag,
github
meetings
and
so
on.
I
want
to
also
point
out
my
thanks
for
Chris
when
to
maintain
this.
B
This
yeah
fasten,
qualities,
inventory
and
github
is
documenting
stuff
there
and
William,
who
gave
quite
a
lot
of
rough
input
regarding
you
with
bits
and
various
people
ones
like
and
in
person,
and
so
on,
who
helped
shape
these
these
topics
and
these
comments
there?
Okay,
that's
I,
think
ready
it.
What
I've
prepared
here
yeah
so
just
leave
it
here
and
I.
Guess
we're
good
to
you
know,
keep
the
conversation
going
in
potentially
have
questions.
A
C
So
I'll
throw
out
I'll,
throw
out
one
one
one
comment
and
then
I'll
throw
out
a
question
and
then
anybody
can
control
I
I
know
we
talked
a
lot
of
here
about
sort
of
developers,
building
functions,
so
application
specific
things.
I
think
there
is.
There
is
a
whole
set
of
people
that
are
using
this
for
operations
too.
We've
we've
heard
a
lot
of
people
that
are
looking
at
repetitive
tasks
that
they
have
that
they
do
for
operational
things,
whether
it's
inventory
management
or
things
for
upgrades
or
whatever
and
they'll
take
up
they'll.
C
B
Right
absolutely
and
I
love
it.
True.
I
heard
a
story
that
I
literally
heard
it
I
didn't
read
it.
So
anyone
who
has
some
relations
with
a
double
yes
lambda
can
confirm
that
or
tell
me
that
that's
not
true,
it
would
be
interesting
to
see,
but
I
heard
the
story
that
AWS
essentially
was
looking
at
patterns.
B
How
people
would
use
ec2
instances
in
one
pattern
that
kind
of
emerged
over
time
which
apparently
then
led
to
the
introduction
of
lambda
was
that
many
especially
ops,
really
a
task
or
essentially
ramping
up
an
ec2
instance,
then
doing
something
for
a
minute
or
two
and
inferring
it
away
again,
and
they
looked
at
that.
That's
the
rumor
here
like
how
can
we
make
that
pattern
better
and
more
efficient?
Also
for
for
them
in
terms
of
utilization?
B
B
B
They
have
pots
already
running
or
containers
in
these
parts
really
running
on
standby,
so
they're
kind
of
like
hot
they're
already
running
and
when
the
function,
when
they're
the
trigger
actually
triggers
the
the
function
execution,
they
don't
need
to
be
spun
up
anymore,
so
how
they
do
that
differs,
but
in
general.
Yes,
all
of
them
essentially
use
the
ones.
We
have
talked
about.
First
qualities
use
containers
there.
A
So
I'm
just
gonna
ask
the
question:
I'm
kind
of
curious
about
the
monitoring
of
service
applications
ads
in
me.
We're
we're
all
getting
used
to
monitoring,
pods
and
containers
and
all
of
that.
But
this
is
a
function
kind
of
boggles
of
minds
how
you
keep
track
of
all
the
moving
parts
and
and
what
sort
of
monitoring
fast.
B
Right,
so
we
have
already
seen
that
this
big
transition
from
you
know
this.
Traditional
monitoring
of
hosts
I
was
saying
where
you
know
it's:
okay,
since
the
VM
or
whatever
lives
for
weeks
or
a
month,
it's
okay
to
kind
of,
like
think
of
monitoring
or
whatever
else,
and
obviously
with
containers
that
doesn't
work
anymore
because
of
my
lifetime
of
the
containers.
So
you
need
to
be
much
faster.
They
are
being
able
to.
You
know,
have
the
monitoring
and
logging
in
place
immediately,
and
that
is
obviously
also
true
and
even
more
true
for
functions.
B
The
other
aspects
and
actually
haven't
seen
much
there
either
is
that
of
the
street
racing.
In
this
context
this
has,
or
any
you
know,
became
become
sort
of
very,
very
popular
in
container
orchestration,
so
things
like,
for
example,
Jaeger.
What
any
really
any
open
tracing
based
system
is
becoming
kind
of
like
the
standard
people
need
that,
and
obviously
that
is
also
true
and
valid
for
for
functional
service.
So
you
want
to
know
that,
especially
if
you
have
a
couple
of
functions,
working
together
and
dependence,
is
there
dependency
graph?
Where
is
your
bottleneck?
B
And
for
that
you
need
at
the
end
of
the
day
again
it's
really
tracing,
so,
whereas
monitoring
and
logging
might
be
relevant
for
a
particular
owner
or
developer
for
a
single
function,
looking
at
an
entire
application,
it
is
made
up
of
ten
or
hundreds
of
functions,
I
think
a
distributed
tracing
tooling
around.
That
is
almost
the.
Must
it's
not
optional,.
A
B
It
very
very
much
depends
on
on
what
you're
what
you're
looking
at.
If
you
look
at
examples
here,
some
of
them,
you
know
they
were
not
developed,
how
to
put
it
nicely
with
with
qualities
in
mind
or
they
kind
of
like
evolved
in
parallel
and
where
then,
you
know,
optimized
or
made
comfortable
with
so
that
they
run
properly
on
communities.
Others
were
more
or
less
like
Cupid
secret
example.
They
were
really
born,
as
could
use
applications.
B
It
really
depends
that
the
main
thing-
and
this
is
really
more
probably
interesting
for
someone
who's
operating
that
and
probably
wants
to
extend
that.
The
main
thing
is
that
the
interface
for
the
developer
is
essentially
here
is
how
I
can
package
and
upload
my
function,
for
example,
or
my
functions
plus
resources
here.
Is
the
zip
file
put
that
to
that
HTTP
put
that
at
endpoint,
for
example,
I
have
a
nice
UI
and
whatever?
How?
How
do
I
define
the
trigger
is
how
easy
is
that?
How
powerful
is
that?
B
A
B
C
I
well,
I,
think
I
think
there's
a
couple
things
and
we're
not.
We
won't
sort
of
skip
it,
but
it's
not
a
not
a
simple
question.
I
think
you
know
you're
how
you
how
you
charge
people,
you
know
whether
it's
a
chargeback
model
or
you
know,
sort
of
known
pricing,
for
you
know,
amounts
of
time
or
whatever
the
the
the
guidance
I
always
give
to
people
is,
you
know,
go
go
look
at
what
the
public
cloud
does
because
in
essence,
your
end
customers
will
use
that
as
a
as
a
baseline
right.
So
they
will
say.
C
Ok,
the
cloud
gives
me
an
option
to
charge
things
by
the
monthly
charging
by
the
hour
by
whatever
you
know.
In
the
case
of
something
like
lambda,
they
tend
to
do
it
by
like
number
of
transactions
that
you
spin
up.
So
a
thousand
or
a
million,
or
something
use
that
as
a
starting
point,
you
don't
have
to
use
that
as
your
what
you
actually
do,
but
it
gives
you
some
sense
of
okay.
This
is
maybe
what
people
are
going
to
use
to
make.
C
Comparisons
of
you
know,
should
I
use
your
service
or
should
I
use
a
different
service.
The
other
thing
is
I.
Think
you're
gonna
find
that
the
more
you
can
you
can
chunk
it
up
into
into
things
that
people
will
typically
use.
Like
you
kind
of
have
to
understand
some
use
cases,
you
may
have
a
completely
different
charging
model.
If
you
say
we
will
run
your
application
and
then
you
can
use
something
like
functions
for
certain
operational
things
right
for
upgrades
and
monitoring.
C
A
B
Right,
I'm
not
aware
that
there
might
be,
you
know
somewhere
deeper,
hidden,
something.
The
thing
is
that
in
general,
the
attitude
there
is
a
little
bit.
It
was
like,
like
container,
you
should
not
necessarily
know
or
or
make
assumption
where
stuff
is
running
now.
There
might
be
a
good
reason,
which
especially,
is
true
for
data.
If
you
wanna
have
data
locality,
for
example,
you
you
know,
you
have
a
data
on
a
certain
node
again
in
the
case
of
communities.
B
It
certainly
can't
be
useful
and-
and
that
is
maybe
something
that
could
actually
be
part
of
the
orchestration
as
well
to
to
make
that
you
know
to
explicitly
state
dependencies
in
a
declarative
way
that
you
say
you
know
this
group
of
functions
or
these
two
functions
or
whatever
they
run
best
or
better
or
whatever.
If
they
are
close
by
and
or
with
with
data,
and
to
be
more
concrete
here,
for
example,
you
have
this
data
locality
essentially
built
into
all
these
big
data.
B
Things
like
you
know,
spark
and
clink
and
whatever,
where
the
basic
idea
is
that,
rather
than
shipping
data
across
in
the
cluster,
your
shipping,
the
code
to
where
the
data
is
so
it's
really
when
you
say
locality.
The
question
that
pops
to
mind
for
me
is
mainly
what
kind
of
locality
is
necessary.
Excuse
for
your
use.
A
B
Right
right,
so
the
main
point
being
that
you,
as
a
user
user
being
the
developer
here
in
this
case,
you
don't
need
to
take
care
of
that.
But
you
really
just
say
here
is
my
source
code:
no
jail
is
Python
or
whatever,
and
there
is
some
mechanism
to
package
that
and
give
that
to
the
respective
framework
and
what
happens
then
is
really
up
to
the
framework,
how
they
essentially
inject
that
they
build
an
imager
on
that
if
they
inject
their
life
and
so
on.
C
It's
it
and
it
sort
of
explains
why
you
know
if
you
took
this
at
a
really
high
level-
and
you
said
well,
you
know,
surveillance
on
kubernetes
kubernetes
is
containers.
I
should
be
able
to
run.
You
know
any
language
I
want
anything
I
want
to,
and
then
you
look
at
the
project's
themselves
and
they
only
support
certain
languages.
C
Part
of
the
reason
for
that
is
you
know
some
languages
wouldn't
make
sense
for
small
functions,
but
up,
but
in
other
cases
you
know
managing
the
behind
the
scenes
of
how
do
I
get
it
from
just
code
into
something
that
will
run.
That
has
to
be
a
function
or
you
know
a
capability
of
the
platform
right.
So
you
know
in
the
in
the
openshift
world
right
what
we've
done
with
containers
is.
C
We've
said
you
can
give
us
a
full
container
or
we
can
use
something
like
Sui
that
all
of
these
things
have
to
have
some
sort
of
like
a
Sui
type
of
function.
That's
going
to
get
you
from
your
code
into
into
some
operational
state.
So
like
again,
this
is
going
to
be
kind
of
project
dependent
which
languages
it
supports,
and
you
have
to
have
a
little
bit
of
awareness
of
that.
But
yeah
we'd
like
to
put
to
your
point
like
as
a
developer.
The
whole
goal
of
this
should
be.
B
Some
do
some
do
expose
that
some
allow
you
to
do
that,
but
in
general,
that's
exactly
to
go
right.
You
you're
not
dealing
with
any
kind
of
provisioning,
including
container
image
building
or
if
I
there
was
one
more
thought.
I
had,
maybe
that's
kind
of
like
at
least
to
my
site.
The
kind
of
final
thought
on
that
happening.
B
I
said
I
totally
believe
in
some
surveillance,
but
being
aware
of
it's,
not
a
silver
bullet
doesn't
solve
all
your
problems.
There
are
used
cases
where
you
probably
just
probably
not
a
really
good
fit,
and
in
that
sense
it's
really
one
tool
in
the
toolbox
and
that's
what
I
like
about
that
is.
You
know
UNIX
philosophy.
B
In
terms
of
toolbox,
you
have
different
tools
for
different
things
and
and
fast
being
one
tool
in
the
toolbox,
and
there
are
other
cases
where
you
have,
where
you're,
probably
better
off
with
with
permanently
running
container
and
then
paying
the
price
for
it.
Essentially,
you
have
to
deal
with
building
the
container
image
and
so
on
and
so
forth.
It's
one
tool
and
it's
a
very
useful
tool.
C
Think
we've
covered
a
lot.
You
know
I,
my
big
takeaway
is
you
know
this
is
this
is
an
exciting
space
in
terms
of
you
know,
we've
evolved
to
where
there's
communities
around
this,
so
people
have
some
choices,
which
is,
which
is
great,
gives
us
some
choice
of
project
but
also
say
choice
of
cloud
or
platform.
You
know
I
think
it
should
be
pretty
obvious.
This
is
still
pretty
early
days
right.
We,
you
know
there
isn't
necessarily
you
know
100
percent
standards
and
we
still
have
lots
of
projects,
so
you
know
be
cognizant
of
that.
C
C
But
if
not,
you
know
jump
in
one
of
the
projects
and
see
what's
going
on,
okay
Mike,
my
only
last
my
only
last
one
is
if,
if
at
some
point
somebody
in
one
of
your
meetings
or
one
of
your
developers,
says
we're
going
to
take
all
the
applications
we
have
today
and
we're
going
to
move
those
to
just
server
list,
because
we
don't
have
to
pay
for
anything
like
throw
up
the
biggest
red
flag.
You
you
possibly
can
I,
don't
say
that.
C
B
Just
to
clarify
or
to
phrase
it
in
a
more
positive
way
when
you
said
contributing
very
often
people
think
you
know
what
we
mean
by.
That
is
the
writing
code,
though
ever
not
at
all
right.
You
can
just
grab
one
of
these
things
pick
one
and
do
something
with
it.
Try
try
it
out
of
your
environment
and
and
give
it
back
to
a
community
blog
about
it
say:
hey
here's.
My
use
case
here
are
my
experiences
with
two
legs
and
that
helps
all
of
us
to
see.
B
You
know
that
case
is
better
to
see
how
could
such
an
Orchestrator?
What
would
be
the
responsibility
there,
so
contributing
really
does
not
mean
that
you
write
me
to
write
or
contribute
to
one
of
these
projects.
Of
course
we're
happy
if
you
do,
but
it
really
just
means
use
it
and
share
that
usage
experience
with
the
rest
of
the
community,
but
everyone
who
I
couldn't.
A
Have
said
it
better,
so
thank
you
for
that
Michael
and
collaborating
and
giving
feedback
doing
demo,
as
all
of
these
things
are
essential
to
moving
this,
these
projects
forward
and
all
of
our
projects
for
us.
So
thank
you
guys
for
having
this
conversation
with
us
today
and
yes,
someone's
asking
me
to
deep
dive
session
showing
how
it
all
works
on
humanities.
A
Under
the
hood,
though,
I
think
there
are
a
few
resources
around
a
demo
link
here
for
open
whisk
on
open
ship,
but
I
think
we
will
try
and
probably
post
read
that
summit
when
everybody's
breathing
again
set
up
an
open
whisk
on
an
open
shift.
Deep
dive
expose
some
of
the
underbelly
of
how
it
all
works
on
the
kubernetes.
So
thank
you
guys
again
for.