►
From YouTube: Kubernetes SIG Multicluster 2020 May 26
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).
B
B
B
B
A
A
So
one
of
the
things
that
we
are
seeing
the
two
IQ
work
we
are
using
confettis
is:
we
are
mainly
using
two
federal
resources
right
to
manage
where
resources
are
being
deployed,
which
clusters
and
and
a
bit
managing
the
lifecycle
and
but
while
when
trying
to
investigate
deeper
into
into
the
core
qfn
I
found
that
that
is
like
service
DSS
record
and
has
been
shown
in
many
demos
on
qfn
I.
My
bedroom
with
this
controller
is
like
is
doing
the
task
that
probably
at
the
time
when
kill
said
appeared,
was
quite
missing.
A
It
might
be
a
really
an
important
feature
right
now,
but
I
don't
know.
It's
only
me
or
I
I
see
that
there
are
many
other
services
on
government
as
eco
system
that
could
be
used
regardless
that
matter
of
having
services
across
multiple
classes
and
how
these
will
import/export
that
make
a
reference
to
to
the
services
cap
by
Jeremy,
so
yeah
I
mean
I,
think
maybe
fits
of
the
past.
B
The
the
folks
from
Huawei
were
interested
in
having
that
feature
again
in
in
queue
fed
or
Federation
v2
I
I
will
just
call
out
that
that
feature
like
all
the
other
cheap
fed
features
is
alpha.
So
I
I
completely
understand
the
question,
since
we
are
doing
work
that
has
a
high
degree
of
overlap
with
it.
B
My
original
hope
had
been
that
we
would
that
we
would
build
things
that
you
could
take
or
leave
as
you
wanted,
without
installing,
like
a
big
monolith
that
gave
you
a
bunch
of
stuff,
so
I
think
the
way
that
we've
been
doing
the
multi
cluster
services
and
the
work
API
stuff
is
a
lot
more
representative
of
what
I
had
originally
hoped
to
do
with
Federation,
so
I
for
me
personally,
if
it
came
out,
I,
don't
think
that's
a
loss.
All.
A
Right
yeah
I
mean
table
to
see,
but
yeah
I
make
sense,
make
sense.
Yeah
I
will
sell
it.
We
I
mean,
if
that
comes
to
to
the
time
being,
a
kind
of
removing
it
or
duplicated
I
will
I
will
sell
it
for
sure
to
the
rest
of
the
of
the
people
in
the
channel
are
all
showing
the
teens
on
the
river
and
the
other
point
that
kind
of
related,
but
not
fully
its.
A
A
So
that
is,
is
there
and
you
you
can
use
it,
but
then
the
next
step,
where,
when
you
need
troubleshooting
resources
to
deploy
across
multiple
classes,
which
so
that
Cupid
or
at
least
the
way
of
the
Federative
resources
structure
needs
you
have
a
property
or
field
in
order
to
be
able
to
troubleshoot
any
status
and
what
I
mean
the
status
mean
an
example.
It
could
be
deployment
right.
A
You
know
deployment
you,
you
need
to
know
whether
you
have
all
the
replicas
running
or
maybe
more
replicas
now
will
ever
or
maybe
not
all
the
replicas
that
have
been
created
because
once
spending
to
be
a
schedule
and
stuff.
So
of
course
the
idea
behind
these
is
to
improve
the
troubleshooting
of
the
resource,
including
somehow
the
status
of
all
these
resources
or
lease
the
resources
that
they
are
not
shall
fear.
A
No,
no
ready
in
order
to
do
that,
I
have
caneva
I
kept,
while
almost
on
is
to
rely
on
some
new
changes
that
has
been
normalized
or
standardized
in
order
to
use
conditions
and
a
certain
interface
on
kubernetes
to
determine
when
our
resources
is
when
a
condition
is
true,
so
that
you
can
easily
independently
of
the
of
the
type
of
the
result
you
can,
you
can
be
sure
to
determine
if
it's
ready
or
not
so
yeah
you
have.
To
sum
up,
the
question
is:
has
ever
come
into
into
the
thoughts
of
queue?
B
B
I,
don't
think
we
need
to
make
a
snap
decision
here,
but
we
should
talk
about
what
we
think
is
the
best
thing
to
do
so
for
from
just
my
own
subjective
opinion,
I
think
cluster
registry,
as
an
idea
is
like
deceptively
simple
to
kind
of
imagine
and
then,
when
you
try
to
build
whatever
you
think
that
is
it's
much
much
harder.
I
have
come
to
think
that
it
was
the
wrong
starting
point
for
the
work
that
we
previously
did
to
start
with.
A
registry
I
would
much
rather
work
backwards.
B
So
if
we
play
the
movie
forward
and
I,
don't
think
Valerie's
here,
that's
too
bad
since
she's
been
working
on
it.
But
if
we
play
the
movie
forward
on
like
the
work
API
stuff
that
Valerie
has
been
has
been
doing
or
driving,
I
could
see
the
concept
of
a
registry
being
involved.
If
we
play
that
a
couple
steps
forward
and
think
about,
how
do
we
schedule
work
to
happen
where,
at
the
same
time,
I
frankly
like
do
not
think
that
the
the
current
state
of
cluster
registry
is
especially
useful
to
anybody?
B
C
B
I
and
I'm
completely
with
you,
that's
that's
sort
of
what
I
was
thinking
when
I
said:
I
wouldn't
want
the
current
state
of
cluster
registry
to
like
weigh
down
any
of
our
current
work.
I
I'm,
not
sure
that
I
think
it
should
be
archived
like
right
at
this
moment,
but
at
a
minimum.
We
should
probably
put
a
note
on
the
cluster
registry,
explaining
what
the
history
was,
and
that
may
be
explaining
that
we
may
be
working
back
up
to.
It
is
something
that
we
could
do.
B
Kubernetes
SIG's
didn't
exist
when
cluster
registry
was
created,
so
it
might
be
something
to
do
to
just
archive
the
one
that
we
have
with
a
note
explaining
the
history
and
if
we
want
to
build
something
like
that,
we
can
make
a
new
one
in
kubernetes
SIG's,
just
some
food
for
thought.
I.
Definitely
think
this
is
something
that,
like
we,
shouldn't
unilaterally
do
or
shouldn't
do,
based
on
one
conversation
in
one
sig
meeting.
C
Awesome,
thank
you
yeah.
So
our
last
few
conversations
have
been
really
productive,
so
I
figured.
It
would
be
good
to
to
kind
of
keep
that
going
here.
So
last
week
we
got
the
the
first
big
PR
merge
for
the
cap,
which
is
great
and
then
I
posted
another
one
with
a
bunch
of
updates.
Based
on
the
conversations
we've
had
here.
C
B
B
C
Like
it
cool
so
yeah,
so
the
service
export
hasn't
changed
much
I
think
in
the
last
little,
while
it's
basically
just
a
status
with
conditions,
but
I
think
some
of
the
conditions
have
kind
of
evolved.
So
I
just
wanted
to
call
out
that
yeah
the
actual
structure
I,
don't
want
to
spend
too
much
time
on,
because
the
plan
is
to
use
the
the
new
standard
conditions
model
from
kep
1623
as
soon
as
that's
implemented.
C
Unless
there's
a
really
good
reason
not
to
it
seems
like
we
should
just
kind
of
follow
suit
and
and
go
with
what
what
will
hopefully
be
the
new
standard.
But
so
there's
a
few
conditions.
We've
kind
of
been
talking
about
I
want
to
start
from
the
bottom,
actually
because
I
think
it's
really
a
headless
conversation
that
we
had
a
couple
weeks
ago.
C
So
you
know
get
to
how
we
can
show
that
on
a
service
import
layer.
But
I
think
basically
the
thinking
is
that
a
ace
we
should
have
a
condition
for
headless
that
is
true
or
false,
based
on
the
derived
state
of
that
service
import,
but
also
that
this
is
a
good
place
to
kind
of
service.
If
there's
any
disagreement,
because
we
are
opening
up
this
opportunity
for
for
there
to
be
a
mismatch
potentially.
C
Think
that's
that's
kind
of
interesting
because
the
headless
services
like
a
service
is
gonna
be
headless
or
not.
It's
probably
a
mistake.
If
you
have
any
disagreement,
I'm
not
sure
I
really
see
a
case
for
you
know
where
you'd
want
to
have
a
service
that
that
you
know
becomes
headless
or
becomes
a
supercluster
I'd
be
like
that
really
starts
feeling
like
a
new
service.
So
this
would
be
kind
of
an
error
state
that
you
that
you
want
to
have
some
some
way
to
diagnose
this
issue.
When
you
create
a
service
export,
you
look
at
status.
C
B
C
C
D
The
service
x4
tends
to
be
written
by
some
person
or
from
the
human
derived
side
right.
It's
it's
some
human
or
CI
system.
That's
indicating
the
services
to
be
exported
I've,
not
seen
high
fidelity
of
humans
using
owner
FS
properly.
Are
we
assuming,
then
that
we
would
come
back
through
and
write
the
owner
F
from
the
controller?
It's
not
I,
don't
know
if
it's
the
worst
thing
ever
I
just
don't
know
what
the
precedent
there
is
yeah.
D
It
seems
a
little
bit
sketchy
to
me
that
we
would
come
through
on
something
that
you
know
if
I
went
and
I
created
this
thing
and
you
come
in
and
say,
yeah
well,
I'm
actually
gonna
take
it
over
and
you
know
if
you
do
something
else,
I'm
gonna
delete
this
thing
for
you,
it
doesn't
I,
don't
know
it
doesn't
feel
right
to
me.
Yeah.
A
B
Can't
readily
think
of
any
precedent
for
an
operation
like
that
off
the
top
of
my
head
and
since,
in
my
own
head,
it's
entirely
possible
that,
like
service
and
service
export
might
be
delivered
by
different
pipelines.
It
seems
like
we
probably
don't
want
one
of
those
things
being
deleted
to
make
the
other
get
G
seed,
like
that's
nice
like
breaking
the
boundary,
but
between
pipelines
that
would
be
delivering
it.
Yes,.
D
B
C
D
So
so
I
went
through
the
your
P
ARDS
this
morning
and
I
guess
you
haven't
seen
my
comments
yet
I
have
a
general
question
on
conditions.
It's
not
just
for
you.
It's
for
David
and
everybody
who's,
trying
to
write
about
conditions
which
is
sort
of
what
principle
makes
us
decide
whether
something
is
a
condition
or
not
like
it
feels
like
invalid
service
type,
isn't
a
condition.
It's
really
just
a
status
field
like
I,
don't
know
I'm
applying
my
own
sort
of
handle,
yeah.
B
D
Asked
David
in
his
writing
to
think
about
that
and
I
know.
Evan
Anderson
from
the
key
native
world
is
trying
to
also
write
about
conditions
and
I.
Keep
asking
the
same
question
not
getting
a
really
good
answer
like
what
is
the
principle
that
stops
me
from
turning
every
single
status
field
into
it
into
a
condition.
C
Yeah
I
mean
the
one
so
yeah
I,
don't
think
there
really
is
a
consistent
principle.
I
think.
Normally
things
become
conditions.
If
they,
you
know,
if
it's
basically
a
bullying
that
has
some
human
consumable
message
and
may
or
may
not
actually
need
to
be
exist
so
like
it
may
be,
it's
a
condition
if
it
if
it
doesn't
always
apply,
but
then
there's
things
like
you
know:
pod,
ready
and
whatnot.
That
are
obviously
that
don't
follow
that
at
all
I,
don't
I,
don't
know.
There's.
C
D
Some
clear
anti
patterns
and
how
it's
been
used
in
the
past,
but
I'm
really
looking
for,
is
how
we
should
continue
to
use
or
AB
use
it
in
the
future.
Like
my
my
feeling
and
I,
don't
know
how
much
traction
this
is
going
to
get,
but
the
principle
that
I
sort
of
go
with
is
similar
to
what
you
just
said.
If
it's
optional
condition
might
be
reasonable,
if
it's
something
you
can
state
about
every
single
instance,
it
probably
doesn't
need
to
be
a
condition.
D
C
So
I
think
I
think
it's
more
than
that,
though.
It's
also
that
that
human
human
readable
message
right
like
if
because
because
I
agree
with
you
about
time
time
stamps
you
know,
maybe
if
every
field
should
have
a
time
stamp
and
that's
something
we
care
about.
We
should
that's
that's
a
separate
problem
assault
but
but
conditions
also
provide
you
know
the
the
human
readable
message
as
well
and
that's
something
that
we
probably
wouldn't
want
on
every
field.
That's
what
it's.
C
D
So
it
export
it
is
actually
the
interesting
one.
The
timestamp
is
interesting
right
because
exported
is
not
a
one-way
thing
right,
like
you're
exporting
continuously.
Every
time
this
end
point
slices
change.
You
need
to
re
export,
so,
like
I,
think
I
asked
in
one
of
the
drafts
of
the
Catholic.
What
does
the
exported
really
mean
like
there?
Isn't
it's
not
a
one-way
latch,
so
the
time
stamp
there
I
think
is
what
makes
export
and
interesting
otherwise
I
would
say
export.
It
doesn't
belong
at
all
because
it
actually
doesn't
have
any
consistent,
meaning
right.
C
I
completely
agree,
and
so
the
timestamp
of
the
last
change
and
if
there's
anything
any
clusters
that
it
wasn't
exported
to,
for
example,
it
might
be
something
that
you
want
to
communicate
there,
but
still
that's
a
case
where
this
this
seems
like
you'll,
always
have
this
on
every
on
every
export.
So
maybe
maybe
it
isn't
a
condition
like
I,
don't
think,
there's
anything
wrong
with
having
an
exported
time
stamp
right.
C
B
C
So
so
I
guess
the
things
that
I
did
I'd
care
about
that
kind
of
trying
to
rephrase
what
you're
saying
are
probably
like
the
time.
Obviously,
the
time
it
was
last
read,
probably
the
observed
generation
from
the
controller
and
then
whether
or
not
it
was
successful
like
if
there
was
an
error
and
if
there
wasn't
what
the
error
was.
D
So,
in
the
interest
of
time,
I'm
gonna
propose
that
we
should
actually
set
up
a
formal
API
review
like
I.
Don't
think
I've
gone
through
this
with
my
fine
toothed
API
reviewer
hat
come
on
wait
comb
hat
anyway
and
it
might
actually
make
sense
to
get
someone
who
isn't
as
close
to
this
as
I
am
to
review
at
the
API
point
of
view.
I
don't
want
to
volunteer
anybody,
but
Jordan
pops
to
mind,
so
it
might
make
sense
to
schedule
in
the
very
near
term,
like
a
more
comprehensive
like.
D
B
B
B
This
feels
like
so
like
with
the
strange
attractor
of
like:
what's
the
decision
plane,
here's
one
thing
that
was
in
my
head,
as
we
were
like
talking
about
that
question
I
feel
like
if,
if
you
start
out
like
in
your
API
design
and
say
something
is
like
a
status
field
and
then
you
add
the
fields
for
conditions
pragmatically,
it
should
probably
just
be
a
condition
and
I
could
see.
I
could
see
this
being
a
condition.
C
Yeah
this
one
seems
like
it:
it's
usually
not
going
to
be
there
like
something
went
wrong.
This
is
a
mistake
that
you
know
that
kind
of
only
required
some
time
part
of
conditions
like
invalid
service,
yeah
I
agree.
It
seems
like
a
condition
like
most
of
the
time,
any
time
you've
deployed
anything
properly
or
you
didn't
use
external
name.
This
would
be
it.
It
would
be
a
valid
service,
so
this
doesn't
need
to
always
be
on
status.
That's
that's
kind
of
my
take
I.
B
B
D
B
C
D
I
I
know
that
David
with
his
standardized
condition,
stuff
and
I-
know
Eric,
toon
and
others
are
trying
to
come
up
with
actual
guidance
for
like
when
and
how
to
use
conditions
and
I
will
align
with
whatever
we
decide.
Overall
I'll
disagree
and
commit
at
some
point,
but
we
haven't
actually
got
consensus
on
it.
Yet
I
think
this
is
fine.
We
can
leave
it
the
way
it
is
for
now,
but
as
we
proceed,
we
should
keep
an
eye
on
that
discussion
because
I
think
you're
an
exemplar
of
it
right.
C
But
agreed
that
we
shouldn't
do
that
now
we
should
have
a
kind
of
follow-up
on
on
what
that's
looking
like
and
then
the
last
one
on
here,
just
cuz,
it's
been
in
here
since
the
beginning.
Is
this
initialized
and
honestly
I?
Don't
really
think
this
adds
a
lot
of
value,
I
think
the
the
other
three
kind
of
communicate,
the
information
that
you'd
want
like
if
a
service
is
not
exported
and
there
was
an
error-
that's
where
you'd
have
it
communicate
any
initialization
problems,
I.
C
D
D
C
C
It
will
be
true
if
everything
is
headless
and
it
will
be
false,
otherwise
it
will,
but
it
only
needs
to
have
any
message,
and
maybe
this
is
a
better
event
if,
if
there's
any
disagreement,
this
is
something
where,
like,
if
I,
create
a
service,
export
and
I
have
a
cluster
ID
and
I'm
exporting
a
cluster
I'd
be
serviced
to
an
otherwise
headless
super
cluster
service.
I
would
want
to
know,
basically
this
be
able
to
look
at
that
export
status
and
see
that
I'm
breaking
something.
D
C
I
mean,
but
presumably
this
isn't
it.
This
is
a
short-lived
scenario,
because
I
really
can't
see
a
case
where
this
would
be
like.
You
wouldn't
want
this
right.
So
so
yes,
it
would
be
chatty,
but
something
has
gone
very
wrong
and
and
like
your
deployment
is,
is
almost
definitely
broken
somewhere
and
not
doing
what
you
expect.
If,
if
you
have
a
disagreement,
so
let.
C
I
think
so
putting
it
on
import,
because
I
was
thinking
about
this
too,
like
what
so
this
this
is.
This
is
not
what
we
would
check
to
see
if,
if
the
consumer
should
treat
it
as
a
headless
service
right,
like
I'm,
assuming
DNS
or
anything,
would
use,
you
know-
and
cue
proxy,
of
course,
will
use
the
service
import.
But
this
is
when
I
create
a
service
export
right
after
creating
and
if
I
look
at
the
status
and
see
see
what
it's
done,
I'd
want
to
see
if
I
basically
broken
anything.
C
Otherwise,
if
we
put
this
status
on
service
import,
which
would
be
an
option,
I
would
have
to
then
go
check
an
import
after
creating
an
export
to
see
if
it's
still
functioning
and
that
that
that
seems
wrong.
Like
I
would
like
to
look
at
the
thing
I
created
to
know.
If,
if
its
functioning,
the
way
I
expect.
B
C
D
Okay,
so
again
we
have
a
I,
don't
know
a
unstated
assumption
that
somebody's
gonna
go
off
and
look
at
this
versus
something
like
an
event
which
would
be
reoccurring
right,
like
right
at
least
an
event.
I
can
throw
monitoring
against
well,
like
I
could
say,
alert
me
if
there's
a
number
of
events
right
or
a
famous
any
events
of
this
type,
I,
don't
know
I'm
just
thinking
out
loud
again.
It
feels.
B
D
D
No
I
just
don't
really
know
what
the
done
thing
is
anymore,
because
I've
lost
track
of
all
the
different
ways.
People
are
using
and
abusing
conditions
yeah.
So.
D
Is
enough,
medicine
like
that
seems
like
this
would
be
a
match.
Here
would
be
the
controller
again.
They
don't
need
to
be
mutually
exclusive.
That's
a
good
point,
but
the
controller
would
be
throwing
an
event
against
the
service
exports.
Saying:
hey,
you're,
the
your
your
oh,
it's
the
right
phrasing
you
are
in
the
you
are
being
overridden
your
preference
is
being
over
in
them
right
yeah.
It's
not.
C
D
D
C
I
mean-
maybe
maybe
we
do
need
to
document
this
this
event,
because
it
seems
like
like
a
more
critical
one
but
yeah
I.
Think.
B
So
since
we
just
have
a
few
minutes,
left
Jeremy
I'm,
not
sure
if,
if
you
ever
recounted
to
the
group
like
what
what
type
of
reception
you
had
gotten
from
engaging
sidon
Signet
work
about
changes
to
the
cube
proxy.
Maybe
we
went
over
that
last
time,
but
and
if
we
did
that's
my
bad
but
I
wanted
to
make
sure
that
we
had
read
out
the
latest
status
of
that
and
yeah.
C
No,
we
haven't
gone
over
it
actually
because
that
meeting
with
them
was
right
after
our
last
the
last
time
we
talked
about
this,
so
reception
was
pretty
good
for
having
cube
proxy
understand
the
CR
D.
So
this
means
we
would
have
an
out
of
tree
client
and
Ciardi
for
the
MCS
API
will
probably
have
to
set
up
in
staging
so
that
it
can
be
consumed
by
Q
proxy,
but
basically,
yeah
Q
proxy
will
will
understand
service
import,
the
Sierra
TF
present
and
configure
supercluster
IP
routing
as
necessary,
just
like
it
does
with
service.
B
C
Also
I
think
that's
the
direction
that
things
are
going
right
like
we.
You
want
more
more
core
ish
concepts
to
be
as
long
as
they're
optional
to
be
expressed
by
a
CRT,
so
it
kind
of
makes
sense.
I
guess
the
the
one
thing
to
be
aware
of
is
I
believe
will
be,
will
be
the
first
example
of
this,
but
hopefully
not
the
laughs,
but
I
think
we're
also.
C
C
B
C
This
meeting
I
know
I
think
this
was
really
productive.
I
think
this
is
good,
I
think
anything
else,
I
think
yeah.
We
should.
We
should
probably
set
up
a
more
detailed
API
review.
You
know,
obviously
sooner
would
be
great
because
I
think
every
time
we
get
together
these
discussions,
they've
been
very
helpful.
I,
don't
know
what
makes
the
most
sense
for
for
doing
that.
How
we
go
about
that
I.