►
From YouTube: Kubernetes KubeBuilder Meeting 20200326
Description
KubeBuilder Meeting for 2020/03/26. See https://sigs.k8s.io/kubebuilder for more details.
A
A
B
Sure
how
do
you
want
me
to
share
the
screen
wrong?
Yeah
go
ahead.
Let
me
find
this
one.
Anybody
see
a
little
bit.
B
Okay,
so
that
yeah,
the
fish'
PSA
is
just
like.
We
have
two
new
releases
controller,
run
time:
zero,
five
two,
which
was
cut
today
and
you
have
two
eight
zero.
Five
two
two
has
a
new
helper
to
configure
longer
options
via
P
flex.
There's
two
bug
fixes
one
is
a
break
and
changing
M
test,
which
we
didn't
mean
to
to
add
so
and
that
got
fixing
846
and
then
the
delay
in
the
rev
limiter.
I
think
this
is
only
for
dynamic
restaurant
program
memory
correctly,
yes
and
then
zero
to
eight
has
a
new
resource
memory.
B
Infield
in
the
are
back
marker
and
there
is
a
new
bit
of
bug
fix,
which
was
actually
like,
also
breaking
change,
which
we
didn't
meet
when
the
policy
is
not
set.
That
didn't
like
a
look
at
the
other
release.
We
have
match
policy
now
in
the
web
book
scaffolding
and
you
can
set
it
to
I,
think
exact
or
equivalent
and
yeah.
That's
all
so.
I
have
any
questions
and
it
releases.
A
A
A
A
C
C
Fur
plugins
that
have
been
added
on
top
of
key
builders
base
plugins,
and
so
that
could
also
take
some
confuse,
but
over
the
next
probably
next
week.
I
think
that
the
feature
branch
will
be
ready
for
PR
against
master
and
if
we
could
get
as
many
reviews
on
that
as
possible.
That
would
be
great
because
this
is
a
fairly
big
change
to
that
plumbing
of
key
builder.
C
Caveats
or
drawbacks
that
you've
encountered
and
post
about
them,
because
we
want
this
to
be
as
robust
as
possible
for
people
to
be
creating
plugins
and
extending
to
builder
as
they
see
fit
and
also
going
forward.
There
will
be
a
phase
to
proposal
submission
that
deals
with
pipelining
plugins
together
so
taking
the
bay
scaffolding
of
KU
builder
and
extending
it
before
writing
it
to
disk
or
being
able
to
modify
scaffolds
in
memory
and
then
commit
them
to
disk.
C
D
So
there's
this
interesting
issue
we
ran
into
where
we
tried
to
have
to
try
to
create
a
type.
That's
like
a
map
of
up
conditions,
basically
and
if
you're
aware
of
what
conditions
are
and
in
communities
they're
there
arrays.
So
what
we
want
to
do
is
have
a
map
that
is
keen
on
the
condition
type
and
then
we
can
marshal
it
out
to
an
array
and
make
dealing
with
conditions
a
lot
easier,
because
we
want
set
based
semantics.
D
So
one
thing
we
ran
into
and
controller
tools
is
that
when
we
try
to
sort
of
control
and
tools
added
the
type
of
override
in
the
CR
d-generation
annotation-
and
it
turns
out
that
when
you
use
that
in
this
case,
what
happens
under
the
hood
is
that
CRD
that
gets
generated
you
get
type
equals
array.
But
then
the
other
field,
in
the
validation
section,
don't
line
up
with
how
an
array
is
supposed
to
look
so
I'm
curious.
What
we
want
to
do
about
this.
D
It
seems
like
for
the
particular
case
that
I
have
I
think
there's
a
potential
fix
that
we
can
make,
but
it's
just
like
the
one
of
many
many
possible
permutations
of
going
from
one
type
to
a
different
type
and
try
to
deal
with.
All
of
those
seems
a
little
crazy,
but
I
don't
know
if
anybody
has
seen
this
issue
or
knows
what
I'm
talking
about,
but
I'm
interested
in
solving
this
one
way
or
another.
So.
A
D
A
We
have
a
whole
class
of
issues
that
is
basically
things
with
custom.
Jason
serialization
are
really
hard
to
make
it
actually
work,
and
so
I
think
this
falls
under
that
class
of
things.
Let
me
see
if
I
can
find
the
issue,
because
there's,
like
other
similar
things
with
like
types
that
we
don't
support
or
that
have
custom
jason
serialization,
but
don't
have
like
mark
so
don't
have
like
have
private
fields
that
don't
have
markers
on
their
pub
or
not
markers
tags
on
their
public
fields.
A
Really
well
yeah
yeah.
Here
we
go
I
found
the
I
will
drop
the
issue
into
meeting
notes.
A
Do
the
equivalent
of
a
market
type
as
being
treated
as
opaque
by
our
conversion
system?
We
should
probably
do
that
automatically
if
there's
a
custom
serialize.
If
there's
a
custom
like
marshal
Jason
on
marshal
Jason,
which
we
can
detect
and
controller
tools,
we
can
say
if
someone's
implemented
Marshall
and
on
marshal
Jason
like
we
really
can't
say
anything
about
Jason
schema
and
then
probably
a
corresponding
marker.
That's
just
like
alright,
no
holds
barred,
specify
the
raw
Jason
schema
or
something
because,
like.
A
There
there's
there's
stuff
that
you
want
to
do
in
those
cases
where
we
don't
really
generally
want
to
expose
a
marker
for
them,
and
so
it's
probably
faster
than
figuring
out
exactly
what
markers
in
semantics
we
want
to
expose
to,
like
specifically
in
that
case,
allow
people
to
just
specify
custom
jason
schema.
Do.
D
You
think
that
so
in
this
case
like
we
have,
we
have
like
a
validation
field
that
says
I
expect
to
see
an
array
of
conditions
and
then
there's
a
bunch
of
items
that
have
this
other
schema.
I
would
be
nice
and
I.
Don't
know
if
you're
suggesting
this
or
not,
but
it
would
be
nice
if
the
underlying
element
type
like
all
that
still
works,
just
fine,
it's
just
the
container.
That's
causing
problems
right
like.
E
D
A
A
Specifically,
for
your
case,
this
type
is
not
going
well.
That's
not
true,
specifically
for
this
case,
you
sort
the
array
yelling
yeah,
we
sort
yeah
all
right
cool,
just
wanted
to
make
sure
so
for
people
listening
in
that
may
may
not
be
clear
why
I
asked
that
kubernetes
types
and
an
extension
types
really
are
supposed
to
be
what's
called
round
trip
of
all,
which
means
that
you
can
deserialize
them
into
a
corresponding
go
type
and
then
ricci
realize
them
back
into
jason
and
assuming
that
the
go
type
has
all
the
fields
to
deserialize
it.
D
D
D
D
Yeah
here
we
go
this
one
I,
don't
know
as
much
about
personally.
This
came
up
during
one
of
our
SDK
triage
call,
I
think
it's
related.
It's
basically
a
similar
like
we're
trying
to
do
something
with
this
type.
That's
not
quite
getting
parsed
correctly
the
controller
tools.
It
looks
like
there's
some
activity
recently,
so
the
main
reason
I
have
this
on
the
agenda
is
just
see.
If
anyone
here
knows
what
the
status
is
or
is
working
on
this
or
who
knows
what
we're
gonna
do
with
it.
A
So
I
can
I
can
give
some
background.
The
main
reason
why
it's
paused
right
now
is
because,
like
in
general
maps,
aren't
something
that
we
encourage
people
to
use
in
kubernetes
and
are
mostly
in
a
state
where,
unless
you're
doing
like
one
of
two
specific
types
of
maps,
you're
really
not
supposed
to
have
been
using
them
in
the
first
place,
according
to
the
kubernetes
api
conventions,
and
so
like.
This
is
one
of
the
cases
where
you
are
probably
not
supposed
to
be
using
them.
A
So
I've
been
saying:
no
I,
don't
think
we
should
merge
this,
because
I
want
to
I.
Generally
philosophy
has
been
to
try
to
minimize
the
amount
of
things
that
produce
an
invalid
like
an
unwrapped
non
recommended
Fabrice
schema,
so
that,
like
you,
know,
people
don't
hopefully
looking
at
ID
people
towards
building
the
Wrights
schemas
with
with
controller
tools.
D
D
A
If
you
need
anything
more
complicated
than
a
a
map
from
string
to
string,
which
is
like
basically
the
only
allowed
type
map
from
string
to
string
and
technically
map
from
string
to
string
list
for
some
legacy
cases
yeah,
you
should
be
be
using
a
a
indexed
list.
I
will
loosely
call
it,
which
is
which
is
slices
with
a
particular
key
like
key
field,
basically
called
name,
and
we
have
semantics
and
reply
and
stuff
like
that
to
handle
those
cases
as
as
kind
of
ordered
maps
and
that's
most
of
what
we
do
in
kubernetes.
A
A
A
F
F
F
A
G
G
A
A
A
little
scary
that
the
et
test
did
not
didn't
catch
that
it's
possible,
so
I
have
like
immediate
theory.
Is
that
like
go
imports,
overzealous
behavior
on
inserting
and
removing
imports
inserted
an
import
on
some
systems
or
something
that
fixed
it
like
where
it
could
find
and
test
it
inserted
and
chest,
and
if
it
can't
find
out
tests
that
it
doesn't
answer
em
test
because
we
run
go
imports
to
sort
and
filter
I
think
was
yeah.
We
still
do
that
what
it
still
does,
yeah
yeah.
So
sorry
there
was
at
some
point.
A
G
A
F
Then
I
also
have
the
next
one
as
well,
which
is
just
a
call
for
some
more
reviews
on
this.
Thanks
for
insert
pokin
in
there
today,
I
think
I
saw
a
comment
that
got
in
there
just
looking
for
some
finer
reviews.
It
seems
like
most
most
of
it
is
unless
the
turn
has
gone
away
and
it
looks
like
we're
kind
of
centering
on
something.
F
So
if
we
can
get
some
extra
extra
eyes
on
this,
just
to
make
sure
it's
okay
and
then
move
it
forward,
hopefully,
since
it's
not
entirely
a
break
or
it's
not
a
breaking
change
at
all,
it
would
just
be
a
new
feature
and
I.
Don't
know
if
there's
a
best
practice
if
I
should
be
resolved
in
conversations
and
things
like
that,
I've
been
avoiding
after
right
now,
most
part.
A
F
A
That's
fine
to
leave
them
open
if
you
want,
and
definitely
as
a
room
as
just
like
a
general
message
for
people.
If
there
is
stuff
like
that's
really
really
urgent,
like
feel
free,
if
it
seems
like
you're
being
even
if
it
seems
like
you're
being
annoying
or
something
like
you're,
not
if
you
have
to
ping
us
like
every
day
and
be
like
by
the
way
like
there's
a
breaking
change
on
master
there's
a
breaking
change
in
a
release
like
that
is
a
fine
thing
to
like
be
like.
We
need
to
fix
this
now.
B
If
there
are
no
that
are
done
items,
we
do
backlog
grooming
going
once
twice
three
times
so
forth
today,
I
think
we'll
just
do
control
the
run
time.
I've
briefly
look
at
control
of
tools,
but
there
is
44
open
and
they're
pretty
much
like
the
whole
backlog
doesn't
have
done
anything.
So
we
probably
should
take
that
synchronously
and
yeah
and
figure
that
out
but
or
schedule
a
new
backlog
grooming.
B
We
have
24
open
in
control
at
one
time.
If
that's
okay,
would
you
all
Nicola
I'd
like
to
start
with
it's
your
18?
Well,
in
the
118
two-brains
dependencies
update,
there
was
some
chatter
about
like
opening
master,
2:06
and
updated
dependency
to
118,
0
18,
so
I,
do
you
think
we
should
just
like
open
master
right
now
for
changes
and
like
target
118
4
0,
6
I.
B
B
B
A
A
B
A
Yeah
yeah,
we
definitely
I
think
we
need
more
investigation.
E
B
A
Without
being
a
breaking
change,
which
is
that
we
like
use
information
from
the
manager
to
construct
the
host
field
yeah
because
I'm
assuming
this
is
a
case,
I
like
probably
need
to
ask
if
this
is
a
case
where
you
want
to
identify
which
controller
it
was.
But
if,
if
that's
the
case
like
we
can
probably
like
use
information,
we
already
have
a
like
leader
election
key
or
something
like
that.
A
Oh,
that
is
no
name
on
which
the
event
is
generated.
Yeah
interesting,
I,
don't.
B
A
I'm
curious
as
to
what
the
semantics
of
that
are
in
kubernetes
I.
Don't
actually
remember,
I
wonder
if
it's
only
used
for
pod
events
and
similar
like
events
that
are
inherently
associated
with
the
node
or
if
it's
or
if
it's
used
to
distinguish
like
which
is
the
active
leader
for
like
cube
controller
manager
or
what's
going
on
there,
that
yeah
I
think
that
means
more
investigation.
B
A
Yeah
this,
this
isn't
like
you
shouldn't,
be
detecting
this
based
on
on
kubernetes
version.
You
should
be
detecting
it,
based
on
like
what's
available
in
the
rest.
Mapper
like
this
is
what
discovery
is
like
you
should
be
saying
is:
is
fee
1
beta,
1
deployment
available
in
discovery?
Not
not
I,
I
know
a
priori
that
that
fee
1,
beta
1
is
not
available
in
1/16,
or
something
like
that.
No.
B
A
B
E
B
A
B
B
B
A
B
A
B
A
Yeah
they're
handling
updates
quickly
enough
that,
like
multiple
small
updates,
cause
multiple
reviews
and
they
want
to
not
they
want
to
not
do
that,
basically
I
think,
but
we
should.
We
should
ask
and
clarify
on
that.
Oh
you
mean
even
per
like
individual
and
I'm,
not
per
like
multiple
items.
Yeah
I
think
that
make
something
like,
even
if
there's
10
updates
in
this
time
period,
only
ever
run
reconcile
once
every
10
seconds.
A
I
B
A
H
A
I'm
not
sure
what
the
a
good
way
around
this
is
like.
It
is
a
little
bit
of
a
foot
gun,
but
unless
we
want
to
have
like
a
separate
marker
for
there
is
there
like
there's
a
high
level
web
poker,
which
seems
like
a
little
bit
too
strong
of
the
tie
between
like
semantics
in
control
or
runtime
and
semantics
and
controller
tools.
I,
don't
really
know
what
we
can
do
about
this.
B
D
A
G
G
They
shall
be
faces,
so
it
is
more
like
a
Fisher
Price
I
believe
it
that
may
be.
The
implementation
will
be
point
notching
the
controller
at
the
time
which
I
thought
was
easier
illustrated
in
the
contribution
time,
so
I
don't
like
it.
It's
basically
are
you
to
like
it
to
able
to
use
the
client
ago
provider
by
the
control
different
time,
no
matter
if
the
resource
it
has
are
not
the
watch.
G
So
if
you
don't
have
the
permission
to
watch
a
resource
II,
because
this
could
be
simulated
by
removing
the
permission
from
the
operator
as
well.
So
if
they
don't
have
the
permission
show
what
that
resource
will
be
like
in
the
same
way
being
which
we
forget
or
connected.
So
my
idea
was
like
he
skipped
the
cast
in
desperation
because
I
wonder
things,
that's
because
she's
with
the
wall,
she
is
required
because
of
the
cache,
because
the
catch
is
using
cheese.
G
A
That
made
sense
so
I
think
I
mean
you
can
always
fall
back
manually
back
to
manager,
get
API
reader
to
use
a
live
client
instead
of
using
the
cache
clients.
Do
that
I
think
the
two
cases
you
mentioned,
no
permission
to
watch
and
watch
not
implemented
or
slightly
separate,
and
because
no
permission
to
watch
is
a
lot
harder
is
if
there
are
two
different
paths,
is
it
attacked
right?
No
permission
to
watch
requires
sending
a
self
subject
access
review
to
figure
out.
A
G
The
the
other,
the
PII
hater
I
mean
the
client
enter
the
other
one.
We
are
able
to
build
the
object
Luigi.
We
cannot
simply,
for
example,
no,
oh
yeah
yeah
you
to
shake
you
say
just
she
would
you
like
it
we're
sorry
sake.
We
defined
the
workaround,
which
I
could
you
not
to
find
a
workaround
it'll
what
day
the
resource
or
and
other
operations?
No,
oh,.
A
G
I
think
it
has
another
way
to
attend
to
the
same
needs,
but
she
you'd
be
very
nice
if
he
each
was
obstructed
for
the
developer.
Do
you
know
what
I
think
he
would
you
be
great
you
like
it?
We
could,
along
the
main
city
or
KDS
the
source.
Is
notifications
or
something
like
that.
I,
don't
know
we
cheese,
but
could
you
we?
Could
you
I
believe
it
that
you
we
couldn't
find
a
way
to
obstructed
it?
You
know
justice,
keep.
Okay,
don't
have
any
permission
to
watch
which
I
can
engage.
I
came
up
to
date.