►
From YouTube: Kubernetes KubeBuilder Meeting 20200409
Description
KubeBuilder Meeting for 2020/04/09. See https://sigs.k8s.io/kubebuilder for more details.
A
Okay,
we're
recording
welcome
to
the
q
builder
controller
runtime
community
meeting
today
is
april
9th
I'm
eric
strzinski,
I
will
be
hosting
today.
So
currently
there
are
no
agenda
items
chris.
You
seem
to
be
adding
one
right
now,
so,
while
you're
doing
that
I'll
just
give
a
quick
update
on
plugins
work
in
queue
builder,
so
I'm
sure,
as
many
of
you
are
aware,
I've
talked
about
this
once
or
twice.
A
A
There
is
one
more
thing
that
needs
discussion,
but
it's
something
that
can
be
done
alongside
the
pr
itself
to
master.
So
after
this
I
will
post
a
link
in
the
agenda
to
that
pr
and
and
also
in
the
kubernetes
operator,
slack
channel.
So
if
you
get
a
chance,
please
take
a
look
through
that
test.
It
out
give
feedback.
B
Yeah
put
in
this
pull
request
a
couple
days
ago.
I
thanks
david,
I
think
for
taking
a
look
at
it
so
far,
I
poked
a
couple
other
folks,
if
you
could
just
give
a
review
of
that,
I'm
looking
for
I'm
looking
for
feedback
right
now,
specifically
on
the
ux
of
the
implementation,
because
it
differs
slightly
from
what
we
talked
about
in
the
the
actual
implementation.
Do
you
mind
clicking
on
that
from
your
screen
or
yeah?
B
Just
so,
you
can
kind
of
see
what
I'm
talking
about,
but
basically
there's
there's
two
different
paths
for
implementing
this:
that
we've
discussed
during
the
design
review
in.
In
essence,
what
I'm
looking
for
is
some
feedback
on
the
user.
Experience
for
implementing
component
config
retrieving
an
option
struct
for
configuring,
the
manager,
whether
that
be
updating
or
creating
a
new
one
from
scratch
and
then
mutating
it
after
the
fact
yeah.
So
I'm
looking
for
feedback
on
that
on
the
user
experience
and
then
yeah.
If
anybody's
interested.
B
Yeah,
so
this
is
moving
us
towards
following
the
community
standards,
around
management
for
components
within
kubernetes,
so
using
component
config
style
configurations
for
controllers
built
out
of
or
runtime.
So
this
takes
the
current
implementation
of
creating
a
new
manager
where
you
pass
in
an
option
struct
into
it
and
provides
a
way
to
have
a
common,
a
common
interface
that
every
that
that
configurations
can
get
pulled
out
of
yeah
just
makes
it
easier
to
configure
controllers
using
files
instead
of
flags.
B
Yeah
so
things
like
within
the
that
at
the
end,
like
the
last
parameter
passed
into
this
function,
that
controller
v1
alpha
one
that
default
controller,
config
you'll
notice
in
there
that
that
standardizes,
the
a
type
that
you
can
load
in
for
pulling
out
options
out
of
a
file,
and
it
has.
B
If
you
go
under
files,
you
can
actually
see
what
that
struct
looks
like,
but
it
basically
references
a
component
base
style
or
component
base
struct
to
pull
out
the
leader
election
configuration.
So
those
are
standardized
around
the
same
ones
that
we
use
for
cubelet
and
cube
api
server
and
and
the
scheduler.
C
I
have
one
question
around
that
so
there's
there
are
some
options
that
can
be
changed
during
runtime
and
there
will
always
be
some
options
that
can
not.
Does
the
component
config
somehow
differentiate
among
these,
or
is
the
idea
to
just
have
it
loaded
once
during
startup
and
then
never
reload?
It
again.
B
B
I
think
long
term
we
could
look
into
having
this
be
able
to
be
like
dino
dead,
the
dynamic
public
config,
but
there's
obviously
some
things
that
would
need
to
be
reworked
out
like
how.
How
would
you
reconfigure
a
client
it
just
wouldn't
really
work,
so
we
could
look
into
it
after
the
fact.
I
think.
C
B
Yeah
agreed
agreed,
and
I
think
that
would
be
a
fun
addition.
Addition
to
this.
My
word
is
over
complicating
the
first
implementation,
especially
because
controller
authors
might
not
realize
that
that
would
be
the
the
case
if
we
supported
that
out
of
the
box
as
like
a
default
flow,
so
yeah
also
you'll
notice
in
there
that
the
the
design
and
implementation
differ
from
like
this
pull
request.
118
that
you're
looking
at
it
differs
slightly
from
there.
So
a
plan
is
like.
B
I
want
the
feedback
on
the
on
the
user
experience
and
then,
before
this
pull
request,
gets
accepted,
I'll,
go
and
update
the
actual
design
dock
to
implement
it
in
case
somebody
calls
that
out
just
wanna
make
sure.
A
If
but
there's
nothing
else
well,
there
is
nothing
else
in
the
agenda.
So
if
there
are
no
other
questions
or
comments,
concerns
anything
give
you
another
43
minutes
back
for
your
day,.
B
So
I
this
might
not
be
the
right
channel
to
ask
this,
but
I'm
trying
to
actually
gather
some
information
that
I
don't
know
the
underlying
origins
of,
but
I've
been
looking
at
a
lot
of
implementations
of
controllers
where
scheme
is
is
configured
and
and
added
to
and
a
lot
of
the
like
base
com,
client
go
implementations,
all
use,
util
runtime
must
and
which
causes
things
to
panic.
If
the
like
add
to
scheme
doesn't
work,
I'm
not
sure
if
I'm
explaining
this
properly.
B
But
it's
interesting
because
I'm
looking
in
like
in
qbuilder
and
out
of
box,
it
doesn't
do
anything,
it
just
ignores
all
errors
and
I'm
curious
what
the
implementation
is.
There
there's
a
pull
request.
I
have
open
in
in
qbuilder
specifically
to
add
this
to
the
scaffold
and
I'm
curious
if
it
was
like
a
a
specific
reason.
We
don't
do
that.
B
Cool
yeah
in
the
in
in
the
agenda,
I
just
dropped
in
a
link
to
the
other.
Pull
requests
that
I
have
open.
Yeah
mostly
just
saw
that
going
across
other
projects
and
figured
we
could
have
a
conversation
about
it.
A
Sure,
because
I
think
it's
appropriate
to
check
errors
if
there
are
actual
errors
being
returned
next.
D
B
Okay,
I
don't
think
it's
easy
to
disagree
with
that
yeah.
I
didn't
know
if
there
was
if
there
was
like
some
long-term
history
behind
some
things
causing
errors
like
if
it
was
whether,
when
you
were
registering
like
conversion
web
hooks
or
anything
like
that,
like
that
we're
not
doing
within
cube
builder,
at
least.
C
Right
so,
okay
cool-
I
don't
know
about
the
historical
thinking,
but
the
whole
scheme
is
a
map
underneath
and
unless
someone
has
changed
that
within
the
last
six
months
or
something
it
doesn't
have
any
protection
against
concurrent
additions.
C
B
Okay,
that
makes
sense
yeah,
because
I
saw
I
saw
the
only
case
that
I
could
get
it
to.
Panic
was
basically,
if
I
re-registered
two
schemes
or
registered
like
a
crd,
that
had
the
same
name
as
a
common
type
in
or
a
standard
gvk
in
kubernetes.