►
From YouTube: OpenFeature/Flagd K8S SIG, June 30, 2022
Description
Meeting notes: https://www.google.com/url?q=https://docs.google.com/document/d/1u_1qnS0PiYvZxwtsBgZrsX6DrRKfpG56q-Kh0-qn_pU/edit%23heading%3Dh.wh8mkiotns4b&sa=D&source=calendar&ust=1655043850692349&usg=AOvVaw3jgs22qY0Eh_ZvErxDUON9
OpenFeature website: https://open-feature.github.io/
A
C
B
B
Hold
on
my
I'm,
I'm
muted
try
again.
I.
C
B
Yeah,
just
for
the
until
saturday
and
we're
playing
back
so
todd
just
got
out
here,
though
so
we're
meeting
with
a
few
teams
and
yeah
I've
been
out
here
almost
two
weeks,
though
so
I'm
pretty
exhausted
and
ready
to
head
home.
B
Yeah
yep,
so
definitely
looking
forward
to
that
yeah.
B
Lars,
I
think
I
think
I've
seen
you
before
were
you
from.
Are
you
from
dev
cycle?
Am
I
getting
that
right?
Ebay,
ebay,
that's
right!
That's
what
it
is!
Yeah.
A
Right
yeah
but
yeah
so
first
meeting
for
me,
so
I
don't
know
what
to
expect
yeah,
I'm
I'm
I'm
a
java
developer.
Basically,
so
I'm
a
bacon
developer,
but
java
is
my
main
language
for
now
over
20
years,
so
hope
to
be
able
to
contribute
to
open
future.
A
B
Perfect
yeah
so
yeah
just
to
I
guess,
set
a
little
bit
of
context,
we'll
see
we'll
see
where
it
kind
of
conversation
goes
today,
but
this
one
this
meeting
typically
is
focused
on
just
like
the
the
kubernetes
and
the
flag
d
portions
of
it,
and
then
the
other
meeting
is
the
one.
That's
specific.
Like
the
broad.
You
know,
I
guess
open
feature
topics,
so
you
know
we
may
have
time
to
address
any
potential.
B
You
know
questions
that
you
have,
but
that's
typically,
you
know
what
we
talk
about
most
of
the
time
for
these
these
particular
meetings.
B
You're
gonna,
be
subscribed,
yeah
I'll,
my
battery's
about
to
die,
but
I'll
do
my
best
to
describe.
A
B
Can
also
you
could
also
just
copy
it
down
after,
if
you
want
to
use
it
yeah,
so
I
might
just
do
pen
and
paper.
So,
okay,
yeah,
I'm
not
even
sure.
If
we'll
get
that
far
daniel,
we'll
see
how
far
the
conversation
goes,
but
I'm
not
sure
how
long
we'll
go
for
so
we
don't
actually
have
an
agenda
so
I'll
just
rely
on
people
to
bring
things
up.
B
The
the
only
thing
I
can
think
of
starting,
I
guess
that
might
be
worth
starting
with,
is
there's
been
a
decent
amount
of
work
done
in
flag
d,
especially
since
the
last
meeting,
so
I
can
summarize
that
at
a
really
high
level-
and
that
will
segue
into
I
guess
the
one
thing
that
I
was
thinking
of
talking
about.
B
So
in
summary,
we
have
implemented
a
basic
json
evaluator,
so
that
the
flags
are
stored
in
a
specific
format
and
which
is
itself
maintained
in
a
schema's
repository
which
also
previously
contained
our
apis,
open
api
specification
for
flag
d
and,
in
addition
to
defining
static
flags
in
a
way.
That's
that's.
B
You
know
going
to
be
familiar
to
you
if
you're
familiar
with
with
pretty
much
any
feature
flagging
system-
and
it
doesn't
look,
it
looks
like
you'd,
expect,
there's
variance
and
a
default
variant
and
basically
an
enum
specified
if
the
flag
is
enabled
or
disabled.
We
only
made
it
minimum
because
we
wanted
to
support
additional
possible
options
later
without
having
to
break
the
contract
and
then
most
recently,
we
added
rules
which
are
defined
via
json
logic
and
into
that
schema.
B
So,
basically,
now
you
can
do
what
I
would
consider
to
be
like
a
basic,
a
really
basic
definition
of
a
flag,
including
dynamic
evaluation,
just
by
just
by
defining
a
proper
json
structure
and
last
week
at
the
at
the
bigger
sync,
I
demoed
that
end
to
end
with
the
operator
just
deploying
it
and
using
it
to
power
our
playground,
app,
which
you
guys
may
remember-
and
I
did
a
similar
demo
today
internally
at
dynatrace
that
also
included
a
rule-based
evaluation
which
again
only
got
merged
a
couple
days
ago,
so
that's
kind
of
where
we're
at
now.
B
One
thing
I
I
think,
I'd
like
to
see
happen
relatively
soon
is
grpc
service
implementation.
So,
right
now
we
have
http
as
a
service
implementation,
and
I
would
like
to
see
kind
of
a
similar
thing
in
grpc
just
to
reduce
all
the
overhead
associated
with
http.
B
I
mean
there's
there's
ways
we
can
improve.
Even
the
http
transport,
using
like
persistent
connections
and
stuff
like
that,
but
http
is
always
going
to
be
a
bit
slower
and
it
doesn't
support
bi-directional
so
eventually,
when
we
want
events
to
fire,
if
the
configuration
changes,
http
is
not
really
a
great
way
to
do
that.
So
I'd
like
to
see
that
done
eventually
and
then
one
other
actually
somewhat
related
thing
so
alex
initially
we
in
in
your
implementation,
there
was
a
socket
implementation.
B
We
we
kind
of
took
that
away
as
we
added
some
additional
interfaces
and
abstractions,
but
I'm
thinking
now
it
would
be.
It
would
be
no
problem
to
add
it
back
now
to
add
something
like
that.
If
we
wanted
to
so
I
don't
know
how
you
feel
about
having
a
socket
implementation.
Still
that
would
probably
just
you
know,
dump
all
the
flags
or
or
take
a
certain
interface
and
or
take
certain
parameters
to
like,
like.
B
Maybe
you
could
take
a
a
string
identifier
for
a
flag
and
just
send
back
that
flag
as
a
json
object.
I
don't
know
if
you
still
see
that
as
valuable
or
not,
but
that's
something
I
think
we
could
add
back
if
we
if
we
felt
like
it
if
he
thought
it
was
important,
I
think
the.
D
We
could
make
it
so
that
each
buffer
write
from
the
socket
returns
the
payload
body
of
what
the
json
object
would
be
from
the
http
restful
protocol.
So
you
can
kind
of
just
mirror
that
right,
because
it's
just
a
decapsulated
version
of
that.
So
I
think
it's
it
would
be
interesting
to
do.
D
But
the
greater
reason
was
that
you
could
use
the
same
socket
for
hp
as
well
right
you
could
use,
you
could
have
a
socket
client
and
you
could
then
do
http
transport
on
top
of
it,
which
would
just
mean
we
could
do
a
bit
of
reuse
inside
of
the
application.
D
I
don't,
I
don't
think
it's
worth
doing
unless
there's
actually
a
need
for
it
right
unless
we're
going
to
build
an
end-to-end
example
which
quite
could
be
cool,
I
do
think
we
should
have
a
conversation
about
the
examples
we
have,
because
we've
got
the
kind
of
hdb
sidecar
example.
But
if
we
wanted
to
showcase
flag
d,
is
a
multi-um
level
server,
multiple
service
levels?
I
do
think
that
that
could
be
an
idea
right.
D
You
say:
okay,
well,
we're
going
to
pick
either
socket
we're
going
to
pick
up
pick
our
pc
and
we're
going
to
design
an
end-to-end
demo
with
the
implementation
so
that
people
can
really
get
what
this
is
all
about,
and
also
now,
as
just
a
brief
update
the
I
did
a
lot
of
work
on
the
operator
side
open
feature
operator
that
now
supports
multiple
service
providers.
D
There's
some
work
to
be
done
on
passing
credentials
and
api
api
uris
through.
But
in
theory
the
default
service
provider
is
now
flag
d
and
that's
that's,
checked,
that's
validated
what
service
provider
is,
but
we
can
now
add
service
providers
like
flagsmith,
launch,
darkly
et
cetera,
et
cetera,
and
then
what
would
happen
is
that
on
the
back
end,
it
would
do
a
dynamic
resolve
based
on
the
flag
d
input
arguments
to
the
flag
provider.
So
I
actually
think
that
might
be
a
better
route
of
being
our
next
thing
to
do.
C
Can
you
help
me
understand
that
one
a
little
bit
what's
like
the
order
of
resolving
it?
One
thing
to
worry
about
with
multiple
providers?
Is
it
can
get
abstracted
from
an
end
user?
So
if
the
flag
isn't
in
one
of
these,
you
can
end
up
in
a
weird
case
where
you
don't
know
what,
where
the
value
is
coming
from.
B
So
so
to
me,
it's
this
exactly
the
same
as
a
server-side
implementation,
your
pat,
the
the
the
context,
the
flag
d
provider,
for
example,
that
we
have
passes
all
the
same
information
to
flag
d
that
that
is
available
in
the
in
the
sdk.
So
it
passes
any
user
context.
It
passes
all
of
the
same
objects
across
the
wire
via
like
via
the
rest
protocol.
B
So
it's
it's
it's
it
doesn't
the
map.
The
parameters
you
have
and
the
context
you
have
is
exactly
the
same
as
what
you
have
in
any
like
server
side.
Evaluation,
for
example,.
D
Yes,
yes,
okay,
okay,
I
can
talk.
I
can
briefly
just
walk
us
through
how
this
implementation
works
so
flag
d
at
the
moment,
has
this
this
abstraction
layer
for
service
providers.
Also,
sync
providers
syncs.
This
is
this-
is
actually
am
a
different
sync
provider.
The
service
provider
is
the
service
that
flag
d
offers
outwards
right.
So
that's
hp
only
at
the
moment
we're
talking
about
res
we're
talking
about
grpc
as
well,
etc.
D
However,
the
sync
provider,
which
is
the
internal
naming,
we're
calling
this
is
where
it
gets
its
data
from
right.
So
what
would
happen
is
that
the
kubernetes
operator,
so
this
could
be
a
vendor
and
what
happens
is
when
flagged
gets
created
in
on
the
host
by
the
kubernetes
operator.
It
just
effectively
is
starting
it
with
new
arguments.
Right,
just
likely
start
sync
provider
equals
and
then
what
would
happen
is
we
also
have
some
additional
parameters
that
get
passed
through,
such
as
credential
path.
I'm
thinking
we
actually
store
this.
D
We
actually
put
the
secret
from
kate's
on
the
disk,
so
we
never
have
to
read
it
in
plain
text.
We
just
let
that
get
read
by
the
program:
flag
d
only
or
set
it
as
an
environmental
variable.
We
can
decide
and
then
the
other
thing.
Of
course
we
can
say
something
like
the
restful
endpoint
right.
D
I
think
most
most
providers
come
down
on
a
pretty
similar
set
of
credentials
and
and
configuration,
and
then
what
would
happen
is
that,
rather
than
using
the
the
injected
config
map
that
we
currently
use
at
the
moment
like
that,
where
we
we
update
it
and
sort
of
like
the
case
layer,
if
you
will
manages
that
we
let
the
the
provider
manage
it,
so
you
can
manage
that
through
your
ui.
D
So
I
can
go
to
one
of
my
flagging
isvs
and
click
on
or
off
in
my
ui
and
suddenly
now
I'm
affecting
what's
happening
inside
the
pod.
Does
that
make
sense.
C
D
So
if
that
application
is
a
it's
like
a
flask
server
right
and
that
flask
server
here
has
a
has
a
like,
you
know,
pet
shop
kind
of
thing
and
then
or
whatever,
and
then
this
can
return
back
and
then,
of
course,
it's
just
your
normal
kind
of
flagging
situation
when
inside
the
host
application,
it
calls
the
client
library
for
flag
d
or
in
fact
this
I
think,
todd
in
your
implementation.
Don't
you
have
a
client
library?
Do
you
right?
You
just
use
restful
so.
B
I
think
part
of
the
confusion,
though,
is
in
like
actually
a
lot
of
feature
flag
vendors.
They
do
flag
evaluation
inside,
like
basically
the
host
application
in
this
context.
So,
yes.
B
That
right
right,
but
I
mean
the
question
I
guess
would
be:
is
there
a
need?
You
know
to
some
extent
it
made
sense
in
flag
d,
because
then
you
know
that's
just
kind
of
part
of
this
particular
architecture
and
it
worked
well
in
the
kubernetes
world
but
like
if,
if
you're
already
using
launch
darkly,
for
example,
to
move
it
to
a
separate
process,
may
not
you
know,
buy
you
much
yeah
the
reason.
The
reason
why
I
think
and
and
doug
from
launch
dark
was
somewhat
excited
about
it.
B
I
don't
know
if
it's
the
right
solution
for
everybody,
like
especially
in
especially
in
a
k-8's
world,
like
I
think
you
know
using
the
configmap
injected
version-
is
probably
what
we're
going
to
see
mostly.
But
I
think
this
is
worth
pocing,
because
I
think
specifically
the
use
case
that
that
I
think
was
interesting
to
to
doug.
I
don't
want
to
speak
too
much
on
his
behalf.
B
Was
that,
like
I
think,
alex
is
getting
at
it
allows
the
it
allows
the
layer
that
is
written
in
your
application
code
to
be
extremely
thin,
so
all
you're
doing
in
application
in
in
the
native
application
language,
like
maybe
your
application
is
written
in
erlang
like
something
incredibly,
maybe
it's
miranda
all
you
need
to
do
is
be
able
to
to
basically
communicate
over
a
socket
and
then
there's
no
need
to
have
a
flag
or
to
have
a
launch
darkly
sdk
made
for
that
language.
C
And
for
the
additional
languages,
then
it
makes
sense.
I
guess
like
originally.
You
know
it
was
definitely
more
exciting
with
the
different
architectures.
I
know.
That's
really
is
a
blocker
now
also
itself,
because
it's
the
end
provider
would
start
to
support
that
architecture
but
yeah.
I
could
see
a
limited
use
case
of
this
for
those
other
languages
that
don't
have
support
some
of
the
really
esoteric
ones.
B
Yeah,
I
think
that's
that's
one
one
use
case
another
one
is
like,
if
you
think
of
what's
happening
here,
you're
we're
moving
the
decision
about
about
their
feature,
flagging
implementation
out
of
app
code
and
therefore
maybe
a
little
bit
away
from
the
developer
and
towards
the
operations
person
right
so
like
the
person.
Maintaining
the
cluster
is
actually
making
decisions
about
how
this
flag
is
evaluated.
B
C
Sorry,
I
was
gonna
say
well,
the
abstracting
of
multiple
ones
is
definitely
a
question
I
have
because,
as
a
service
provider
at
the
end
of
the
day,
you
know
we're
there
to
provide
a
good
value.
If
you're
now
combining
multiple
providers
in
your
code
bases
it
gets
kind
of
complicated
in
how
is
it
getting
supported?
If
something
goes
wrong,
is
it
actually
launched
starkly?
Is
it
flagsmith?
Is
it
flag
d,
which
one
of
these
providers
is
called.
B
Yeah
and
I'm
not
necessarily
suggesting
that
people
would
build
an
architecture
like
that.
What
I
can
tell
you
is
that,
right
now
at
dynatrace
we
already
have
that
problem.
It's
already
a
problem
for
us,
because
we
have
multiple
future
flag
sources
of
truth
in
our
even
in
our
in-house
implementations,
so
starting
to
be
able
to
unify
them
behind
a
single
abstraction
at
least
helps
us
move
towards
finding
a
solution
to
that
problem.
B
I
I
don't
necessarily
see
it
as
a
target
as
much
as
it's
a
it's
kind
of
a
reality
for
some
people
and
then
so
it
does
help
in
that
regard.
B
D
I
was
just
going
to
make
that
this
is
opening
up
a
new
market
for
these
flagging
providers,
who
traditionally
have
been
aiming
towards
application
developers
right
and
maybe
even
product
folks
who
want
to
do
ab
a
b
testing.
This
now
lets
you
create
flags
that
control
operators
inside
of
kubernetes
clusters
right,
so
you
can
do
infrastructure
flagging,
turning
off
load,
balancers,
there's
nothing
quite
like
that
exists
currently.
So
I
think,
there's
a
quite
exciting
opportunity
here
to
be
leveraged.
B
Yeah,
that's
an
excellent
point
actually
to
alex.
Who
is
it?
Somebody
asked
this
last
week.
I
don't
know
if
you
were
there
or
if
your
head's,
just
in
the
same
place
yeah
some
somebody
asked
the
ex
they
were
like
hey.
Could
we
do
routing
decisions
in
nginx
based
on
this
concept
based
on
flag
d,
and
the
answer
is
absolutely
you
could
like?
D
Interesting
because
I
was
also
sorry
I
was
just
going
to
say,
I
was
also
looking
at
losing
like
a
wasm
filter
on
flag
d
that
could
do
resolution
based
on
incoming
arguments
and
that's
something
else
we
can
do.
I
know
we've
talked
about
using
queue
as
an
evaluation,
some
dsl
as
well.
So
there's
a
lot
of
ways.
You
can
build
that
out.
A
C
Yeah,
I
do
think
these
will
help
with
the
learning.
Overall
we
do
have
some
customers
using
us
with
nginx.
Today.
I
actually
know
one
customer,
instead
of
using
the
hpa,
actually
uses
launch
directly
because
it's
more
limited,
but
those
are
much
more
one-offs
versus
industry
kind
of
concept.
So
hopefully
we
can
kind
of
move
it
in
the
operational
direction.
B
Yeah
yeah
and
and
like
I
said
I
mean
I
just
want
to
reiterate
like
to
your
original
point.
Dan,
the
the
using
a
vendor
in
flag
d
is
a
valid
use
case,
but
I
think
it's
probably
not
a
primary
use
case.
Right,
like
I
think.
The
primary
use
case
of
flag
d
is
going
to
be
reading
the
config
maps,
but
I
think
that
it's
certainly
it,
but
because
we're
kind
of
already
almost
there,
because
we
have
all
the
abstractions,
because
we
have
the
golang
sdk
and
we're
gonna
have
golang
providers
it
make.
C
B
Yeah
yeah,
I
do
it
does
make
sense,
so
I
don't
think
it's
it's
likely
to
have
like
some
java
application
called
through
flag
d
to
call
the
launch
directly
java
sdk.
It
just
doesn't
really
make
sense,
but
so
yeah
yeah
yeah.
I
think
there's
there's
some
use
cases
but
they're
more
they're,
not
the
primary
use
case.
Like
I
said,
I
would
expect
that
90
of
flag
d
deployments
are
probably
going
to
be
reading
going
to
be
reading
a
config,
but
that
could
change,
because
I
think
this
could
be
a
potentially
powerful
concept.
A
D
I
had
a
few
thoughts
around
that
I
mean
I
don't
know
it's
a
good
thing
to
mention
now
from
the
agenda
at
some
point,
that
is,
to
open
up
other
ways
that
flag
d
can
have
more
dynamic
updates,
and
one
of
them
is
to
read
the
api
server
directly.
So
it's
how
they
have
a
have
an
event
stream
on
the
api
server,
and
then
we
actually
put
the
data
into
edcd
and
we
do
that
through
primitives
right,
so
we
can
kind
of
cut
out
the
middleman.
D
I
mean
it's
actually
the
same
thing
if
you
think
about
it
other
than
having
to
mount
the
volume
and
then
and
then
have
to
to
read
the
path,
we're
doing
we're
cutting
that
out
and
going
straight
to
the
server.
So
I
think
that
might
be
the
next
evolution
of
the
sync
is
to
have
a
an
api
server.
Sync.
B
Yeah
that
actually
I'm
assuming
that
would
probably
cut
down
on
kind
of
suppose
the
latency
almost
of
the
updates.
I
don't
know
if
you've
gone
through
the
demo,
but
it
takes
about
a
minute
sometimes
to
like
the
record.
D
Yeah
I
mean
there
will
definitely
be
use
cases
where
it's
not
possible.
Like
I
know
you
know,
working
at
jp
and
amex
those
sort
of
places
you're
never
going
to
be
given
permission
in
a
tenant
namespace
to
read
off
the
api
server,
because
it's
also
a
cross
issue
right.
You
can,
you
can
hammer
it,
but
there
is
certainly
a
like
a
bunch
of
folks
who
don't
care
about
that.
So
you
could
have
said
that
maybe
that
could
even
be
the
default.
It
would
definitely
be
much
quicker
because
we.
D
Yeah
the
service
account
permissions
would
need
to
be
more
extensive,
yeah.
B
Yeah,
but
that
makes
sense
it's
still
a
really
interesting
point.
It
makes
sense
to
me,
but,
speaking
of
speaking
of
the
reconciliation
time
we
had,
we
had
a
an
issue
to
add:
add
an
annotation
to
our
workloads
like
that.
Would
that
would
represent
the
hash?
I
don't
know
if
that
ever
got
action.
Did
we.
D
That
was
to
to
check
to
see
if
the
hash
changed.
I
thought
I
did
that.
D
Can
check
that
so
that
would
be
in
the
the
mutating
the
the
the
web
hook
right.
B
A
B
What's
I
mean
those
are
all
the
things
I
think
I
wanted
to
talk
about.
I
don't
know
if
anybody
else
has
anything
they
want
to
bring
up,
but
those
were
the
ones.
D
Yeah,
I
think
you
might
be
right,
I'm
not
sure
if
it
is
in
there.
What
was
it
was
too.
It
was
just
store,
the
the
shah
of
the
sorry
that
the
the
the
integrity
check
of
the
payload,
so
you
didn't-
have
to
run
it
through
the
validator
right.
If
it'd
already
been
validated.
B
B
So
in
those
cases
we
didn't
want
to
do
any
reconciliation
of
the
actual
config
maps,
but
the
other
thing
was
that
we
were
going
to
as
part
of
that
add
that
same
sha
to
an
annotation
on
the
actual
workload
on
the
on
the
deployment
itself,
which,
if
I
understand
correctly,
adding
an
annotation
to
a
workload
hypothetically,
should
it
should
decrease
the
propagation
time
in
updating
the
config
map.
So
it
hypothetically
should
just
make
things
quicker.
Was
my
understanding,
yeah.
B
B
D
B
D
Yeah,
so
it
just
just
does
like
a
checksum.
D
D
A
D
Trickier
because
the
we
don't
own
the
car,
we
don't
own
the
deployments
right,
we're
not.
We
don't
have
an
owner
reference
on
his
deployment,
so
we
have
to
scout
in
it's
fine,
but
it's
just
as
you
scale
the
cluster
out
that
becomes
quite
intensive
scanning
every
deployment
in
every
namespace
for
that
annotation
for
those
levels.
B
Yeah
I
mean
I
mean,
maybe
maybe
that's
not
something
we
want
to
do.
Maybe
if
the
only
reason
to
do
that
is
to
is
to
lower
the
reconciliation
time.
Maybe
it's
not
something
that's
worth
doing.
I'm
not
personally
of
the
opinion
that
we
need
to
have
flag
values
be
propagated
faster
than
a
minute.
I
don't
know
if
other
people
disagree
like.
I
think,
if
you
do
need
that,
maybe
you
should
again
consider
a
vendor.
B
D
I
think
we
definitely,
I
want
to
aim
to
try
and
lower
it
as
much
as
we
can,
because
that
is
quite
slow,
but,
let's
just
see
I
mean
I
was
seeing
20
seconds
30
seconds
when
I
was
doing
direct.
B
I
mean
it
depends
on
when
you
do
it,
you
can
hit
it
just
wrong
and
it
takes
about
a
minute
where
it
becomes
slightly.
Annoying
is,
if
you
have
multiple
pods,
you
can
see
it
basically
slowly
roll
out
and
that's
that's
because
a
bummer
like
a
minute's,
no
big
deal.
But
if
it's
10
seconds
between
you
know.
D
B
Yeah
when
I,
when
I
like
it
just
as
an
experiment,
I
added
a
whole
bunch
of
replicas
to
our
workload
and
it's
interesting
what
the
delta
is
so
and
maybe
it
does
have
to
do
with
that
backup
retrieve
because
it
seems
like
there
could
be
a
very
fairly
large
delta
between
the
first
workload
be
having
its
config
map,
be
updated
and
and
the
last
one.
But
again
I
I'm
not
convinced
any
of
this
is
like
a
huge
deal.
Breaker
for
the.
D
I
think
we're
getting
to
a
point
where
we
actually
need
some
user
feedback
as
well
on
this
stuff.
You
know
to
really
decide
that.
A
D
What's
the
what's
the
timeline,
I
mean,
I
haven't
attended
some
of
the
other
meetings,
but
what's
the
timeline
on
sandboxing.
B
Yes,
yeah,
I'm
just
filling
out
the
paperwork.
Now
I've
been
in
europe
for
two
weeks,
so
it's
slightly
delayed.
But
when
I
get
back
it's
gonna,
something
I
knock
out,
but
yes,
it
has
been
accepted
and
then,
when
it
comes
to
spec
and
some
of
the
sdks
we're
pretty
solidly
in
alpha
now
and
we
have
java
and
net
in
a
pretty
good
state
right
now,
I
think
go
won't
be
too
far
behind.
A
D
We,
I
think,
maybe
then
from
like
the
flag
b
perspective
in
the
operator.
We
need
to
just
really
really
focus
on
kubecon
to
get
good
examples
like
more
good
end-to-end
examples,
and
so
maybe
the
thing
we
should
do
is
harmonize
on
the
next
sync
provider.
We
end
up
building
or
that's
our
next
service
next
service
provider
we
end
up
exposing,
I
mean,
maybe.
B
Something
infrastructure
could
be
really
cool.
I
mean
that
that
is
just
such
a
new
concept.
To
some
extent,
you
know
that
it
could
be
a
pretty
cool.
You
know
proof
of
concept.
At
a
conference
like
kubecon,
I
I
mean
actually
maybe
doing
routing
with
nginx
is
a
good
use
case,
because
it's
in
lua
and
lua
is
not
a
not
an
extremely
popular
language
and
we
could
easily
write
a
lua
provider
that
would
just
talk
to
flag
d,
and
then
we
could
do
some
like.
Maybe
we
could
find
some
interesting
routing.
D
I
would
like
to.
I
would
like
to
write
an
admission
controller
that
has
flagging,
enabled
to
check
to
be
able
to
turn
on
and
off
the
ability
to
do
cosine
checking
in
queue,
because
I
think
cosine
with
sigstor
is
a
very
hot
topic
and
I
think
q
is
a
very
new
language,
so
actually
having
a
dsl.
You
can
mount
as
a
config
map
that
could
be
an
alternative
to
just
json
might
be
interesting.
A
D
B
Right,
yep,
yeah
and-
and
I
think
like
that
whole
that
whole
eye
evaluator,
I
think
it
is-
is
the
interface.
So
you
can
really
plug
it
with
anything.
So
you
could
easily
implement
a
q1
which,
which
might
might
be
more
like
cloud
native
than
json
and
maybe
like
more
attractive.
So
I
think
we
could
plug
in
anything.
We
want,
maybe
even
becomes
the
new
default.
I
think.
D
It
would
definitely
get
a
lot
of
interest
because
q
is
such
a
an
interesting
language
to
people.
B
Yeah
yeah
I'd
be
excited
to
see
that
so,
just
to
kind
of
summarize
that
you
know
it's
the
the
mission,
the
mission
controller,
to
toggle,
what
exactly.
D
So
for
a
q
con
demo
or
a
demo
generally,
I
think
it'd
be
interesting
to
have
the
ability
to
show
some
sort
of
infrastructure
capability,
and
so
to
do
that,
you
could
write
an
operator
that
has
an
admission
controller
so,
and
what
would
happen
is
that
it
does
what
we
do
in
some
ways:
it
scans
the
pod
spec
the
object
spec
coming
to
the
cluster.
It
looks
at
the
image
and
if
you
flag
the
feature
on
it,
doesn't
just
look
at
the
image.
D
It
validates
that
that
is
a
signed
image.
It's
a
new
it's
worth.
Checking
out
six
store
is
a
project
that
has
just
been
adopted
by
kubernetes
upstream.
D
A
B
B
Great
but
yeah
that
sounds
like
a
really
cool
demo
and
I
think,
like
all
of
the
subject
matter
around
it
seems
really
hot
and
interesting
to
talk
about
kubecon.
So
I
I
like
that
idea.
A
lot.
A
I
put
a
should
I.
B
B
Okay,
that
would
that
would
make
sense
you
probably
yeah
they're,
probably
heavily
restricted
actually
well,
it
sounds
like
I
mean,
maybe
I'm
not
sure
if
lua
will
allow
you
to
do
like
like
if
well,
I'm
not
sure
if
a
framework
allows
you
to
do
in
a
different
thread
like
some
kind
of
other.
Maybe
it
does
can
do
some
other
kind
of
polling,
but
if
it
can't,
then
that's
just
a
that's,
not
an
idea
we
have
to
implement.
B
D
Okay,
I'll
write
a
quick
issue
up
and
we
can
take
it
offline
and
chat
about
that.
Perfect
cool.
B
Lars
yosa
have
anything
else
specifically
to
add
or
any
questions.
I
know
it's
probably
a
little
off
topic
for
your
lars
compared
to
what
you're
probably
expecting,
but.
A
B
Perfect
yeah,
it
is
pretty
interesting.
I
think
that
that's
why
we
kind
of
split
the
meetings
off
too.
It's
like
this
can
be
very
specific,
for
you
know
some
of
this
maybe
infrastructure
and
like
the
flag,
d
processes
and
kind
of
some
new
ideas
there
and
the
other
one's
more
around,
like
the
spec
itself
and
kind
of
defining
that
all
out
so
both
are
progressing
pretty
nicely,
but
they're,
pretty
distinct
topics.
D
About
the
logo
situation.
B
No
I'm
glad
he
brought
that
up.
Actually,
that's
a
good
point.
I
I
like
have
been
I've
had
that
on
my
brain
for
a
while,
I
kind
of
I
gotta
say
I
really
like
your
idea
of
doing
like
a
a
contest
I
would
be
down
for
that.
I
don't
know
it
would
be
cool
to
have
some
kind
of
half
decent
prize,
but
yeah
I
don't
know
alex,
is
going
to
personally
build
and
ship
someone
like
a
raspberry
pi,
kubernetes
cluster.
D
D
B
B
Yeah,
I
think
that's
good,
I
mean
we
could
start
there.
If
it's
a
flop,
then
we
make
our
our
own
logo
so
yeah
I'll.
Take
that
as
a
an
action
item
then
see
what
we
can
do
on
on
our
side.
If,
if
you
know
anyone
at
canonical
that
feels
like
getting
involved
too,
certainly
let
me
know,
but
it
probably
isn't
too
difficult
to
find.
You
know
some
someone
to
sponsor
a
50
prize
or
something
and
oh.
B
Yeah,
if
anything,
I
think
the
more
difficult
thing
would
be.
You
know
trying
to
get
making
sure
that
we
have
enough
participation
within
the
community
and
that
there's
also
like
even
outside
of
our
community
so
far,
it'll
be
a
good
opportunity
to
involve
new
people.
Also.
D
Yeah
for
sure
the
ultimate
prize
I
can
think
of
would
be
like
virtual
or
you
know,
some
sort
of
conference
ticket
or
something
like
that.
We
have
we
get
loads,
but
unfortunately
we're
going
to
use
them
because
we
send
like
30
people
every
year,
but
I'm
just
trying.
A
D
B
Can
be
vip
access
to
the
open
feature
party
at
cubecon?
You.
D
B
Send
them
some
stickers
so
yeah.
Let's
think
about
that.
I
think
that
would
be
kind
of
fun
like
why
not
we
can
give
it
a
shot.
If
it
doesn't
work
out,
you
know
no
loss,
there
really
yeah.
So
maybe
we
could
have
something
by
by
next
week
community
meeting,
and
that
would
be
my
goal
yeah.
If
I
could
have
something
kind
of
prepared,
we
could
mention
it
there
and
try
to
build
some
buzz
around
it.
Try
to
get
some
people.
B
D
D
D
B
D
The
linux
yeah,
the
cncs
section,
make
sure
that
I
would.
I
would
make
the
thing
a
google
form
right,
you'd
attach
it
and
you
just
say
it
be
in
the
form
please.
This
is
the.
This
is
the
agreement.
This
is
the
license
just
so
you
know,
you
know,
and
you
know
just
link
to
the
the
the
contributor
covenant
that
people
are
expecting,
because
we've
got
that
in
now
as
well.
Don't
we.
A
B
And
that's
probably
also
another
reason
to
keep
the
prize
money
low
or
not
money,
but
like
a
of
low
value,
obviously
probably
getting
the
weird
like
tax
and
other
issues
that
I
don't
feel
like
dealing
with
so
yeah.
I
know
I
think
that
that's
that's
cool
it'd
be
fun
to
get
people
involved
in
that
you
know
all
right,
any
other
topics.
We
should
chat
about.