►
From YouTube: Kubernetes SIG API Machinery 20201104
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
A
We
have
three
topics
in
our
agenda
and
then
I
will
open
up
to
any
question
in
case
we
have
and
trying
to
be
mindful
of
everybody's
time.
I
will
go
directly
to
kevin
our
first
topic
of
today
kevin.
It's
all
yours.
Let
me
know
if
you
want
me
to
open
any
well,
I
will
open
these
documents.
If
you
want
me
to
go
to
any
of
any
section
in
particular,.
B
Sure
yeah,
so
just
to
give
a
an
introduction
of
you
know
what
the
what
what
we're
talking
about
here.
So
currently,
our
controllers,
you
know,
start
informers
but
are
unable
to
stop
them
controllers
start
informers
by
passing.
You
know
a
stop
channel
that
when
closed
will
shut
it
down,
but
there's
no
condition
by
which
the
informer
can
shut
itself
down,
and
this
has
some
unintended
consequences.
B
A
big
one
is
that
dynamic
conformers
are
unable
to
stop
themselves
when
the
crd
they
are
watching
is
uninstalled,
and
so
you
know
this
has
been
brought
up
in
in
a
bug
that
that
I
post
in
the
in
the
background
section
of
this
dock,
and
basically
you
know
the
issue
there.
B
Is
that
or
you
know,
one
of
the
issues
that
arises
because
of
this
is
that
the
cube
controller
manager
via
the
resource
quota
and
garbage
collection
controllers
creates
log
spam
as
a
continuously
404s
retrying
lists
and
watches
on
resources
that
will
never
exist
and
hypothetically
one
could,
you
know,
create
and
delete
enough
crds
to
oom
the
api
server,
and
so
that
issue
is:
is
there
and
there's
a
lot
of
community
feedback
trying
to
get
that
fixed
and
another?
Another
kind
of
motivation
that
we
see
for
fixing?
B
This
is
that
this
greatly
inhibits
the
control
that
that
operators
and
custom
controllers
have
over
the
life
cycle
of
their
controllers,
and
so
you
know
there
are
desired
use
cases
around.
You
know
that
that
don't
fit
the
kind
of
the
model
that
this
was
built
on,
that
the
admin
of
a
controller
is
necessarily
the
admin
of
the
cluster,
and
so
you
know,
controller
admins.
In
a
sense,
they
have
little
control
over
which
resources
are
installed
and
uninstalled.
B
So
ideally,
they'd
have
a
way
to
to
run
a
controller
that
tells
the
controller
to
shut
down
informers
for
resources
that
don't
exist,
or
you
know
create
a
new
informer
when
the
resource
is
reinstalled
and
so
kind
of
what
we're
proposing
here
is
is
a
new
method
on
the
sheridan
former
that
runs
the
informer
and
passes
options
that
indicate
what
conditions
should
trigger
the
informer
to
shut
down,
such
as
when
the
listing
watch
and
the
reflector
fails
like
with
the
404
indicating
that
the
crd
has
been
uninstalled,
and
so
this
you
know
this
creates
an
opt-in
way
to
give
greater
control
over
the
lifecycle
of
informers
without
breaking
current
use
cases
and
then
kind
of
in
the
dock
and
in
the
implementation
I
kind
of
go
through
how
this
is
used
by
you
know.
B
This
can
be
used
by
the
factory,
both
the
the
the
sharedom
former
factory
and
the
metadata
factory,
and
so
you
know
so
far.
It's
I've
used
it.
I've
tested
it
with
controller
runtime
and
with
and
with
the
research
quota
controller.
But
you
know,
that's,
you
know
kind
of
wanted
to
get
feedback
on
this
idea
now,
because
there
has
been
kind
of
feedback
around
this
before
and
then
kind
of
that's
kind
of
the
main
thing
and
a
smaller
second
part
is
that
you
know.
B
There's
there's
also
a
use
case
around
when
a
when
a
you
know
when
certain
resources
are
installed
run,
one
can
run
one
set
of
informers
otherwise
degrade
to
to
another
set,
and
this
would
require
some
form
of
like
tracking
event,
handlers
on
the
informers
and
so
being
able
to
remove
event
handlers
and
see
the
count
of
the
number
of
event.
Handlers
would
also
be
super
useful
for
that
use
case
and
and
kind
of
ties
into
this.
B
This
kind
of
idea
of
greater
life
cycle
control
of
the
of
the
of
the
informers
life
cycle.
So
that's
kind
of
what
what
I've
discussed
here
and
have
given
an
implementation
for
and
just
kind
of
want
to
open
the
floor
to
discussion
about.
A
C
Thanks
kevin,
I'm
always
in
favor
of
like
fixing
the
shutdown
path.
So
I
think
this
is
great
I'd
like
anyone
on
the
call
who
has
thoughts
to
speak
up,
though.
A
I
would
assume
that
solely
you're
in
favor
of
all
these
two.
D
Yeah,
yes,
definitely.
This
is
a
a
use
case
that
a
couple
people
have
asked
for
in
controller
runtime.
In
addition
to
the
obviously
nice
closing
the
closing
the
loop
pun,
kind
of
intended
on
a
resource
quota
and
garbage
collector
controller.
A
C
I
I
wouldn't
expect
anyone
to
be
against
this.
If
anyone
is
the
only
thing
that
that
we
might
see
is
people
who
are
interested
in
it
and
have
like
some
specific
desire
about
how
it's
done.
I
guess
that's
mostly
what
I
would
expect
in
the
way
of.
E
Comments,
the
only
overlapping
thing
I
saw
and
this
I
thought
this
was
a
different
issue
when
you
first
started
going
through
it,
but
it's
not
as
related,
which
is
when
we
get
the
separate
issue,
which
is
when
we
roll
back
in
the
in
the
control
plane
we
restore
from
an
earlier
resource
version.
Basically,
every
single
controller
cache
is
completely
invalid
and
until
either
reconciliation
or
a
complete
reset
happens.
C
There's
actually,
there's
actually
a
way
to
do
that
to
do
that
safely.
It's
complicated,
though.
E
I
don't
wanna
get
into
specifics
here.
I'm
just
yeah.
C
E
C
Do
we
need
a
cap,
I
mean
how
much
how
much
code
is
this
gonna
end
up
being,
or
I
guess
the
I
guess
how
much,
how
much
interface
fluctuation?
Is
this
gonna
end
up
being
and
will
it
will
it
break
existing
clients
or
is
it
like
you
just
update
and
things
continue
to
work
and
you
just
have
some
additional
options.
B
Yeah,
it
shouldn't
break
any
additional
client
or
any
existing
clients.
If
we,
as
it's
currently
proposed,
it's
just
adding
a
couple
methods
on
the
shared
informer
and
not
actually
changing
the
behavior
of
the
current
existing
ones,
maybe
long
term,
that's
not
what
we
want.
Maybe
we
want
the
default
behavior
to
you
know,
accept
these
options
or
something,
but
I
think
I
don't
know
as
it's.
You
know
as
it's
currently
written
it's
just
a
few
more
methods
on
on
the
shared
informer
yeah.
I
might.
I
might.
E
E
Style
closure,
which
is
pretty
standard
now,
like
you
know,
like
leader
election
uses
context
for
closure
stuff
like
that,
like
those
are
the
kinds
of
discussions
and
then
the
implications,
but
I
think
we
could
have
those
in
a
in
a
proposed
api
change
poll
and
not.
C
A
A
Very
good
so
looks
like
there
is
general
agreement.
People
may
have
some
comments
on
the
implementation,
so
I
don't
know
if
we
need
anything
else
here.
It
does
feel
like
green
light,
for
you
to
move
on
sounds
good
thanks,
very
good.
Thank
you.
Everybody!
G
Yeah,
so
we
thought
that
we
started.
I
was
trying
to
stick
with
her.
We
thought
that
the
api
server
does
not
validate
objects
created
with
duplicated
from
the
reference
or
gets
updated
to
the
ways
to
begin
duplicate
references.
G
G
So
I
bring
this
to
the
stick
meeting,
because
I
think
there
might
be
a
data
about
how
people
feel
we
should
fix
this.
I
see
two
options
we
can
fix
this.
We
either
reach
that
we
add
validation
and
reject
requests
which
will
break
clients
who
deficit
or
we
silently
chops
the
fields
and
makes
the
duplicate
case
call
the
good
news
is.
A
G
Client
that
has
this
I'm
aware
of
is
the
upgraded
lifecycle
manager
and
I
think,
if
people
who
already
do
this,
it's
it's
it's
more
obvious
compared
to
other
api
vibration,
because
if
they
print
out
the
object,
they
will
see
the
on
the
reference
took
place.
So
I
think
that's
the
good
news
is
we
don't
expect
a
lot
of
clients
to
do
this,
and
if
we
break
clients
that
this
we
probably
won't
break
blood
yeah.
So
still
I
would
like
from
people
about.
G
Can
we
just
reject
such
a
request.
C
Right
thanks
thanks
howie,
just
in
case
anyone
had
trouble
understanding
your
microphone's
a
little
little
yeah.
I
don't
know,
what's
going
on
there,
basically
real,
really
briefly,
we've
got
some
it's
possible
to
make
duplicate
owner
references
in
metadata.
C
We'd
like
it
to
not
be
possible,
we've
got
the
choice
of
rejecting
requests
and
potentially
breaking
clients
sending
these
or
silently
removing
duplicates.
So
we
wanted
some
broader
input
on
which
of
those
two.
I.
H
Think
you
have
an
additional
option
right.
You
have
the
ability
to
remove
it
and
return
a
warning
right.
C
So
yes
yeah,
hopefully
we
would
return
a
warning
if
we
silently
remove
it
well,.
E
E
E
C
I
G
E
Certainly,
we
could
warn
to
start
with
and
see
it
do.
We
have,
I
mean,
like
adding
the
warning
seems
pretty
reasonable.
H
That
parts
that
part's
easy,
even
if
we
gated
this
functionality
off
for
one
release
and
turned
it
on
later,
I'd
be
okay
with
that
as
well.
But
we
still
face
the
problem
of
what
is
less.
C
Yeah,
I
don't
know
if
the
warning
is
like
like
I'm
not
I
I
I'm
definitely
in
favor
of
it,
but
I
don't
know
if
it's
going
to
be
super
useful
because
most
likely
the
the
people
making
or
the
things
making
these
requests
are
controllers
and
there's
no
human.
That's
going
to
be
looking
at
the
warning
so.
C
E
C
This
this,
the
the
the
example
we
have
linked,
has
known
about
this
issue
for
like
almost
a
year,
and
it
hasn't
been
addressed
so
like,
and
it's
a
it's
like
an
alpha
version
of
it,
so
that
hasn't
been
recompiled
recently.
So
I
don't
think
it
even
knows
about
warnings,
so
I'm
not
sure
that
it's
going
to
show
up
in
their
logs.
E
I
mean
we
generally
bias
away
from
breaking
clients.
Unintentionally,
what's
happening.
H
Now
is
the
data
is
being
poisoned.
It's
breaking
everyone
so
like
it's,
not
a
we're
breaking
a
client
for
no
reason
if
this
client
is
breaking
all
the
clients
and
we
need
to
stop
it
in
in
some
ways,
you
could
even
mark
this
as
like
a
very
low
severity.
Cbe,
someone
who
can
write
an
owner
ref
can
hose
your.
E
Cluster,
if,
if
they're
exactly
duplicate,
I
re,
I
struggle
to
think
about
someone
who's
doing
like
in
in
the
class
of
people
who
might
have
the
logic.
If
there
were
some
fields
that
weren't
duplicates
that
were
kind
of
like
you
know,
they
meant
the
same
thing
right
like
two
controller
refs
or
whatever,
that
intentionally
like
had
slightly
different
values,
but
meant
the
same
thing
versus
completely
identical
duplicates.
E
H
Before
we
transition
and
do
it
yeah,
I
mean
like
this
intersects
with
with
the
pr
that
I
was
looking
at
for
fixing
the
bad
owner.
Refs
can
delete
good
children
problem
right
because
you
end
up
with
duplicates
and
now
how
do
you
choose
the
next
loop
with
a
check
like
it
becomes
a
mess
yeah.
E
H
C
Yeah
yeah,
I
think
mvars,
is
the
big
offender
on
this.
D
E
Yeah,
I
I
guess
the
question
would
be
is
like
a
lot
of
our
other.
We
want
to
do
something
that
we
think
the
right
is
the
right
thing
to
do,
and
it's
pretty
clear
that
we
should
do
it.
Can
we
generate
enough
signal
from
the
community
before
we
trigger
destructive
or
excessive
problems
and
yeah
like
a
one
or
like
warning
one
or
two,
some
level
of
you
know
communication
with
a
period
a
shorter
period
does
seem
like
something
we've
also
justified
before
for
other
changes
of
this
nature.
So.
H
H
Because
that
way,
like
work
can
still
progress
at
least
hopefully
right.
If
you
have
an
updater
who's,
updating
something
different,
you
strip
the
duplicates,
so
you
persist
as
opposed
to
creating
something
different
and
you
can't
persist
because
the
owner
refs
are
bad
in
an
object.
You
create
your
own
arrests
for,
like
that
seems
like
a
really
unfortunate
client
effect,
doing
straight.
Rejection.
I
Wouldn't
start
just
removing
the
duplicate,
not
tell
the
client,
he
did
nothing
wrong.
Yes,
there's
a
warning,
but
it
would
do
it
over
and
over
again,
so
it
would
put
strain
on
the
survey
instead
of
an
error
it
would
run
in
in
some
backup
at
some
point
would
not
be
like
it
would
could
still
harm
other
clients
if
it
escalates.
C
H
You
just
I
like
rejection.
If,
if
we
remove
the
duplicates,
how
would
we
feel
about
making
that
default
behavior
in
120
that
can
be
disabled?
We
have
precedent
for
that
with
introducing
a
pre-deprecated
feature
gate,
and
so
we
would
turn
it
on
pre-deprecate.
It
indicate
we're
gonna
keep
it
until.
C
Yeah,
I'm
fine
with
that,
I
think
really
like.
I
don't
think
it's
going
to
break
very
many
people.
I
I
mean,
I
don't
think
it's
going
to
break
any
anybody,
so
this
is.
This
is
mostly
a
safety
safety
mechanism.
C
To
go
talk
to
evan,
but
but
I
I
mean
I
mean
if
we
just
remove
the
duplicate.
I
think
that
should
still
continue
to
function
right.
I
hope
so.
C
H
I
think
we
lost
hawaii,
I
did
quickly,
I
can't
hear
highway
at
all.
I
did
look
up
those
ssa
merge
strategy
and
it
says
merge
key
is
you
would
and
the
patch
strategy
is
merged.
I
don't
really
know
what
that
means,
but
I
think.
C
Yeah,
that's
actually
that's
actually
not
the
ssa
directive.
That's
the
old
strategic
merge
patch
directive,
so
I
I
do
think
the
dupes
we
saw
were
not
just
uid
duplicates.
They
were
everything
duplicates
yeah
ssa
can
have
multiple
key
fields.
I'm
not
sure
if
this
is
useful,
so
we
should
probably
figure
out
what
exactly
we
want.
The
the
ssa
flags
to
be.
G
C
C
Yeah,
so
a
set
would
mean
that
it
considers
it
unique
if
any
field
is
different,
which
is
not
the
case
that
we
we
observed.
G
Okay-
okay,
sorry
about
that,
okay
yeah
I
can
I
can,
I
can
make
it
drop
the
field
and
have
a
date
on
it,
and
you
are
proposing
that
we
should
validate.
We
should
detour
if
all
the
fields
are
same
or
if
only
if
the
uid
are
the
same.
I
I
thought
we
can
just
do
from
the
uid
if,
besides
uid
fields
are
different,
that
also
doesn't
make
sense.
C
It
doesn't,
but
I
think
it
I
feel,
like
it's
safer
to
dedupe
on
all
the
fields,
not
just
the
uid.
Okay,
yeah
yeah,.
E
E
Yeah
I
mean
like
that's
one
way
to
split
that
because,
like
if
you
were,
if
you're
doing
exact,
if
you
were
intentionally
using
this
and
you
were
modifying
some
fields,
that
actually
did
make
a
difference,
although
I
guess
I'm
struggling
a
little
bit
in
the
use
case,
that
might
be
more
likely
to
contain
those
ones
that
we
were
worried
about
before,
whereas
exact
duplicates.
I
think
it's
pretty
clear-cut,
like
there's
no
value
in
an
exact
duplicate
on
an
owner
reference.
C
E
C
E
I
Okay,
if
we
only
do
on
exact
duplicates,
do
we
have
to
define
the
other
fields
as
keys
for
suicide
prior.
G
Maybe
double
check,
I
will
I
always
go
one
of
my
my
question
is:
I
was
going
to.
I
need
to
talk
with
my
folks
about
how
the
match
field
the
field
manager,
what
should
be
the
field?
Manager's
opinion
about
api
server,
dropping
field
from
the
request,
because
currently
I
think
the
field
measured
district
ignores
any
server.
C
Defaulting
yeah,
I
think
we
would
pre-treat
the
request
so
that
field
manager
should
see
the
request
after
the
duplicates
have
been
dropped.
Okay
and.
G
My
last
question
is,
I
guess,
like
confidence,
since
we
are
proposing,
we
will
put
a
data
on
it.
We
won't
cherry-pick
this
behavior
to
create,
although
I
would
prefer
we
cherry-picked
this
curious
release,
but
I
guess.
H
H
C
Yeah,
I
think
the
most
we
could
do
is
make
it
opt
in
and
119.
F
C
C
Actually,
I
I
I
it
came
to
our
attention.
I
don't
I
don't
know
if
I
want
to
say
exactly
what
the
what
the
route
was.
G
C
Jordan's
been
tagged
in
this
in
this
issue,
so
he
knows
about
it
all
right.
Shall
we
go
on
to
our
last
topic,
or
do
we
have
more
to
cover
on
this
one?
I
think
we
got
the
answers
we
need.
Are
you
good?
How
are
we?
Yes?
Thank.
A
You
very
good
well,
thank
you
daniel!
So,
let's
make
it
super
quick,
so
everybody
can
go
back
to
do
their
other
things
just
wanted
to
go
through
this
list.
These
are
the
open
encasements
that
we
have
for
120..
A
I
don't
think
we
have
all
the
owners
here
in
the
call
I
spoke
with
anduan,
and
I
know
joe
and
jenny
are
here
so
looks
like
built-in
api
type
defaults,
it's
good
for
120.
huawei!
You
are
the
owner
of
sorry
yeah,
it's
good!
Okay!
Thank
you!
Joe
hawaii
cube
api
server,
identity.
A
Okay,
the
other
five.
I
don't
think
we
have
wojciech
here
in
the
call,
so
I
can
probably
check
with
him
offline
same
thing
with
the
other
owners,
unless
I
I'm
not
seeing
somebody
in
my
screen
and
or
anybody
in
the
call
knows
the
status
of
the
other
ones.
C
I
will
the
only
one
I
don't
know
anything
about.
Is
that
last
one
I
don't
know
what
that
is
or
why
it's
ours.
C
So
I
can,
I
mean
it
looks
like
a
stick
storage
thing,
so
I
don't
know
why
we're
tagged
in
it.
C
Maybe
just
accidental.
A
H
A
Well,
this
is
what
I
can
do
I
can
try
to.
I
can
put
a
comment
here,
removing
ourselves
and
saying
that
if
you
know
they
feel
that
that
is
incorrect,
they
can
reach
to
us.
So
we
have
one.
Yes,
whatever
okay.
A
C
A
Last
one
remember:
we
are
a
little
bit
more
than
a
week
away
from
code.
Freeze,
I
put
a
link
here.
You
have
all
the
dates
so
try
to
keep
that
in
mind
and
then,
if
anybody
in
this
call
or
anybody
you
know
of
the
owners
of
these
features
or
people
that
has
collaborated
is
interested
in
writing
a
blog
post
for
120
under
feature,
please
reach
out
to
me
or
to
daniel
or
to
david,
but
you
know
start
with
me.
A
I,
the
basic
contributor
c,
is
interested
in
you
know
finding
those
people
so
keep
that
in
mind.
I
will
try
to
remember,
remind
the
sea
a
couple
more
times.
Okay,
but
it's
all.
I
had.