►
From YouTube: WG Component Standard Office Hours 20191210
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).
A
Okay,
welcome
everyone
to
the
tuesday
december
10th
working
group
component
standard
office
hours.
This
is
our
hour
for
folks
to
ask
questions,
hopefully
get
answers
and
you
know
whatever
people
need.
So,
let's
get
started,
I
think
alex
you
want.
You
had
some
stuff
from
this
morning.
You
wanted
to
discuss.
A
B
Yes,
yes,
all
right,
yeah
regarding
very.
A
Nice
little
display,
in
your
background,
by
the
way.
B
Yeah,
it's
it's
like
traditional
eastern
german,
like
wood,
art,
woodwork
kind
of
stuff.
Obviously
they
do
it
like
in
the
iron
mountains.
A
B
Nice
all
right
yeah
regarding
the
controller
manager
first
have
some
newbie
questions,
so
you
mentioned
that
you
say
like
we
should
look
at
all
these
steps
necessary
to
get
to
beta
is
that
right.
A
B
So,
do
you
mean
like
move
the
the
external
types
to
v1
beta1.
A
Yes,
that's
I
want
to
figure
out
what
that's
like
the
most
direct
signal
that
we've
actually
been
able
to
move
things
forward,
and
I
want
to
figure
out
what
the
actual
blockers
are
for
that,
and
I
think
I
think
it's
basically
the
subset
of
config
in
the
type
matches
like
what
we're
comfortable
with,
and
you
know
it
works
to
load
from
a
file.
A
A
So
what
we
did
in
the
cubelet
right
was
we
we
implemented
all
this
like
complicated
backwards
compatibility
logic,
so
that
we
could
incrementally
migrate
flags
into
the
config
and
it
might
be
better
to
to
try
and
just
do
it
as
like
a
one-off
here,
you
can
either
you
use
either
or
but
not
both
kind
of
migration,
because
then
we
won't
need
that
layer
and
and
legacy
flag
was
the
idea
around
like
oh
well.
A
So
I
think
I
think
the
one
thing
that
one
thing
that
blocked
cue
proxy,
for
example,
from
moving
forward
with
the
either
or
thing,
was
that
we
hadn't
come
up
with,
like
any
real
solution
to
the
instance
specific
config,
and
we
thought
there
might
be
a
few
different
ways
to
do
it.
And
so
we
didn't.
We
didn't
necessarily
have
a
way
to
do
all
the
flags
at
once.
A
Oh,
that
was
just
like
stuff
like
hostname
override
note.
Nodeip
are
fields
that
are
set
differently
on
every
instance
of
a
of
a
like
process.
A
B
A
So
having
some
sort
of
solution
to
that
is,
is
kind
of
a
blocker
to
being
able
to
do
the
like
all
at
once,
applied
to
config
migration
and
that's
kind
of
what
pushed
us
on
the
keyword
side
into
doing
the
incremental
stuff
that
and
the
cubot
had
a
bunch
of
other
options
that,
like
people,
weren't
sure
if
they
wanted
moved
or
not.
The
cuba
was
a
lot
more
complicated
of
a
case.
B
Solved,
okay,
so
I
mean
like
from
when,
when
the
cubelet
config
was
alpha,
and
so
at
that
point
we
couldn't
use
it
like,
we
couldn't
generate
it
from
from
a
config
file.
Is
that
correct
when
it
was
on
alpha.
A
We
used
it
in
in
some
tests
like
in
some
of
the
no
dd
tests,
because
it
was
convenient
for
some
of
those
things.
But.
B
A
A
That's
using
angle
brackets,
instead
of
colons
to
separate
keys
and
values
right
there,
all
these
kinds
of
things
that
in
yaml
you
would
just
want
to
use
like
a
normal
list
or
a
map
to
represent,
and
so
a
big
part
of
it
was
sort
of
writing
like
like
a
lot
of
the
parsers
for
the
flagstaff
were
not
actually
like.
Just
in
the
flag,
parsing
part
of
the
code
base.
A
They
were
like
buried
way
down
but
like
the
strength,
value
actually
gets
consumed,
so
part
of
it
was
like
lifting
all
of
those
out
into
like
here's,
this
flag
type
and
here's
that
flag
type
that
parses
this
thing
and
then
parsing
it
directly
into
like
a
map
in
the
config
object.
Instead
of
sort
of
that
really
really
really
late
parsing
and
that
part
of
it
was
lifting
validation
out
right
because
validation
always
happens
after
parsing.
So
a
lot
of
the
validation
was
really
buried.
A
Deep
in
the
code
too,
and
supported
was
lifting
that
out
into
like
the
normal
config
validation
function,
then
there
was
all
the
global
flag
stuff
as
well,
so
that
was
when
we
discovered
that
there
we
had
all
these
third-party
libraries
that
were
just
registering
global
flag
values
on
that
sort
of
global
command
line
and
then
we're
like.
Oh,
this
is
bad
because
random
libraries
can
add
stuff
to
our
command.
A
May
or
may
not
support
so
that
wasn't
getting
all
that,
like
global
flags
go
stuff
so
that
those
had
to
be
explicitly
registered
and
we
were
no
longer
going
to
parse
just
the
global
command
line
flag
set.
So
that
was
another
thing
that
I
think
you
might
want
to
double
check
what's
in
controller
manager
and
make
sure
it
is
like
explicitly
doing
that.
But
I
think
things
should
be
explicitly
doing
that
at
this
point.
A
So
that's
probably
not
as
big
of
a
amount
of
work.
What
else
a
lot
and
then
a
lot
of
it
was
just
like
trying
to
migrate
as
many
things
as
possible
before
moving
there,
and
so
there
were
things
that
were
already
deprecated.
Then
we
left
those
out.
There
were
the
instance
specific
things
like
host,
name,
override
and
node
ip
that
we
weren't.
There
was
some
discussion
around
like.
A
Oh,
maybe
we
could
like
be
smart
and
auto,
detect
these
and
get
rid
of
them
all
together
and
that
that
kind
of
stalled
things,
I
think,
probably
the
real
solution
to
that
is
just
to
like,
have
two
config
files,
and
there
was
there
was
somebody
who
signed
up
to
write
a
cup
for
that.
D
A
D
Oh
okay,
yeah
there
was
one.
There
is
one
comment
I
think
in
the
cape
that
it
somebody
mentioned
it
would
make
actually
difficult
for
you
know
starting
multiple
binaries,
which
cubecontrol
manager
kind
of
manages
like
cloud
controller
manager,
but
that's
obsolete
now,
but
somebody
mentioned
that
it
single
like
the
config
files,
when
we
are
basically
trying
to
merge
everything
into
a
single
config
file,
maybe
two
now
to
make
things
difficult
to
kind
of
have
different
configuration
for
different
binaries.
D
A
Yeah,
that's
a
that's
and
alex,
and
we've
talked
about
that
a
bit
too
like
there's
this
overarching
issue
of
when
you
have
a
controller
manager
style
binary
host
right.
This
binary
is
a
host
for
a
bunch
of
other
controllers.
A
It
abstracts
away
some
things
for
them
and
then
it
adds
some
efficiency
because
they
can
share
memory
for
things
like
informers.
A
A
B
A
I
want
to
start
with,
like
a
high
level
description
of
what
what
needs
to
be
done
in
order
for
us
to
be
confident
enough
to
move
controller
managers
config
to
beta.
So
it
would
be
like
it's
possible
to
serialize
the
config.
If
that's
still
a
problem,
you
know
hidden
structures
have
been
extracted
into
like
more
typical
vml
structures.
A
We
are
we've
addressed
any
backwards.
Compatibility
concerns
with
migrating
from
command
line,
5
to
configuration,
and
that
could
mean
we
we
just
say:
hey,
you're
only
allowed
to
use
one
or
the
other.
It
could
mean
we
implement
like
more
complex
flag
to
config
mapping
logic
that
you
know
enables
command
lines
to
to
be
interleaved
with
configuration,
and
I
think
there's
there's
like
maybe
another
question
of
like.
Are
these?
You
know
the
apis
that
we
really
like
it's
like
one
thing
we
do.
A
We
did
in
the
cubelet,
because
migrating
to
config
is
technically
crossing
a
version
boundary.
You
can
change
defaults
of
things
that
had
bad
defaults
and
stuff
like
what
that
is
an
option,
but
it
will
slow
down
your
work.
If
you
go
down
that
rabbit
hole
of
like
trying
to
discuss
what
the
best
shape
for
every
existing
option
is,
and
I
think
it
will
be
faster
to
get
to
component,
I
think
if
we
don't
try
to
redesign
the
whole
api.
A
And
so
I
I
would,
what
I
would
do
is
I
would
write
a
short
document.
Maybe
one
page,
or
maybe
two
pages
on
like
here
are
the
things
that
need
to
be
true
for
us
to
move
to
beta
and
then
make
that
a
like,
just
like
there's
like
move
q
proxy
to
v101
cap
like
write
a
move
controller
manager
to
p1
beta1
cap.
Here,
all
the
things
that
we
need
to
do,
and
that
can
be
our
template
for
like
managing
this
work.
B
All
right
so
like
what
would
be
some
good
like
a
lot
of
these
things
you
mentioned,
I
still
need
to
look
like
into.
I
don't
really
know
how
how
does
work
and
work
together.
So
is
there
like
any
like
previous
prs
or
documents,
like
you
mentioned,
the
the
cubelet,
you
know
which
describe
these
kind
of
things
when
someone
probably.
A
Probably
the
best
document
that
covers
all
of
that
is
that
version
component
configuration
file
stock,
which
has
the
section
where
it
describes
kind
of
everything
we
did
to
the
cubelet.
But
I
want
you
to
read
that
with.
I
can
send
you
a
link
to
that
as
well.
Here.
A
D
Just
to
confirm
so
I'll
I'll,
so
orbitech
is
you
right,
alexander
yeah
yeah.
It's
me
exactly
yeah.
The
names
are
different,
so
I
think.
A
So
I'm
gonna
share,
which
I
got
working.
A
So
this
is
the
dock
and
it
basically
goes
through
all
those
problems
like
the
global
flags
thing.
We
already
have
api
groups
created
for
most
of
these
now,
so
that's
not
as
big
a
problem.
A
A
So
it's
a
little
bit
different
and
we
might
have
to
do
some
problem
solving
around
the
best
way
to
kind
of
map
between
that
and
the
the
component
config.
And
I
think
that
might
be
the
reason
that
cube
proxy
really
wanted
to
just
say
use
one
or
the
other
with
cuba.
It
was
pretty
easy
to
have
like
a
direct
mapping
from
flags
to
the
config
because
they
were
just
these
simple
structs,
so
you
might
have
to
change
that.
C
A
And
remember
we
have
time
too,
because
we'll
be
we'll
be
off
for
three
weeks.
You
get
the
time
to
kind
of
get
your
heads
around
this
and
get
a
plan
together,
but
that's
the
main.
A
D
Yeah
I
was,
I
was
under
the
impression
that
the
serialization
is
the
biggest
problem
coming
to
the
first
problem
we
need
to
tackle,
but
it
seems
it
is
resolved.
D
Cool
so
alex
are
you:
are
you
familiar
with
the
code
base
of
control
manager.
B
B
You
also
mentioned,
like
a
queue
controller.
Just
so
for
understanding
cube.
Controller
manager
is
like
a
binary
host
right.
It's.
A
A
So
the
reason
to
do
that
is.
A
The
reason
to
do
that
is
that
the
main
reason
is
that
all
those
controllers
rely
on
informers
to
get
updates
from
the
api
server,
and
you
can
collapse
it
down
to
a
single
informer
cache,
that's
shared
among
all
the
controllers,
if
you
host
them
in
the
name
by
an
area,
and
that
saves
a
lot
of
memory.
D
A
A
In
this
case,
this
struct
is
just
like
embedding
the
component
config
struct
looks
like
the
internal
type
and
reading
the
flags
into
that
so
yeah.
I
think
it's
definitely
worth
kind
of
sketching
out
with
what's
actually
happening
here,
which
is
different
from
the
keyboard.
A
A
A
A
There's
this
map
of
all
of
these
controller
initializers.
A
All
of
this
information
is
what's
sort
of
common
between
different
subcontrollers.
A
lot
of
it
is
kind
of
runtime
information,
so
like
your
plumber,
but
then
there's
also
this
component
field,
so
the
that
I'm
guessing
is
just
all
of
them
together
and
so
maybe
there's
a
better
way
to
divide
what
each
controller
can
see.
C
A
Everything-
and
that
is
embedding
the
internal-
this
is
an
internal
type,
embedding
the
internal
types
of
all
of
these
other
ones.
I
think,
when
we
talk
about
breaking
it
up
into
separate
sub
structures,
what
it
means
is,
instead
of
having
this
one
giant
superstructure
controller
manager
configuration
that
is
hosting
the
config
for
every
single
one.
A
B
Yeah,
I
think
that
was
the
one
of
the
comments
don't
mention
tonight
or
amateur
of
possible
or
stefan,
if
that
is
like
practicable
ryan.
A
A
For
the
you
know-
and
it's
kind
of
it's
kind
of
up
to
you
guys,
because
it's
like
how
much
work
do
you
want
to
do
up
front
because
we
could
split
it
up
and
we
could
implement
multi-doc
support
so
that
it's
still
possible
to
write
them
all
in
a
single
file.
And
you
know
slurp
them
all.
In
from
one
place,
we
could
figure
out
how
to
how
to
how
to
map
them
to
the
controllers.
That
they're
meant
for,
like
one
option,
was
to
key
on
the
kind.
A
A
You
know
at
least
a
v1
alpha
2
with
that
and
kind
of
get
everything
else
in
place
where
we
can
say
now.
Our
choices
move
to
beta
with
what
we
have
or
break
this
up
before
we
move
to
beta,
but
there's
so
many
other
problems
to
kind
of
solve
outside
of
the
like.
How
do
we
put
these
up
that
I
think
it's?
It
might
be
worth
solving
those
first,
if
you
want
it
to
be
a
little
more
expedient.
D
I
think
for
me
it's
like
I
want
to
like
start
doing
something
so
that,
because
once
I
you
know
get
used
to
the
code,
then
I
can
be
more
confident
with
respect
to
what
I
can
do
and
now
that
I'm
not
that
experienced
and
not
I'm
not
confident
that
how
much
I
can
take
it
and
how
much
I
can
do
it.
A
B
D
A
B
Yeah,
I
think
I
think
making
it
serializable
would
be
a
good
first
step
just
to
like
get
to
know
how
the
stuff
works
and
what
steps
are
necessary.
You
know,
and
just
like,
dig
through
all
the
different
individual
conflicts
of
all
those
controllers
and
just
see
what's
going
on
there.
So
I
think
this
could
give
some
confidence
and
once
we
know
what's
going
on,
maybe
we
can.
We
have
a
better
image
of
you
know.
Is
it
really
like?
A
Right
what
I
want,
what
I,
where
I
want
to
get
to
is
where
we,
especially
you
guys,
understand
the
problem.
Well
and-
and
you
know
like
the
specifics
of
how
it
actually
works,
and
then
you
can
make
an
argument
either
way.
It's
what's
important.
Is
that
you're
able,
on
your
own
to
make
that
argument.
B
Cool
all
right
I'll,
definitely
read
the
doc.
You
posted
on
the
cubelet-
and
you
know
like
with
that
in
mind,
like
look
at
the
look
at
the
controller
manager
and
then
probably
like.
While
I
read
this
I'll,
take
some
notes
and
some
questions
will
come
up
and
we'll
probably
meet
again
here
with
some
more
info
on
some
more
questions.
A
Yeah
sounds
good
I'll,
be
I'll,
be
on
vacation
the
next
three
weeks,
but
I'll
still
be
periodically
dropping
into
slack
to
see.
What's
going
on,
yeah
no
worries.
B
Sorry
so
connect
your
names.
Can
I
call
you
lala
yeah.
B
All
right,
yeah
nice,
so
I
mean
yeah.
What
we
can
do
is
like
you
know
we
just
we
just
read,
read
this
stuff
and
maybe
we
stay
in
touch
and
I'll
take
some
notes.
Maybe
you
can
take
some
notes
and
we
compare
them
and
see.
If
we,
you
know,
we
can
come
up
with
some
some
ideas
on
how
to
move
this
forward,
and
maybe
we
can
create
a
shared
document
on
what
our
ideas
are,
and
you
know,
exchange
a
little
bit.
B
It's
it's
yeah.
I've
been
I've,
been
taking
a
couple
of
notes
during
this
talk.
It's
if
you
look
in
the
meeting
notes,
it's
linked.
A
Make
sure
you
guys
include
at
tbh
also
yeah
in
those
discussions.
A
I,
for
I
think
his
github
handle
is
different
from
his
slack,
his
slack
but
they're
both
they
should
both
be
in
that
sign
up
sheet
from
way
back
when
yeah
he's.
D
So
I
think
that
the
challenge
with
this
one
is
because
it's
a
lot
of
history
in
this
effort,
so
there's
a
lot
of
unwanted
information.
I
think
so.
I
think
we
can
just
cut
the
crap
and
just
say:
okay,
these
are
the
things
we're
gonna
do.
These
are
three
things
three
or
four
things,
not
much
more
than
three
or
four
things.
Let's
finish
this:
let's
do
it
first
one
thing
at
a
time
and
because
when
I
was
trying
to
read
all
this
document,
there's
so
much
of
documentation
like
sometimes
it's
overwhelming,
like
yeah.
A
It's
just
what
happened
and
you
can
make
a
good
decision
without
necessarily
knowing
all
of
it
and
other
people
will
chime
in
with
their
opinions.
So,
like
don't,
don't
feel
like
you're
going
to
miss
something
because
people
will
show
up
to
tell
you
what
they
think
is
important,
and
it's
much
better
that
somebody
be
paying
attention
to
this
and
have
an
opinion
and
be
driving
progress.
A
Then,
like
everyone
waiting
for
some
sort
of
consensus,
there's
really
there's
really
no,
like
you'll,
see
a
lot
of
things
in
kubernetes
that
are
sort
of
claimed
to
be
driven
by
consensus.
But
under
the
covers,
it's
always
like
one
or
a
few
people
like.
Actually,
you
know
trying
to
build
that
consensus.
So
yeah.
Everyone
eventually
agrees
to
something,
but
somebody
is
going
around
and
having
the
discussions
with
them
and
saying
hey
like.
Would
you
be
able
to
compromise
on
this
or
that
in
order
to
put
like
the
overall
package
together?
A
B
Yeah,
that
sounds
good.
I
think
the
first
first
part
is
like
understanding
the
problem
well
enough,
yes,
which
I
think
we'll
need
to
to
do
some
reading,
and
you
know
just
looking
things
asking
some
more
questions
and
then
on
the
way
like
trying
to
come
up
with
some
some
points
we
identify,
which
can
be
good
for,
like
attacking
first.
A
Yeah,
perfect,
okay
and
then,
and
then
the
only
other
thing
is,
you
know,
try
and
try
and
stay
in
touch
with
people
who
are.
You
know
either
signed
up
to
solve
some
of
the
other
sub
problems
like
if
you
just
take
a
look
at
that
mentorship
sheet.
If
you
see
that
you're
you're
so
like
instant,
specific
config
might
be
an
example,
if
you
see
you're
solving
a
problem
somebody's
signed
up
for
on
there
like
ping
them
and
try
to
make
sure
you're
in
line
with
whatever
they
were
thinking.
A
Boss
ross
has
his
move,
cube
proxy
debata
kept
as
well,
so
that
might
be
worth
taking
a
quick
look
at
nice.
D
Okay,
so
yeah,
I
think
that's
it
one
once
one
such
one
suggestion
from
for
alex
is
so
the
word
document.
I
I
just
modified
one
line,
but
then
I'll
next
time
I
will
do
only
suggest
so
that
only
I,
the
idea
is,
you
know
when
I
do
suggest
some
modification.
You
will
get
to
know
what
I'm
trying
to
say
and
you
can
accept
it,
and
you
can
actually
suggest
that
I
I
basically
come
to
know
what
kind
of
changes
you're
doing
to
document.
D
So
it's
basically
it's
kind
of
mental
map
of
what's
what's
changing
in
the
document.
Basically,
I
think
will
help
both
of
us.
B
Oh
yeah
yeah
sure
awesome.
We
can
use
this
document
to
to
move
forward
and
just
organize
this.
D
B
So
far
so
good,
I
think
we're
we're
set
just
so
I
mean
maybe
regarding
kubecon,
you
mentioned
that
the
the
maintainer
track.
It's
it's
like
different
right,
it's
on
in,
like
january,
where
this.
A
Yeah,
so
I
emailed
nancy
asking
what
we
should
do
for
maintainer
track
talk,
because
it
said
the
rules
were
separate
and
she
said:
don't
use
the
cfp
wait
until
january
and
she's
gonna
send
an
email
out,
so
we'll
wait
until
january
for
that
and
then
yeah
alex.
I
think
you
wanted
to
maybe
co-present
right
so.
B
I
I
would
love
to
yeah
I
I
would
really
like
to.
I
just
do
why
I'm
asking
again
is
because,
like
my
company
is
now
like
organizing,
like
travel
plans
and
tickets,
so
I'm
I'm
kind
of
like
holding
off
on
that.
I'm
saying
like
you
know
I
might
be
on
a
speaker
ticket
and
I
have
some
friends
in
amsterdam
which
I
can
stay
so
there's
no
problem
just
just
so.
A
You
know
I'll
give
you
a
refund
if
you
get
made
into
a
speaker
after
you've
purchased
a
ticket.
So
don't
don't.
A
D
B
B
Cool
yeah
then
yeah.
Well
as
soon
as
that
comes
up,
we
can
have.
We
can
talk
again
about.
You
know
how
we
structure
it.
Maybe
you
can
guys
can
come
down
as
well.
We've
been
nice,
you
know
to
meet
up.