►
From YouTube: Kubernetes SIG API Machinery 20170816
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
B
C
What's
the
intention
of
this
presentation
is
give
you
and
status
update
on
the
org
and
was
doing
and
I
will
get
it
from
there.
So
kubernetes
api
have
nice
features,
it
dress
it
stressed,
but
it
has
some
more
features
like
watch.
You
know
port,
forwarding
and
stuff
like
that.
But
the
good
thing
is:
we
have
a
fan.
Api
need
sugar
soon
it
should
be
be
tree
and
what
was
so
funny
here
for
people
that
they
don't
know.
What
that
is.
Is
the
machinery
level
interface
for
ascribing,
producing,
consuming
and
visualizing?
Yes,
cool
API.
C
It
has
definitions
for
a
cisco.
Is
api
objects
they
have
pads
for
operations.
It
does
support
that
an
authorization
and
authentication
and
have
extensions
that
we
use
for
some
of
our
things
that
we
yeah
editing.
Kubernetes,
we
decided
to
go
at
every
days.
We
decided
to
go
with
cogeneration
for
non
go
clients
because
having
the
swagger
or
open
api
would
be
a
very
nice
thing
to
use
as
an
equal
to
code.
Generators
and
code
generation
is
very
good
because
you
can
design
the
client
you
can
generate
it.
C
C
We
use
auto
rs4
c-sharp.
We
decided
that
we
need
approval
for
any
other
languages
to
use
anything
other
than
swagger
coach.
Just
like
a
quark
and
I
said
why
why
we're
using
code
generators?
It's
also
less
box.
That
means
you
know.
When
you
add
things
manually,
there
is
also
always
a
chance
that
you
miss
the
approach.
C
This
structure
that
we
decided
to
have
coins
is
having
one
single
generator:
repo
akane-chan,
when
it
is
clients
instruction
that
has
all
code
generator
in
it,
and
it's
not
actually
it's
not
running
in
containers.
It
supports
all
of
the
languages
that
swagger
coach
and
supports,
then
for
each
language
we'll
have
a
base.
C
That's
basically
the
share
utilities
that
we
need
to
for
the
client
works
with
cornetist
disabilities,
things
like
loading,
hugh,
config
and
or
supporting,
watch
or
supporting,
set
or
any
type
of
sockets,
and
then
it's
the
official
language
repo
that
is
basically
auto-generated
in
eating
in
a
nine-year-old.
It
will
be
completely
auto-generated
and
only
depends
on
the
base
and
then,
with
this
design,
a
user
can
come
with
their
own
open,
API,
spec
use
of
a
display
could
be
for
a
CRT
or
a
customer
care
server
user
API
server.
C
The
way
that
the
generation
works
is
actually
generating
operation,
classes
and
definition
classes
based
on
the
operation
and
classes
as
an
IPA
objects.
It
also
has,
it
also
will
generate
a
rest,
client
and
configuration
object.
I
call
these
two
an
old
one
generation,
because
there
is
only
one
of
them
operations
and
definitions
are
Om
based
on
number
of
API
objects
and
operation.
You
have
and
then
we
add,
support
for
these
are
manually
configured
or
watch
the
alert
actually
rest
client
and
configuration
to
be
customized
and
be
part
of
the
utility
the
base
package.
C
C
We
also
have
some
badges
for
for
clients.
Capability
fighters
must
tell
you
what's
in
supports,
promise
means
you
can,
they
can
know
if
you
convict
and
they
can
run
in
in
cluster,
but
they
don't
support
things
like
watch
or
exact.
If
they
support
to
act
watch
they
are
going
to
be
silver.
If
they
supporting
proto
encoding
and
they
generate
examples,
we
call
them
even
our
first
client
fighting
is
not
good.
Yet.
B
Many
I
got
a
question
on
this
page.
Actually,
something
has
come
up
from
the
sig
COI
team
is
they
were
looking
at
what
it
was
gonna
take
to
actually
run
their
Q
control,
extensions
and
the
exceptions,
no
matter
what
language
that
written
in
our
most
likely
gonna
need
to
load
a
queue,
config
file
and
be
able
to
understand
the
different
flags.
So
you
know
how
we
have
flags
that
will
override
the
values
from
Cuba
config
file
in
our
go
client.
Is
that
included
in
bronzers
that
not
included
any
level
today.
C
C
That
means
you
should
at
least
have
bronze
requirements
for
the
capabilities,
and
then
you
should
have
visible
state
stable
releases
instruction
instruction
and
at
least
two
owners
unstable,
better
close
documentation
for
each
platform.
The
application
policy
is
release.
Versioning
should
be
documented
clearly,
and
we
should
have
some
documented
users.
C
A
C
C
Has
one
on
for
release
supposed
to
come
fake
and
partially
X?
They
can
watch
I'm,
say
partially,
because
I
didn't
test
them.
Show
me
yet,
but
they
are
there.
Javascript
is
in
start
of
development,
need
contributors,
I'm
working
on
that
right
now
and
keep
pumping
loader
is
in
progress.
Typescript
is
in
development.
It's
supposed
to
come
together.
C-Sharp
is
also.
C
G
Why
you
have
to
do
executive
tab
to
be
gold?
It
seems,
like
blog
people
could
be
a
long
way
to
the
climb.
It
doesn't
everything,
but
those
that
need
interactively
do
stuff,
especially
attached
like
what
seems
like
it's
the
hardest,
because
it's
screaming,
then
you
could
just
fall
that
you
could
just
Python.
You
know
execute
you,
shell
out,
so
why?
Why
is
not
a
requirement?
Why
not.
C
C
So
it's
we're
supporting
of
sockets.
We
are
not
supporting
SPDY,
so
it's
not
that
hard,
and
also
they
think,
is
that
a
client
can
be
completely
stable
in
a
stable
support
level,
but
only
be
silver.
You
know,
does
a
support,
xx
or
attached
yeah.
We
can
consider
that
if
you
think
it
should
be,
two
sockets
are
so.
B
C
C
B
I
B
C
Okay,
I
couldn't
actually
an
example,
he's
asleep
to
show
people
house
how
easy
it
is,
for
example,
in
fighting
to
use
the
clients
to
basically
important
things.
Don't
you've
configured
reloaded
in
the
application
with
all
the
coffee
needs,
and
then
you
can
just
say:
leave
this
box
blank
spaces
or
we
can
start
watching
on
namespaces
in
a
very
simple
way.
C
Actually,
this
simplicity
is
because
of
this
simplicity,
I
mean
not
having
a
lot
of
dependencies.
There
are
some
people.
Ask
me
why
you
don't
generate
for
client.
Can
regenerate
go
with
the
same
thing.
We
generated
these
clients
and
the
answer
is
yet
yes,
too
early
to
supported
answer's.
No,
you
have
a
graphical
clients,
but
users
can
still
go
and
generate
in
the
client
using
the
same.
E
E
C
C
I
mean
it's
all
about
people
we
are
I
listed.
The
people
are
working
on
clients
here
JavaScript.
It
definitely
needs
more
people.
We
need
to
have
actually
true
owners
for
each
of
them.
We
don't
even
for
Python
I'm.
The
only
honor
I
have
a
lot
of
people
help
in
there
and
fight
with
them,
but
none
of
them.
C
C
B
B
B
So
we've
we
read
have
Stephan
in
particular
and
is
familiar
with
the
tool
because
we
helped
Chow
when
the
initial
reviews
for
writing
at
least
the
ones
upon,
but
we
need
help
getting
the
actual
poles
themselves
merged
and
that
Munder
bot
active
we
want
in
1/7
we
started
creating
branches.
We
have
a
pole
to
create
one
seven
branches
for
people
to
be
able
to
find
their
dependencies
when
they
try
to
vendor
pieces.
There's
a
pole
open
to
do
that.
It's
been
lingering
for
a
while,
and
then
we
have
a
pull
for
cloning
case.
B
Io
metrics
believe
that
one
also
moved
in
one
seven
and
so
they've
been
sitting
out
there
for
quite
some
time.
I'd
really
like
to
get
some
attention
on
it
and
I
was
gonna
gonna
ask
we
can
do
the
reviews,
but
we
don't
own
the
test
in
for
a
tool
and
a
chow
has
helped
in
the
past.
I
was
going
to
ask
for
help
from
Chow
to
to
push
these
so.
J
I
I
hope
they're
I'm
a
match.
Last
Thursday,
he
says
he's
working
on
yet
and
I
think
it
will
make
this
process
faster.
If,
if
people
can
test
it
themselves
and
then
hope,
I'm
a
mentor
me
to
deploy
it
I
had
a
dog
checking
like
one
month
ago.
It
documented
how
to
test
these
taking
recourse
under
your
own
organization.
I
think
that
might
be
helpful
and
I
think
in
the
long
term.
We
somehow
need
to
open
the
commission
of
these
merge
spots
to
non-googlers,
otherwise
it
will
always
here
it
will
always
be
a
for
tonight.
B
Yeah
I
I
think
that
too
you
know
we
like
it
we're
willing
to
open
polls
in
there.
We're
certainly
willing
to
help
do
the
reviews.
We
helped
with
the
initial
review
to
get
this
stuff
in,
but
we
don't
have
actual
rights
to
do
that
final,
merge
of
it.
So
so
in
this
case,
if
you
could
link
in
the
how
to
test
directions,
we'll
give
them
a
try,
these
ones
that
have
lingered
in
the
ones
that
we
have
open
now
right.
We
have
open,
now
closed,
set
off.
J
F
B
B
I
think
it's
a
little
different
if
we
Vince
air,
saying
hey,
we
that
does
this
yeah,
it
said
having
a
pool
and
saying
hey.
Yes,
let's,
let's
create
this
in
all
the
understand,
so
you
know
on
a
similar
note,
we
have
a
new
staging
repo.
We
initially
called
it
Q
gen,
but
after
some
discussion
on
the
on
the
message
board,
Walter
suggested
a
code
generator
and
that
sounded
fun,
so
we're
gonna
have
community
use
code,
generator
and
I've
linked
in
holes.
Here.
B
The
generator
contains
the
generators
that
we
use
for
API
types,
so
it
contains
generators
for
40
copy
and
conversion
and
defaulting
and
clients,
so
that
if
someone
wants
to
be
able
to
build
a
cube
like
API
for
something
like
say,
an
aggregated
API
server,
there
will
be
a
place
where
they
are
going
to
be
able
to
vendor
or
pull
the
generators
without
actually
having
a
vendor
all
of
queue.
So
Stefan
has
gone
through
he's
sniff.
B
The
links
may
be
staging
repo
and
he's
got
to
pull
open
to
rename
it
to
meet
the
name
that
we
agreed
on
and
then
pulls
open
again.
There's
some
secret
google
sauce
that
that
needs
to
be
done
for
two
of
them
they're
linked
in
here
and
don't
you
worry
I
will
bug
people
again
next
week
about
trying
to
get
these
reviewed
and
merged
because
it
would
be.
You
know
it
would
be
very
helpful.
D
B
B
B
Alright,
so
that's
it
for
the
code
generator
couple
releases
we're
going
to
trying
to
remove
cube
isms.
You
know
right
now,
so
some
of
the
generators
actually
hard
code
cube
names
will
be
trying
to
fix
those
pull
them
out
of
defaults.
Make
cube
vendor
pretend
that
a
vendors
these
in
and
interact
with
them
via
shell
script,
like
everyone
else
quote
so
that'll,
be
ongoing.
Work.
J
So
I
have
a
few
outstanding
issues
related
to
the
initializer
presume.
It's
an
idea
to
theta
the
first
one
is
we
I
want
to
change
the
API
key
supporter
to
make
the
initializers
culture
like
audit,
because
if
all
the
initializers
are
running
in
Syria
to
a
long
time,
I
think
Eric
also
has
something
to
say
young.
This
one.
G
J
I,
don't
have
a
direct
number,
but
from
I
will
we
can.
We
can
do
agitated
death
because
all
the
initializer
controller
needs
to
what
for
the
objects
and
watch
latency
can
be
as
long
as
several
eyes
several
seconds
and
then
the
initializer
controller
needs
to
update
it
of
it
and
that
will
take
several
hundred
of
microseconds,
so
millisecond
I
think.
B
That's
actually
listed
inside
of
wojtek's
sli
RSL
aslo
proposal
that
I
believe
merged.
Today
he
has
split
the
metrics
apart
for
ATM
machine
er
II
to
be
able
to
do
lists
and
API
responsiveness
without
initializers
and
then
API
responsiveness
with
initializers.
He
has
goals
for
without
initializers
I,
don't
believe
that
he
listed
goals
for
with
initializers
step.
One
was
measures
what
we
had.
He.
A
B
B
I
I
see
this
as
a
configuration
thing
right,
I
mean
I
wanted
to
be
as
fast
as
possible,
but
ultimately,
if
a
cluster
admin
decides
that
he
wants
to
have
an
initializer
and
that
initializer
happens
to
be
slow,
I
think
it's
important
for
us
to
say
that
it's
slow.
We
should
be
able
to
describe
why
it
is
slow.
G
A
B
A
G
D
F
D
B
So
I'm
in
favor
of
measuring
that
I'm
in
favor
of
looking
to
see
where
the
fat
is
in
favor
of
looking
to
see
where
those
are,
since
it's
built
on
top
of
the
api's
that
we
already
host
I,
don't
anticipate
we're
gonna
get
any
major
surprises
right.
As
Chow
was
hanging
to
write
an
initializer,
you
essentially
have
a
watch
and
you
have
a
patch
and
then
the
sed
watch
has
to
conclude
and
then
the
response
is
sent
right.
So
the
the
overall
flow
is
known.
It
is
not
a
new
flow.
F
B
L
B
L
G
L
L
L
External
web
hooks
is
the
name
that
we
use
for
the
generic
mechanism
for
making
an
admission
call
outside
of
the
system
as
a
synchronous
call.
So
image
and
authorization
are
specific
types
of
generic
web
hooks
that,
if
re
implemented
today
are
just
another
variant,
are
specific
versions
of
the
generic
one.
Okay,.
L
J
L
G
G
L
If
pods
are
gonna,
be
the
thing
that
are
chained,
the
biggest
problem
with
pods
is
that
stupid
things
that
try
to
create
10,000
of
them
and
then
don't
even
look
at
the
fact
that
they're
failing
all
over
is
still
more
of
a
problem
today
then
I
suspect
the
additional
latency
from
this
would
be
like
I've,
never
heard
someone
say.
Oh
my
gosh,
my
10,000
pots
didn't
start
at
5
seconds
in
a
serious
name.
They
know,
but
they're
not
saying
it
today
and
the
things
they're
saying
today
are
high.
J
I
was
in
the
moment.
The
second
topic
is
so
we
so
I
there
there's
a
PR
moving
a
certain
admission
controller
from
from
being
an
admission
controller
plug
into
B
and
the
initialization
controller,
and
the
question
is:
shall
we
keep
the
initialization
controller?
Was
our
regular
controller
C
into
cube
controller
manager
binary
you.
B
Know
in
this
case
I
think
it
was
actually
the
cloud
controller
manager
and
think
that
one's
slightly
easier
to
see
right.
You
have
a
cloud
controller
manager.
The
cloud
controller
manager
requires
that
particular
labels
are
set
in
order
to
do
the
work
that
it
needs
to
do,
and
so
it
made
sense
to
bundle
the
initializer
into
that
process.
I
don't
object
to
having
it
separately,
if
I
guess,
if
someone
feels
really
strongly,
but
that
seems
like
if
you
want
one,
you
want
the
other.
So.
G
B
G
B
G
J
This,
so
that's
actually
the
sir
topic,
so
an
alternative
to
fix
the
replication
controller
is,
to
just
add
a
garbage.
Collector,
that's
well
garbage
collector,
slow
to
initialize
objects,
so
replicate
the
replication
controller,
well
keep
creating
more
out,
but
there
will
be
a
garbage
collector
to
remove
those.
Oh
yeah.
L
L
Failure
mode
of
a
cluster
that
starts
to
slow
down
is
instead
of
eventually
getting
the
work
done.
Yeah
starts
doing
zero
work,
which
does
seem
to
weary
me.
That
worries
me
a
little
bit
like
part
of
the
nice
things
about
the
controller
model
is
that
things
can
go
for
quite
a
long
time
and
then
eventually
reconcile
at
the
moment
we
introduced
timeouts
on
initializers.
We
end
up.
You
know
you
can
end
up
branching.
B
G
B
G
G
I
want
the
controller
were
close,
though
the
cig
apps,
to
take
on
the
job
of
making
sure
that
we
back
off
for
any
kind
of
error
in
the
controllers
as
a
best
practice.
Additionally,
people
will
write
back
stuff,
so
puppy
wait,
minutes
in
the
server
side
and
then
initializers
is
a
sub
k.
So
those
two
things
we
have
to
do:
yeah
I.
G
L
L
201
accepted
it's
weird,
because
what
we're
trying
to
do
is
avoid
breaking
old
clients
and,
like
the
problem,
is
that
old
clients
were
broken
because
they
were
like
this
should
take
place
instantly
and
that's
not
true,
and
so
that
that
isn't
part
of
that
I
think
it's
reasonable
to
say
that
maybe
we
should
just
be
a
little
bit
less
aggressive
around
get
for
uninitialized
resources
in
our
client
tools.
So.
G
L
Good
right
and
that's
already
supportive
done
create
today,
it's
the
you
didn't
make
a
subsequent
get,
and
today
we
tell
you
it's
not
found
because
initialize
deletion,
we
could
optionally
make
it,
but
the
goal
here
is:
you
have
to
be
backwards,
compatible
with
clients
and
that's
the
primary
design
factor
and
the
unfortunate
thing
is:
there's
no
way
to
be
backwards,
compatible
with
clients
that
there's
going
to
be
perfect.
So
forget
the
initial
alpha
erred
on
the
side
of
pretend
it's
not
there.
Well,
we
could
protect.
B
B
L
G
B
All
right,
so
one
other
thing,
I
just
want
to
call
out
so
people
know
and
then
I
think
we're
gonna
call
it
for
today
is
that
the
paging
design
is
open.
There's
a
community
bull
hasn't
been
a
whole
lot
of
comments
in
there
I
suspect
it
just
hasn't
been
called
out.
So
that's
one
I
saw
something
on
mailing
lists
about
it.
Today,
it's
worth
going
through
and
looking
especially
if
you
have
aging
experience.