►
From YouTube: Kubernetes KubeBuilder Meeting 20201008
Description
KubeBuilder Meeting for 2020/10/08. See https://sigs.k8s.io/kubebuilder for more details.
A
Hello,
this
is
the
kubernetes
controller,
runtime
and
controller
tools
meeting
for
thursday
october,
8th
2020.
As
a
reminder,
this
is
being
recorded
and
will
be
uploaded
to
youtube.
Eventually,
so
don't
say
anything
you
don't
want
recorded
for
all
posterity
right.
Last
time
I
looked,
we
had
two
items
on
the
agenda.
Let
me
double
check
to
see.
If
there's
anything
else,
I
think
the
first
was
a
question
about
a
particular
pr
that
just
got
opened.
A
B
So
over
at
operator,
sdk
we're
trying
to
start
having
an
actual
release
cycle
and
we
have
a
release
blocking
issue,
that's
dependent
on
that
pr
being
merged.
So
we
can
bump
our
dependency
on
kube
builder.
B
So
I
just
wanted
to
come
here
and
ask
I
think
camilla's
the
actual
author
of
the
pr.
What
can
we
do
to
get
that
merged
as
quickly
as
possible?.
C
I
think
the
point
is
that
apr,
it's
a
changing
v2.
That
is
not
a
bug
fix,
so
gabi
already
helped
us
by
review
making
the
review
we
did.
It
is
changing
the
v3
plugin,
but
he
before
gets
married.
I
I
think
he
we
need
to
more
one
person
saying
like
okay,
I'm
okay
with
emergencies,
individual,
because
it's
not
about
fix.
We
agreed
that
you
try
to
push
it
all
changing
victory.
D
And
just
to
clarify
for
those
who
might
not
be
familiar,
the
the
change
that
went
into
the
v3
go
plug-in
for
scaffolding
made
it
possible
to
use
a
different
q
builder
installation
method
than
the
one
that
installs
q
builder
via
its
tar
file
release.
So
you
can
now
do
like
go.
D
Get
couponer
and
it'll
install
the
binary
and
the
reason
that
that
hasn't
worked
before
is
because
there
are
other
inf-test
binaries
that
exist
as
part
of
the
queue
builder
release
that
q
builder
v2
plugin
kind
of
expects
to
be
in
the
right
place.
So
the
change
basically
is
instead
of
expecting
things
to
already
be
in
the
right
place.
D
D
The
make
task,
so
the
update
here
is
basically
curl
a
script
from
controller
runtime.
That
knows
how
to
install
and
test
binaries
and
then
execute
that
script.
D
And
what
it'll
do
is
it
download
the
executing
that
script
will
download
the
mtest
binaries
into
the
prop
into
the
local
project
like
in
the
I
think,
it's
in
a
test
bin
directory
and
then
it
sets
up
the
m-test
environment
such
that
when,
if
test
runs,
it'll
use
those
binaries
that
were
just
downloaded.
C
The
advantages
of
that
is
that
we
can.
We
have
our
issue
in
the
computer
for
don't
release
the
binaries
together
anymore,
so
it
could
be
helpful
for
dutch
as
well.
The
other
point
she
is
now
we
asking
for
the
users
using
sudo
to
install
the
binaries
as
some
step
is
that
requires
we
put
you
this
to
the
commands.
A
I
mean
it's,
I
mean
they're,
not
really
arbitrary,
we're
downloading
them
from
a
specific
location,
they're
they're
from
our
our
auto
generated
our
auto
generated
gcp
bucket.
I
believe.
E
Sorry,
I
guess
I
was
thinking-
maybe
they're
I
just
didn't-
haven't
looked
at
the
pr,
but
I
would
hope,
there's
some
verification
to
make
sure
that
they're
not
tampered
with,
etc.
A
Yeah
I
we
should
go
back
and
double
check
the
script.
On
that.
I
also
added
another
comment:
camilla.
I
don't
think
we
should
be
pulling
from
the
master
like
the
head
of
master
as
the
expected
place
to
pull
from
that's
that
never
ends
well.
C
A
We
we
we
can't
merge
this
as
like.
We
have
had
things
that
have
pulled
from
master
before
and
if
we
release
with
that
it'll
break
someone
like
it,
it
it
has.
In
the
past.
We've
we've
like
that
was
that
at
one
point
in
q
builders,
life
that
was
a
major
source
of
issues
accidentally
breaking
like
having
a
release
that
pulled
stuff
from
master
of
control
over
time
and
then
accidentally
breaking
stuff.
When
we
emerged
to
the
master
of
control
at
runtime,
this
makes
sense.
D
Camilla,
it
looks
like
the
setup
and
test
script
and
control
runtime
is
at
least
it's
in
the
most
recent
0.6.3
tag.
A
Yeah
and
we
can
like,
if
we're
expecting
it
to
be
used
somewhere,
we
can
cherry
pick
the
script
back,
but
I
think,
as
as
oran
brought
up,
do
we
verify
that
we
have.
That
is
a
good
point.
Can
someone
verify
that
we
have
verification
in
the.
D
Script,
I
don't
think
we
do,
I
think
it
just
downloads
them
are
there
are
there
are
the
binaries
signed,
like
all,
are
other
signature
files
that
can
be
used.
A
D
A
I
think
the
very
least,
let
me
let
me
double
check
something
but
yeah
we
can.
We
can
fix.
A
A
C
A
A
I
will
I
can
do.
I
have
one
more
comment
that
I
might
want
to
make
later,
but
I
need
to
do
a
little
more
reading
through
this
pr.
So
my
my
other
concern
here
is
potentially
this
is.
C
A
Better
and
this
my
my
initial
reading
of
it
is
this-
should
not
cause
issues
for
existing
users,
because
existing
users
will
have
already
downloaded
files,
so
we
should
be-
and
this
only
happens
on
scaffold
initial
scaffolding.
A
I'll
add
another
comment,
but
I
think
this
should
also
be
in
a
separate
make
file
target.
D
So
do
you
think
it
would
I
don't
I
don't.
I
probably
need
to
look
at
it
into,
but
should
we
check
to
see
if,
like
the
user,
local
q
builder
files
already
exist
and
if
so
then
just
use
those
or
do
you
think
we
should
get
away
from
that
and,
like
always
just
use
the
local
directory.
A
D
C
A
A
If
not,
we
can
move
on
to
the
next
issue.
I
think
this
one
was
from
kevin,
so
kevin.
Why
don't
you
give
the
background.
G
Yeah,
so
basically,
we
want
to
start
discussing
more
fine-grained
control
over
life
cycle
management.
There
was
a
pr
up
for
this
a
few
months
ago,
that's
kind
of
stalled,
and
so
we
wanted
to
keep
the
ball
rolling
on
that.
Where,
basically
you
know,
the
idea
is
that
we
we
don't
have
great
control
over
over
controllers
in
the
manager.
G
We
can
add
them,
but
we
can't
really
remove
them
and
we
can't
restart
controllers,
and
so
the
proposal
from
from
oren's
pr
was
to
allow
for
removing
individual
informers.
And
you
know
that's
that's
definitely.
You
know
one
approach,
but
there
was
a
lot
of
discussion
in
the
pr
and
so
basically
just
trying
to
organize
that
discussion
here
and
see.
G
If
you
know,
if
that's
the
approach
we
want
to
take
and
if
so
how
we
can
move
forward
on
that
or
or
if
there's
alternatives,
and
if
we
need
to
like
go
to
api
machinery
and
and
get
changes
from
them
or
or
or
something
like
that,
but
basically
opening
up
the
discussion
on
that
and
and
and
trying
to
keep
that
going.
And
so
you
know
if
you've
seen
the
pr.
G
The
idea
is
that
you
know
we
basically
just
expose
a
remove
a
remove
call
on
on
informers,
and
so
you
can
remove
any
any
informer
by
by
the
object
and
the
discussion
around.
That
was
basically
that
there's
a
few
concerns.
G
One
is
that
we
we
open
ourselves
up
to
silently
for
controller
silently
degrading
if
the
if
the
given
informer
is
removed
for
that
controller's
watch,
and
so
you
know,
there's
discussion
around
around
potentially
doing
ref,
counting
or
or
or
you
know,
creating
a
cash
per
watch
and,
and
that
seemed
to
be
the
main
concern
and
kind
of
what
I
wanted
to
bring
up
to
discussion.
D
D
Maybe
you
can
chime
in
a
little
bit
on
this
if
you
have
any
thoughts
but
around
there's
another
issue:
that's
about
sending
list
options
to
the
server
when
we're
setting
up
informers
and
there's
a
possibility
of
having
multiple
informers
for
a
given
kind
that
have
different
list
options.
Potentially,
so
I'm
wondering,
depending
on
how
that
goes
like,
what's
going
to
be
the
interplay
between
these
two
things,
something
we
might
want
to
consider.
B
Yeah,
that's
an
interesting
point
to
bring
up.
I
do
imagine
that
in
some
shape
or
form,
the
the
informers
map
to
a
controller
will
be
mapped
by
the
controller's
name.
So
you
should
be
able
to
remove
those
informers,
but
it's
a
matter
of
how
to
do
that
and
yeah.
I
don't
know
what
the
implementation
for
actually
removing
a
controller
or
an
informer
would
look
like
so
okay
comments
that,
but
that
is
important.
D
That
may
be
like
you,
you
have
a
watch
or
you
have
a
controller.
That
only
cares
about
maybe
secrets
that
have
a
certain
label.
So
you
don't
want
to,
and
let's
say
you've
got
a
cluster
scoped
watch.
You
don't
want
to
necessarily
watch
every
single
secret
across
the
entire
cluster.
You
just
want
to
watch
the
ones
that
you
care
about.
E
E
E
Maybe
the
list
options
are
part
of
that
key,
but
I
think
that's
maybe
a
separate
topic.
I
think
the
concerns
around
removing
conformers
or
around
reference
counting
and
making
sure
that
there's
not
another
informer
that
another
controller.
It
still
depends
on
that
informer,
and
so
maybe
that's
one
topic
we
should
drill
into
and
think
about
where
the
best
place
to
implement.
A
A
Or
a
lot
cleaner,
rather
so,
not
quicker,
because
you
know
you
need
to
get
the
api
machinery
changes
merge,
but
we're
we're
going
to
talk
with
some
folks
and
see
if
we
can
figure
out
what
kinds
of
things
they
would
be
amenable
to.
E
A
Yeah,
that's
one
of
the
things
we're
talking
about,
I
think
getting
getting
a
count.
At
least
we
can
do
outside
of
api
machinery
because
we
wrap
informer
in
an
interface
before
we
return
it.
So
you
could.
E
A
Those
right
so
you
could
add
event
handler
ad
event
handler
could
could
increment
a
ref
count
and
then
we
could
have
a
method
for
like
decrementing
ref
count
that
we
add
and
controller
to
call
it.
When
we
stop
and
then
maybe
you
don't,
maybe
the
you
know
the
event
handler
doesn't
get
removed.
A
But
if
you
know
you're
stopping
your
controllers
all
at
once
for
a
particular
thing
then
like
maybe
we
get
away
with
doing
that
and
then
the
refs
account
eventually
reaches
zero,
but
yeah
actually
being
able
to
remove
the
event
handlers
as
we're.
Removing
the
controllers
would
be
real
nice,
so
hopefully,
by
next
meeting,
we'll
have
a
better
answer
as
to
whether
we
can
get
the
ball
rolling
on
that.
A
We
don't
we
don't
like,
so
to
be
clear,
we
don't
like
rap
rap,
but
we
only
return
a
particular
interface.
That's
a
subset,
so
you
could.
E
A
Yeah,
basically,
we
wanted
to
have
some
stability,
because
if
we
expose
the
raw
informer,
if
the
interface
of
the
like
the
raw
informer
or
adds
new
methods,
we
are
now
bound
to
those
methods
and
then,
if
it
then
changes
those
methods
or
removes
those
methods
yeah,
it's
still
bound
to
those
methods
which
are
the
thing
that
happens
quite
frequently
in
upstream
kubernetes.
So
we
wanted
to
expose
like
the
minimal
subset
that
we
needed
for
controller
runtime
to
work.
E
I
understand
so
interesting
because
that's
another
option
that
maybe
we
didn't
discuss,
which
is
you
if,
if
we
were
open
to
wrapping
the
actual
former
implementation,
then
we
could
proxy
the
you
know
we
could
have
the
single
event
handler
be
that
rapper
and
then
proxy.
The
the
the
events
to
to
the
event,
handlers
that
were
reference
counting.
A
We
yeah
that's
a
good
point.
We
could
have
our
own.
We
could
have
our
own
event
handler
dispatch
here.
I
that's
something:
let's,
let's
that's,
definitely
a
good
idea
to
keep
in
mind,
especially
like
so
like.
If,
if,
if
sig
api
machinery
comes
back
and
says,
we
are
not
interested
in
this,
then
obviously
like
that
could
be
a
real
good
avenue
to
approach.
If
they
said
we
are
interested
in
this,
but
it
might
take
a
while.
Then
that's
also
a
good
avenue
to
approach.
If
we
can
get
it
in
quickly.
E
E
Yeah,
I
hadn't
thought
about
that
before.
That's
that's
interesting
when
and
where
do
you
think
that
you're
planning
to
broach
the
conversation
with
them.
A
G
Yeah,
I
think
I
mean
yeah.
I
think
I
think
we'll
get
an
initial
feel
of
some
of
the
some
of
the
api
machinery
people
tomorrow
and
should
have
an
idea
of
what
we
want
to
bring
up
at
the
next
apm
machinery
meeting.
E
G
Did
you
also
want
to
solidly
discuss
the
the
thought
that
that
we
were
talking
about
with
servicing
the
watch
error
handler
from
from
client
go
through
the
informer?
This.
A
Is
this
is
kind
of
a
little
bit
of
an
implementation
detail,
but
I
was,
I
was
thinking
like
along
the
same
lines.
One
of
the
things
that
would
be
nice
is
a
little
bit
more
fine,
and
this
is
like
another
thing
that
we
need
to
change
and
upscaling
theory,
but
like
a
little
bit
more
fine-grained
notification
and
handling
of
what
happens
when
an
informer
encounters,
an
error.
A
One
of
the
things
that
we've
seen
from
some
of
our
users
is
either
a
desire
to
more
precisely
control
the
back
off
of
the
underlying
informer
on
certain
errors
or
like,
and
this
ties
back
into
the
stopping
controllers
kind
of
a
desire
to
know
when
a
particular
crv
is
removed,
depending
on
like
what
the
relationship
is
between
a
particular
controller
or
a
particular
manager
binary
and
like
the
the
end
user
of
a
cluster.
A
It
might
not
be
unexpected
that,
like
the
end
user,
just
like
removes
the
crd
and
then
it's
gone
for
you
know,
oh
days
weeks,
whatever
right
like,
and
in
that
case
you
probably
want
to
stop
the
controller
just
sit
there
until
it
gets
re-added
and
so
like.
A
Obviously,
you
can
do
that
by
watching
discovery
documents,
but
it
one
potential
avenue
that
might
be
interesting
and
like
kind
of
a
better
notification
mechanism,
or
at
very
least
like
lower
latency,
would
just
be
to
like
propagate
propagate
the
error
up
from
the
informer
right,
because
the
informer
gets
a
very
particular
error
when
that
happens,
and
so
we
could
like,
we
could
bubble
that
up
and
there's
there's
already
like
a
low-level
private
hook
for
like
controlling
what
gets
called
when
there's
a
particular
error.
A
So
I
was
hoping
that
potentially
maybe
we
could,
we
could
make
use
of
that
from
api
machinery
and
like
make
that
public
and
then
maybe
maybe
bubble
that
up
and
control
our
rent
time
just
so
that
folks
can
can
make
decisions
like
that.
I
don't
know,
did
any
of
that.
Make
sense
to
people
and
people
have
any
thoughts
on
that.
E
It
makes
sense
to
me:
I
have
similar
use
cases
and
again
the
way
we're
approaching
that
today
is
we
have
these.
You
know,
I
think
it
was
in
the
document.
It
was
called
like
a
meta
controller,
which
is
handling
what
needs
to
happen
when
those
crds
come
and
go
right
on
behalf
of
the
actual
controller,
that's
doing
the
work,
and
so
I
think
the
propagation,
the
error
propagation
idea
is
is
is
very
interesting.
E
Another
thing
we
could
discuss
is:
do
we
want
to
institutionalize
this,
this
meta
controller
concept
for
for
crds
and
the
ability
to
to
start.
You
know
baking
in
that
ability
to
start
and
stop
controllers
into
controller
runtime
with
something
that
knows
how
to
watch
for
presence
of
of
registered
kinds.
A
Yeah,
so
I
think
that
was
one
of
the
things
that
kevin
originally
submitted
in
in
his
design
proposal,
and
I
think
there
was
some
initial
hesitance.
I
don't
know
if
alvaro
is
here
today.
I
think
alvaro
had
some
thoughts.
A
I
seem
to
recall,
but
I
think
some
folks
had
some
initial
hesitance
as
to
like
whether
or
not
we
needed
to
institutionalize
or
whether
or
not
it
was
better
to
just
like
make
sure
we
have
the
support
and
then
and
then
like
just
have
make
sure
it
was
easy
and
then
just
have
people
write
the
meta
controller
themselves,
but
I
think
at
the
very
least,
we
should
have
hooks
for
doing
that
kind
of
stuff
and
make
sure
we
have
good
support.
A
Also,
as
an
aside,
we
should
come
up
with
a
different
name
than
metacontroller,
because
it
may
get
real
confusing
with
the
actual
thing
called
metacontroller
right.
It's
a
different,
related
project
called
metacontroller
and
I
have
a
feeling
eventually,
someone's
gonna
come
across
this
blog
and
get
real
confused.
So
we
should.
We
should
probably
come
up
with
a
different
name
for
that.
H
And
yeah,
and
by
the
way
in
the
meeting,
so
the
main
concern
I
had
with
what
the
initial
pull
request
that
actually
did
call
changes
and
did
is
that
this
was
very
specific
to
starting
and
stopping
controllers
based
on
presence
or
absence
of
custom
resources.
H
And
what
I
would
prefer
is
that
we
have
sort
of
a
two-step
thing
where
first
we
generically
allow
to
start
and
stop
controllers,
and
then,
in
the
second
step,
we
would
think
about
the
best
way
of
doing
this
for
custom
resources
which
might
or
might
not
be
such
a
controller
controller.
How
you
ever
want
to
call
it,
but
I
guess
I
would
like
to
have
the
two
things
like
allowing
it
in
general
and
doing
it
specifically
for
custom
resource
the
definition.
Patents
split
out.
A
That
seems
pretty
reasonable
to
me,
yeah,
okay,
that
seems
pretty
clear
and
I
definitely
think
that's
something
that
we
can
do
we
can.
We
can
do
this
in
two
phases.
A
That
does
also
give
us
more
more
room
to
experiment
around
with,
like
what
we
have
is
the
interface
for
like
getting
like
running
the
controllers
and
or
get
it
managing
the
controller
like
automatically
managing
the
controller
lifecycle
like
making
it
easy
to
start
stop
on
presence
or
absence
of
resources.
A
G
So
I
guess
I
don't
know
what
what
do
we
like?
What
do
we
want
to
move
forward
on
or
what's
like
it
sounds
like
like
adding
ref
counting
on
our
end
or,
like,
I
guess,
agreeing
on
on
what
the
interface
would
look
like
on
our
end.
Is
I
don't
know?
Is
that
the
next
step
here
or
what
do
you
think.
A
Yeah,
I
think,
and
probably
like
oren
and
alvaro
can
hopefully
supplement
this
a
bit,
but
I
think
the
tldr
that
I
got
out
of
that
discussion
was
like
you
said.
The
next
step
is
to
like
the
next
steps
which
we
can
do
in
parallel
is
a
figure
out
what
we
want
the
interface
to
look
like
from
our
end,
like
the
high
level
interface,
the
users
interact
with
for
like
ref,
counting
on
informers
and
and
starting
stopping
controllers,
and
how
those
two
are
tied
together
or
maybe
lack
thereof.
A
But
I
have
a
suspicion
that
they'll
probably
be
integrated
at
least
a
little
bit
figure
out
what
the
interface
looks
like
without
the
guts
and
figure
out,
something
that
we're
like
happy
from
a
ux
perspective.
Stuff
like
that
and
then
in
parallel.
Also
we'll
talk
to
api
machinery
and
see
what
kind
of
support
we
can
get
there
and
then
once
we've
kind
of
resolved,
those
two
or
maybe
like
packed
up
a
prototype
involving
two
of
those
because,
like
we
can,
we
can
hack
on
api
machinery
before
before.
A
We
get
a
official
answer
and
just
like
see
how
the
prototype
actually
feels.
We
can
come
back
and,
like
actually
start
put
together
a
little
proposal
that
has
the
ideas,
implement
that
and
then
like
the
final
step
after
that
we'll
be
seeing
if
we
can
figure
out
the
ux
around,
like
oren,
said
institutionalizing
like
the
controller
controller
or
or
something
similar
right
like
making
it
easy
for
people
to
do
that,
but
that'll
be
like
the
last
step.
Does
that
does
that
adequately?
A
Summarize
is
my
my
impression
of
the
discussion
correct
according
to
other
folks.
A
All
right,
I
I
actually
I
have
one
more
thing
to
suggest,
and
ironically,
my
suggestion
currently
may
avoid
my
ability
to
to
do
this
in
the
future.
I
don't
I
don't
like
I'm
totally
fine
backing
off
on
this.
If
other
people
really
don't
like
this
idea,
but
I
was
wondering
if
like
maybe,
we
should
have
a
policy
that
we
put
in
our
agenda
and
notes
that
we
not
that
we
lock
the
agenda
but
that
if
an
agenda
is
empty
by,
I
don't
know
like
9pm
on
wednesday
pacific.
A
We
say
that
the
meeting
or
or
we
can
change
that
time
or
whatever.
So
it's
more
convenient
for
all
the
folks
to
attend,
but
like
we
say
that
the
meeting
is
cancelled
just
so
that
we
don't
end
up
with
like
like
the
meeting
where
we
all
say,
do
we
have
any
agenda
items?
No,
let's
go
just
I
I
know
like
so
in
the
past.
I
I
know.
Sometimes,
if
you
don't
like,
have
that
policy
very
clearly
documented
it
can
get
frustrating.
A
I'm
I'm
gonna
be
honest
here.
It's
also
a
little
earlier
than
I
normally
get
up,
and
so
it's
a
little
bit
frustrating
waking
up
early
and
then
not
not
having
a
meeting
so
are
do
other
people
have
thoughts
on
that
like
are.
Is
that
like,
if,
if
I'm
being
unreasonable
here,
it's
like
definitely
or
just
like,
you,
don't
think
we
should
do
that.
Definitely
feel
free.
To
tell
me,
like
that's
fine,
I'm
fine
to
be
told
that,
but
I'm
curious
what
people
thought.
I
J
Expert
requiring
an
agenda
makes
sense
to
me
rules
to
be
defined
but
makes
a
whole
lot
of
sense.
E
E
A
I
I
mean
like
I:
can
I
or
or
one
of
the
other
like
we
should
probably
have
like
an
official,
a
semi-official
like
chair
role,
not
not
just
just
to
like
make
sure.
We
have
some
points
of
contact
for
people
on
stuff
like
this,
but
we
can
probably
have
someone
responsible
for
going
in
at
whatever
time
we
we
decide
and
just
writing
on
the
meeting
agenda
is
happening
or
is
not
happening.
A
Does
that
seem
reasonable
and
maybe
yeah
maybe
get
people
potentially
update
the
meaning
invite
too,
but
at
the
very
least
we
can
probably
have
someone
go
and
just
write
on
the
meeting
like
this
day
is
not
happening.
No
agenda
items
were
added.
A
A
Yeah,
I
know
I
I
think
ideally
we'd
also
want
to
have
the
meeting
cancelled.
I
think
that
hinges
on
who
has
edit
access
to
the
meeting.
Invite.
D
Yeah,
I
agree
that
we
should
do
that.
Looks
like
right
now.
Guest
permissions
are
only
for
inviting
others
and
seeing
the
guest
list.
I
can
certainly
there's
another
permission
for
modifying
the
event,
but
that
would
mean
that
every
single
guest
would
be
able
to
modify
the
event.
So
I
don't
know
if
there's
a
google
calendar
way
of
limiting
it
to
a
certain
group
of
people,
I'd
be
totally
happy
to
let
anybody
modify
it,
though
so,
whatever
we
decide
yeah,
I
definitely.
A
A
I
think
we
have
like
a
q
builder
admins
google
group,
and
we
can
add,
we
can
add
the
people
who
who
are
like
does
have
the
desire
to
be
responsible
for
managing,
like
you
know,
doing
the
the
role
of
the
sig
chair,
basically
like
managing
stuff,
like
meetings
and
stuff,
we
can
have
those
people
added
to
to
that
group.
A
I
don't
know
if
that
works,
that's
just
a
thought
off
the
top
of
my
head.
I
need
to
double
check
to
see
if
that's
actually
a
thing
that
you
can
do,
but
if
it
is,
then
that
might
be
an
option
we
can
we
can
we
can.
We
can
hold
off
on
officially
making
that
the
policy
or
anything
until
we
we
figure
those
details
out.
One
other
thing
I
wanted
to
make
sure
is
folks
in
time
zones
is.
A
I
don't
I
don't
know
what
time
zones
people
are
in
and
definitely
don't
feel
pressured
to
say
or
whatever,
but
if
you
are
in
a
time
zone
that
is
not
est
or
pst
or
somewhere
in
between
and
like
you
would
like
to
you
know
you
don't
like
this
for
whatever
reason
or
you
have
concerns
about
like
the
time
at
which
we
close.
A
Definitely
please
either
speak
up
here
or,
if
you'd
rather
speak
up
in
private,
as
for
just
about
anything,
feel
free
to
send
me
a
message
or
or
an
email,
but
select
message
is
slightly
preferred.
There.
A
A
That's
actually
like
the
beginning
of
my
day,
when
I
start
thinking
about
like
what
kind
of
stuff
I
want
to
put
on
the
agenda
for
the
next
day,
and
so
like
that's
really
tricky,
can
we
move
it
by
a
few
hours
or
something
so
that,
like
it
intersects
better
with
my
morning,
and
I'm
not
like
racing
first
thing
in
the
morning
to
add
item
agenda
items
before
the
agenda
closes
or
something
I
just
wanted
to
make
sure
we
weren't
in
that
situation.
A
I
think
we've
had
polls
in
the
past
on
meeting
times
and
it's
real
hard
to
get
the
entire
world
into
one
meeting
time
zone
slot.
But
from
what
I
understand,
this
works
decently
well
for
most
of
the
regular
contributors.
A
But
again,
if
you
have
thoughts
on
that
feel
free
to
ping
me
on
slack-
and
you
know,
I
can
always
send
a
message
about
that
as
well,
and
I
think,
unless
anybody
has
any
more
thoughts
on
that,
I
think
I
I
didn't
notice,
there's
another
agenda
item
from
two
more
agenda
items,
two
more
three
more
from
camilla.
A
So
take
these
away.
I'm
sorry.
C
No,
I
actually
have
it
one
two
three
for
me.
I
asked
for
housing
he's
one
before
my
first
one,
because
I
have
it
too
much,
so
I
can
push
for
the
other
reaching
as
well
on
egp
all
discussed
today.
The
first
one
is
just
about
the
kobe
and
rbc
proxy
image.
We
you.
We
confirmed
that
she
is
beautifully.
You
helped
me
with
that
sorry,
but
she
I
I
believe
that
I
don't
have
a
permissions
to
push
the
image
for
the
cloud
so
for
cheese.
C
I
used
to
need
your
help
and
she
another
thing
that
she
I
think
it's
nice
to
share
is
that
I
spoke
with
the
owner
of
this
project
and
he
told
me
that
he
is
seeing
how
to
donate
the
prostitute,
who
brc
to
go
bernard's
ark,
so
she's
going
to
fascinate
the
process,
although
we
do
she's
automatically
in
the
future,
but
she
I
think
we
need
to
get
some.
A
Yeah
that
that
that
makes
sense
in
kind
of
reverse
order
of
what
you
said,
yay
on
the
donation
definitely
like.
If
you
need
me
to
like,
let
if
he
needs
me
to
like
show
up
and
comment
on
on,
like
an
issue
being
like.
Yes,
we
in
controller
runtime
are
in
cube.
Overland
use
cube
our
back
proxy
and
you
know
it's
super
useful
and
we
would
really
appreciate
it
being
a
part
of
one
of
the
kubernetes
projects
like
happy
to
do
that.
A
For
the
permissions
for
the
permissions,
we
can
probably
put
together
a
cloud
build
file,
so
hopefully
the
idea
is
that
nobody
is
manually
pushing
q
bar
back
proxy
in
the
future,
because
that
requires
like
a
lot
of
individual
people
to
have
permissions
and
stuff
like
that.
But
instead
we
just
have
a
bunch
of
cloud
bill
configurations
and
we
can,
just
you
know,
tag
a
release
on
a
branch
and
it
gets
built
so.
C
A
Yeah
I
mean
we
can
we
can
always
that
yeah
okay,
we
can.
Can
you
ping
me
if
you
haven't
already
ping
me
offline
on
slack
and
it
can
be
boring.
Don't
worry.
A
C
Really,
thank
you
really.
Thank
you
for
all
all
your
help
and
with
the
you'll
help
me
with
the
donation
as
well.
Thank
you.
A
Of
course,
yeah,
and-
and
thank
you
for
for
for
tackling
this
as
well-
I'm
making
sure
all
of
our
infrastructure
is
up
to
date
and
stuff.
That's
real
important
all
right.
What
what
were
the
other
two
agenda
items
you
had.
C
I
Well,
thanks
camilla.
This
should
hopefully
be
quick.
I
felt
a
this
issue
last
week.
Basically,
I'd
like
any
event
that
triggers
a
reconcile
to
get
logged
as
kicking
off
the
reconcile,
since
the
reconcile
function
itself
doesn't
have
access
to
the
event.
I
If
you're
watching
a
bunch
of
different
resources,
it
can
be
tricky
to
know
which
one
was
actually
causing
it.
The
reconcile
this
was
originally
filed
for
the
ansible
operator
sdk,
and
I
thought
maybe
this
would
be
useful
generally,
but
it
also
seemed
a
little
bit
too
obvious,
so
that
there's
probably
a
reason
that
we're
not
already
doing.
H
This,
I
think,
generally
having
this
is
not
a
bad
idea,
because
it
helps
to
debug
certain
issues,
but
it
needs
to
be
on
a
very
high
lock
level,
because
there
are
controllers
that
watch
literally
thousands
of
objects
and
get
maybe
hundreds
of
events
per
seconds
and
it
completely
screws
the
logging,
if
you
lock
every
event
on
those.
A
Alvaro
took
the
words
right
out
of
my
mouth,
those
that
is
exactly
what
I
was
going
to
say
wholeheartedly
agree.
I'm
also
going
to
do
kind
of
a
side
plug
one
of
the
things
that
we've
been
kind
of
thinking
about
in
controller
runtime
for
a
while,
but
have
never
actually
gotten
around.
A
I
think
to
like
fully
fleshing
out.
I
think
someone
did
a
prototype
a
while
ago.
It
would
be
great
to
have
tracing
support
in
controller
runtime
separately
from
this,
like
completely
agree
with
everything
that
was
just
said,
we
should
have
this
at
a
high
log
level,
but
also
this
is
the
kind
of
thing
where
I
think
you
could
probably
benefit
a
lot
from
tracing
too.
So,
if
someone
has
any
interest
in
putting
together,
it
looks
like
we
just
got
a
chat
message.
A
If
someone
has
any
interest
in
putting
together
a
a
proposal
for
tracing
or
or
working
on
tracing,
it
would
be
real
cool,
especially
now
that
we
have
the
context
argument
on
reconcile.
L
Just
linked
something
that
came
up
yesterday
in
the
cafe
meeting,
the
like
brian
from
we
works,
like
has
already
done
something
like
this.
I
did
mention
that
like
now,
we
have
context,
so
maybe
we
can
reach
out
to
see
if
we
can
do
something
in
control
the
runtime.
I
know
that
they
they
were
interested
to
implement
open
tracing
in
controller
runtime.
So
maybe
this
could
all
work
out.
A
A
I
I
think
someone
cut
out
there
for
a
moment
also.
I
think
we.
J
I
think
he
was
looking
to
see
if
we,
if
he
needs
to
hold
off
on
his
current
request,
for
the
tracing
probably
had
more
background
than
I
just
gave.
A
Oh,
I
think
I
think
we
can
probably
go
ahead
with
with
putting
in
the
logging.
We
just
need
to
make
sure
it's
like
at
the
max
log
level
and
or
it's
like
on
its
own,
both
both
at
the
max
log
level
and
probably
like
in
its
own
specific
logger
or
whatever.
A
So
it's
really
easy
to
filter
out,
if
even
if
you're
logging
at
the
max
log
level,
because
like
was
mentioned,
some
things
watch
a
bajillion
resources
or
reconcile
real
fast
and
that's
a
great
way
to
cause
your
log
volume
to
explode.
But
you
know
module
of
that.
I
think
it's
fine
to
go
forward.
A
There
there
is
not,
there
is
not
a
max
log
level.
Technically,
certain
loggers
have
a
max
log
level,
which
is
something
that
we've,
I
think
mostly
tried
to
fix,
but
also
in
kubernetes.
We
effectively
have
a
de
facto
max
log
level
of
10..
I've
never
seen
anything
go
above
10,
and
I
think
our
official
guidance
is
that,
like
10
is
10
is
please
stress,
test
my
log
ingestion
system.
A
A
A
Yeah,
that's
that's,
probably
a
good
plan,
okay,
so
unless
anybody
has
any
like
last
minute,
you
have
like
30
seconds
to
say
something
if
you
want
to
say
something
but
otherwise
feel
free
to
follow
up
on
slack
or
on
our
mailing
list,
and
I
will
see
you
all
in
two
weeks.