►
From YouTube: Kubernetes SIG Multicluster Jul 13
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
C
C
I
was
resolving
to
do
these
so
I'll
go
ahead
and
do
it
now
welcome
everybody
to
july
13th,
yes,
13th,
meeting
of
kubernetes
multi-cluster,
laura
laurens
dominant
on
the
agenda
today
go
ahead.
Laura.
E
Thank
you.
Hopefully
you
guys
are
seeing
the
meeting
notes.
Yes,
okay,
cool,
crystal
clear,
crystal
clear,
great
okay,
so
I
have
two
things
I
wanted
to
give
some
quick
kept
updates
and
then
I
wanted
to
follow
up
on
what
jeremy
brought
up
last
time
about
multi-cluster
controller
coordination.
Some
changes
changes
going
on
in
leader
election
package,
blah
blah
blah.
Naturally
I
made
slides
for
that
too,
so
we
can
look
at
them
all
together
but
anyway,
so
the
quick
cap
updates
for
multi-cluster
dns.
E
I
finally
addressed
some
reviewer
comments
from
like
way
back
in
april,
and
I
don't
know
if
miguel
is
on
the
call,
but
he
took
a
look
and
made
some
comments
to
you.
So
thank
you
for
looking,
so
I
need
to
do
another
round
of
requesting
approvals
again,
so
I
sent
a
note
to
john
and
for
cluster
id,
I'm
catching
up
on
quite
a
large
backlog
of
action
items.
E
Still
I
started
with
updating
mcs
api
grad
criteria,
which
we
talked
about
back
on
5
11,
so
finally
getting
that
in,
but
the
next
steps
these
have
been
talked
about
before,
but
just
sort
of
reiterating
for
update
purposes
change.
E
Cluster
claim
to
cluster
property
everywhere,
actually
in
the
cap
cube
builder
and
a
copy
of
it
go
of
it
into
this
repo
that
we
made
make
the
api
review
request
and
also
circle
back
on
that
crt
bootstrapping
stuff
that
what's
happening
in
sig
arch
again,
I'm
in
the
slack
channel
about
it,
but
I've
been
out
of
office
and
that
group
has
also
been
alternately
out
of
office.
E
So
it
hasn't
moved
that
far
since
we
last
talked
about
it
a
while
ago
so
happy
to
take
any
questions
about
that
too,
but
just
wanted
to
give
a
quick
update-
and
this
also
helped
me
put
all
of
my
action
items
in
one
place
instead
of
in
90
different
places.
So
that
was
also
beneficial
and
then
from
more
of
the
discussion
side.
I
want
to
open
up,
as
mentioned
jeremy
brought
up.
This
pull
request.
E
Yeah
not
materially,
since
I
gave
my
last
update
well,
I
guess
one
thing
that
happened
since
the
last
update
was
which
was
on
the
announcement
thread.
So
possibly
people
saw
it,
but
if
you
us,
what
am
I
trying
to
say
announcement?
There's
a
sig
architecture.
E
That's
spread
that
cc'd
sig,
no
sorry,
sig
api
machinery,
red
that
cc'd
sig
multi
cluster
and
the
latest
was
that
team.
Saying:
okay,
we're
gonna
write
a
cap,
we're
also
that
is
going
to
focus
on
this
this
and
that-
and
this
is
about
being
installation
using
the
controller
manager
specifically
and
they're,
also
working
on
a
poc
related
to
that
to
to
see
what
they
can
write
about
it.
You
know
what
I
mean
yeah,
but
then
they
were
out
of
office
and
then
I
was
out
of
office.
B
Page
because
my
impression
is
that
really
all
we're
waiting
on
there
we're
not
waiting
on
a
like
definitive
solution,
just
just
the
idea
that
we
will,
we
will
eventually
be
able
to
bootstrap
crds
if
necessary,
and
we
don't
necessarily
need
that
as
a
blocker.
We
just
need
to
know
that,
like
one
day
that
will
be
a
possibility,
then
we're
not
closing
doors
right.
E
Yeah
yeah,
I
think
that's
true
and
I
think
the
problem
that
we've
had
is
that
this
that
that
felt
to
be
the
case
without
anything
like
concrete
behind
it,
and
then
this
initiative
has
tried
to
put
something
more
concrete
around
it
and
has
received
a
lot
of
pushback
for
the
ideas
that
are
being
presented
and
then,
but
not
a
lot
of
sort
of
opening,
for
what
is
you
know
going
to
work
for
this
case?
You
know
what
I
mean.
E
I
don't
know
if
I
said
that
very
clearly,
but
if
there's
some
more
no's
happening
to
things
than
yeses
or
yes,
let's
try
this
next,
and
so
I
think
our
interest
is
to
make
sure
that
either
that
yes-
and
let's
keep
talking
about
the
next
thing
is
still
happening
or
get
a
like
confront.
The
situation
be
like
it
sounds
like
there's
a
lot
of
no's
going
on.
Does
that
mean
that
this
is
no
longer
a
pressure
valve.
A
B
I
think
that's
a
that
sounds
like
a
pretty
reasonable.
Take
my
I
guess.
My
the
the
other
thing
that
I
want
to
keep
in
mind
here
is
like
we
don't
actually
have
like
an
a
concrete
use
case
for
requiring
bootstrapping
of
cluster
property.
Yet
right
so
I
also
like
I.
I
think
we
definitely
need
to
make
sure
that
we're
not
closing
doors
unnecessarily,
but
I'm
I'm
also
want
to
be
cautious
that
we're
not
you
know,
creating
blockers
that
don't
necessarily
need
to
be
blockers.
C
Yeah
yeah
and
I
think
that's
the
right
state
of
mind
to
be
in.
C
Are
we
so?
Are
we
going
to
talk
about
this
now
we're
going
to
talk
about
the
leader
election
thing
like
I
guess
we're
talking
about
this
yeah
like
there
isn't
anything
about
multi-cluster
service
discovery
or
any
other
thing
that
we
have
in
progress?
That
means
that
we
have
to
have
cluster
property
crd
installed
as
part
of
cluster
bootstrap
right
like.
E
That's
the
same
as
when
jeremy's
saying
we
don't
have
any
known
use
cases
like
they're,
not
concrete
and
but
and
we
have
a
we
have
a
vibe.
I
guess
and
bible
was
not
enough.
E
E
Yeah
to
change
the
mechanics
of
anything,
but
I
I
totally
understand
what
y'all
are
saying.
I
don't
think
we
need
to
block
a
ton,
but
I
do
think
there's
interest
from
all
of
us-
and
I
know
tim
is
interested
too
and
keeping
a
pulse
on
it.
It's
not
the
biggest
item
on
the
agenda
on
the
action
item
backlog
in
my
opinion,
if
that.
B
E
E
Maybe
in
I
guess
we're
going
to
bi-weekly,
but
in
two
times
or
so
is
probably
my
guess.
E
If
that's,
I
see
my
little
audio
going
up
a
lot,
but
I'm
going
to
change
the
topic
to
the
prioritized
leader
election
and,
as
mentioned,
I
put
together
a
few
slides
of
what
I
think
is
going
on
what
I
think
the
questions
need
to
be
but
feel
free,
especially
jeremy,
since
you
brought
it
up
to
interject
if
I'm
off
base,
but
just
want
to
sort
of
seed,
an
intro
for
everybody
based
on
what
I
know
and
then
try
and
like
guide
the
discussion
a
little
bit
so
yeah,
okay,.
B
B
This
is
like
a
seems,
like
a
pretty
interesting
use
case
like
you've
got
controllers
running
in
clusters
all
over
the
world,
and
you
want
to
prefer
the
geographically
closest
region,
but
it
gets
kind
of
you
know,
there's
a
bunch
of
nuance
here
like
if
you're
this
this
seems
reasonable
and
this
and
this
would
work,
but
if
the,
if
the
controller
is
in
the
same
cluster,
you
know
if
you'd
need
multiple
clusters
in
the
region.
B
For
this
really
to
be
valuable,
it
seems
like
if
you're,
if
your
controller
is
just
in
your
asia
cluster,
then
you
know
that
you,
it
would
only
be
available
when
the
cluster's
up
and
it
would
you
know
leader
election-
would
be
tough
within
that
cluster.
B
If
the
cluster's
down
and
if
your
controller
crashed
when
the
cluster
was
up,
it
would
probably
be
rescheduled.
So,
like
there's
a
lot
of
nuance-
and
I
think
this
just
kind
of
drew
attention
to
like
well,
this
pr
seems
like
it
actually
solves
this
nicely.
In
some
scenarios
there
are
a
lot
of
kind
of
cross
region,
cross
cluster
controller
or
workload
patterns
that
I
think
we're
kind
of
surfacing.
B
C
So
I've
been
scanning
this
while
we've
been
talking.
I
definitely
want
more
information
on
like
the
geo
spread,
one
like
closest
to
what
like
is.
What
does
multi-region
controller
mean?
Does
it
mean
that
one
instance
of
the
controller
services,
multiple
regions
like
I,
I
have.
B
So
many
questions,
so
I
can
answer
a
bit
of
this
because
I've
been
speaking
with
the
istio
folks
and
that's
where
this
is
coming
from
originally
as
part
of
their
support
for
mcs.
B
But
the
idea
is
like
istio
has
decentralized,
or
it
has
a
bunch
of
different
controllers
that
are
that
use
leader
election
to
figure
out
which
clusters
they
they
manage.
B
So
there's
a
bunch
of
different
control,
plane,
workloads
that
are
responsible
for
managing
the
multi-cluster
services
across
clusters
and
the
the
trick
for
them
is
figuring
out
which,
which
instance
should
be
responsible
for
configuring,
the
actual,
like
the
actual,
the
actual
istio
configuration
each
cluster,
and
so
that's
where
they
they
need
some
leader
election,
and
you
know
you
could
use
naive
leader
election
with
each
cluster.
But
you
end
up,
you
know
if
you
have
a
global
deployment
with
with
clusters
all
over
the
world.
B
You
end
up
in
a
scenario
where
you
know
your
us
instance
is
controlling
europe.
Your
asia
instance
is
controlling
the
us
and
it
just
you
know
it
would
still
function,
but
it
it
gets
kind
of
ugly.
C
C
Is
there
any
issue
like
in
within,
like
in
istio
repo,
that
we
can
learn
more
about
this,
like
the
context
you
just
gave
was
very
helpful,
I
wonder
if
it's
written
down
anywhere,
it's
not.
It
doesn't
appear
to
be
present
like
in
the
pull
request
at
all.
B
So
I'll
I'll
see
if
I
can
dig
it
up
and
I'll
I'll
ping,
some
of
them
to
to
join
next
week-
or
I
guess
two
weeks
ago
as
well.
C
Cool
all
right,
I
didn't
mean
to
interrupt,
please
feel
free
to
proceed.
Laura.
E
Cool
yeah
all
right
thanks
for
the
background
jeremy,
a
little
of
my
own
background
too,
for
some
flavor
of
what's
going
on
here,
specifically
besides
that
general
background,
this
specific
pull
request
is
to
edit
the
leader
election
package
to
do
something
kind
of
generic
provide
a
place
for
people
to
put
arbitrary
keys
and
a
key
comparison
function
so
that
there
could
be
some
more
custom
logic
going
on
for
leader
election
based
on
whatever
library
user
writes
into
that
key
location
and
whatever
that
key
comparison
function
can
do
so.
E
What
the
discussion,
I
think,
is
what
we
want
to
advise
people
who
are
implementing
these
controllers,
how
to
use
it
either
this
or
if
we
want
to
recommend
something
else,
or
you
know,
I
don't
know
where
this
might
go,
but
if
they're
running
multiple
mcs
controllers
at
once,
so
this
is
kind
of
the
ping
over
to
this
use
case
that
jeremy
got
cc'd
on
in
this
specific
pull
request
and
a
little
bit
more
background.
That,
I
think,
is
helpful.
This
is
like
lifted
from
the
mcs
api
docs.
E
So
this
is
a
reminder
for
everybody,
but
just
looking
for
some
things,
we
say
explicitly
about
the
mcs
controller
in
the
definition
of
the
mcs
controller
and
the
mcs
api.
Sort
of
this
bolded
part
is
the
interesting
thing,
which
is
that
we
say
it
might
be
a
single
controller,
multiple
decentralized
controllers
or
like
a
human.
E
D
E
Endpoint
slices
and
deal
with
cluster
connectivity
issues,
and
then
maybe
you
know
your
leader
election
thing
here
right
so
that
that's
some
general
background
just
to
make
sure
we're
on
the
same
page.
I
think
you
know
some
potential
starting
questions
here.
It
sounds
like
there's
some
interest
and
what
we
were
just
saying
about.
You
know
what's
some
more
background
about
what
istio's
problems
are,
but
I
think,
there's
sort
of
one
direction
this
conversation
could
take,
which
is
for
this
specific
pr.
E
You
know
what
is
there
something
we
might
recommend
in
terms
of
what
this
key
is
or
what
this
key
comparison
function
is
and
then
something
else
I
wanted
to
bring
up.
Is
people
have
been
talking
about
raft
and
then
rash
talks
about
paxos,
which
are
apparently
two
popular
consensus?
Algorithms,
maybe
everybody
else
knows
about
them,
but
they
were
new
to
me
and
since
that
comes
up
a
lot,
I
just
wanted
to
clear
the
air
on
what,
if
anything,
does
that
have
to
do
with
either?
E
You
know
this
in
general,
or
you
know,
from
more
relevantly
from
mcs
api
perspective,
and
one
thing
I
did
like
while
I
was
looking
around
etcd,
has
this
cool
page
with
lots
of
problems
in
diagram
form
that
they
had
to
solve
so
that
just
might
give
us
some
food
for
thought
at
some
stage
as
well.
E
So
that's
kind
of
my
set
the
stage
jeremy.
Do
you
think
these
questions
represent?
You
know
something
you
want
people
to
talk
about
in
this
group.
B
Yeah
yeah,
I
think
this
is
exactly
the
kind
of
thing
I
was
I
was
thinking
about
is
basically
do
we
want
to
either
suggest
or
create
patterns
for
coordinating
multiple
clusters,
and
I
think
you
know
raft
shows
up
as
interesting
because
well
because
fcd
has
a
fairly
raft
implementation
that
seems
like
we
could
do
something
with
to
to
help
kind
of
create
a
standard
deployment
pattern
or
something
or
for
controllers.
B
C
Yeah,
I
I
I
think
that
this
is
a
very
interesting
problem,
that's
relevant
to
our
scope
and
sig.
There's
a
continuum
of
things
that
we
could
do
like
the
the
least
we
can
do
is
nothing
a
something
that
is
probably
helpful.
That
is
short
of
writing
code
is
like
classifying
and
documenting
these
patterns.
So,
like
here's
another
one
that
I've
seen
folks
that
are
implementing
things
that,
like
chat
between
multiple
clusters,
is
like,
if
you're
implementing
some
kind
of
hub.
C
Figuring
out
how
to
like
factor
your
api
so
that
you
don't
absolutely
like
ddos
an
api
server
with
like
status
updates,
is
isn't,
like,
probably
arguably,
an
adjacent
problem
to
the
leader
election,
one
that
I
think
that
we
there's
I
I
would
be
willing
to
bet
you
all
a
crisp
two
dollar
bill
that
and
I'll
show
you
I'll
show
you
which
one
I
bet
you.
C
Look
at
that
that
there
are
enough
folks
that
have
faced
this
that
could
come
and
speak
to
us
about
what
they
did
and
enough
folks
that
are
either
struggling
with
it
now
or
will
struggle
with
it
in
the
near
future
that,
like
the
leader
election
one
and
like
the,
how
do
you
factor?
C
How
do
you
factor
your
status
rights
to
hub
type
things
to
get
the
right
mix
of
responsiveness
and
and
like
overall,
controllable
qps?
I
think
that
that
both
of
those
are
things
we
can
explore
here.
I'd
be
very
interested
in
having
the
the
gentleman
that
opened
the
the
pull
requests
come
and
tell
us
a
little
bit
more
about
the
leader
election
thing,
because
it's
the
kind
of
thing,
for
example,
that
I
think
a
picture
would
be
worth
a
thousand
words.
C
The
background
you
gave
jeremy
was
really
interesting
but
like
when
I
think
of
work,
artifacts
artifacts
that
we
could
produce.
That
would
help
the
community
like
a
one
that
requires
no
coding,
but
is
still
probably
really
helpful,
is
like
categorizing
different
different
classes
of
these
types
of
situations
like
forget
about
the
solution,
but
just
classifying.
C
These
are
the
types
of
problems
that
we've
seen
probably
can
be
very
inspirational,
or
at
least
like
validating
to
people
in
the
space
that,
like
they're,
not
the
only
ones
that
have
had
to
deal
with
this,
something
that's
more
work
is
producing
code
to
assist
people
I'd
be
open
to
any
of
those.
C
That's
an
excellent
question
and
there's
absolutely
an
overlap
there.
So
we
should
keep
in
mind
that
that,
like
there's
other
groups
and
individuals
outside
the
typical
audience
for
this
particular
like
sig
meeting
that
are
likely
interested
categorizing
things,
unlike
collecting
use,
cases,
is
probably
helpful
to
like
at
a
minimum
like
understand
if
other
folks
are
interested
in
the
same
variants
like
of
these
different
problems
like
there's,
there's,
probably
three
or
four
other
big
areas
that
are
adjacent
to
all
of
these.
C
So
maybe
that
makes
sense
as
like
a
way
to
proceed
to
kind
of
collect
and
categorize
them
have
have
illustrations
to
like
help
communicate
exactly
what
they
are
like.
I,
I
would
imagine
that,
like
the
situation
where
you
have
like
going
back
to
the
leader
election,
one
where
you
have
like
n
number
of
controllers
m
number
of
clusters
probably
has
very
different
properties
at
different
values
of
eminence.
C
B
E
That
their
trick
is
to
figure
out,
which
instance
is
responsible
for
the
issue
configuration
for
each
cluster,
and
I
think
in
our
case,
if
we
had
multiple
mcs
controllers
that
need
to
determine
like
who
wins
in
a
certain
case.
You
know
this
is
the
list
of
things
it
has
to
do,
which
is
a
little
bit
seems
a
little
bit
more
decentralized
from
this
very
brief
information
about
istio.
E
I
guess
I
guess
what
I'm
trying
to
seed
the
group
for
is
brainstorming
like
is
there
specific
properties
about
how
the
mcs
controller
should
work
that
are
cases
of
needing
to
do
leader
election
a
certain
way
that
we
might
want
to
start
writing
down
now.
E
C
So
I'd
be
very
curious.
I
think
stephen
is
on
yeah
stephen,
like
is
there
anything
that
comes
to
mind
from
work
on
submariner
that,
like
like
anything,
any
cash
hits
l1
cash
hits
that
come
to
mind
that
might
be
nuggets
that
you
think
would
be
useful
to
share
around
like
the
leader
election
thing
or
or
anything
else
in
the
neighborhood
of
what
we've
been
discussing.
D
Yeah
well
so,
when
and
submariner
itself,
we
don't
do
cross
cluster
leave
the.
As
far
as
I'm
aware,
we
do
in
cluster
leader
election
to
choose
among
multiple
gateways
when
multiple
gateways
are
available.
D
I
think
we
just
used
the
cd
algorithm
there,
but
yeah
that's
something.
We've
been
discussing
on
and
off
to
support,
multiple
brokers,
basically
for
failover
and
so
the
yeah,
the
the
the
consensus
there
is
probably
to
go
with
raft.
Ultimately,
because
that's
something
that
we're
fairly
familiar
with
having
implemented
it
in
open
daylight.
C
E
I
don't
know
if
this
is
covered
here.
I
guess,
but
another
thing
that
I
was
talking
to
somebody
else
about
was:
if
your
controller
is
multi-tenant
and
you
want
to
like
reduce
blast
radius,
if
that's
like
a
means
by
which
you
might
choose
leader
election
differently
for
different
tenants,
if
you
have
multiple
versions
of
the
controller
so
same
as
like,
fundamentally
the
same
as
different
regions
geographically,
but
now
it's
like
I'm
multi-tenant
and
I
want
to
optimize
different
tenants,
different
ways.
E
Well,
I
can
also
leave
this
slides
or
we
can
make
more
of
a
doc.
I
might
be
the
only
person
who
loves
slides
as
much
as
I
do,
but
feel
free
to
edit
or
or
ping
in
the
slack
or
added
to
the
notes
too,
as
we're
talking
about
it
because,
fundamentally,
I
think
at
this
stage
we're
definitely
looking
for
you
know
concrete
cases,
even
if
they,
even
if
it
ends
up
that
they're
very
similar.
You
know
like
that
they
have
the
same
problems.
E
I
think
that's
still
information
that
we
can
either
package
up
as
paul
says
with
you
know
a
couple
pictures
and
say
these
are
the
cases
that
are
concrete
for
mcs
controllers
that
have
been
happening
in
the
wild
and
if
we're
going
to
come
up
with
something
general
for
kubernetes
decentralized
controllers,
then
you
know
here's
this
info.
I
think,
even
if
they
all
end
up
being
kind
of
the
same
pattern,
that's
still
information.
That's
useful.
B
E
My
like
obituary
should
be
in
google
slides
form
with
google
draw
diagrams
because
I'm
such
a
slides
fan
so.
E
Cool
well,
is
there
anything
else?
These
other
questions
like
is
there
things
people
feel
specifically
about
this
pr
or
still
digesting,
or
is
that
jeremy
dude?
You
said
that's
kind
of
too
narrow
or
paul
too,
do
you
think
that's
too
narrow
of
a
purview
of
with
all
the
like.
You
know
what
we
were
just
talking
about
the
different
stages
of
things
we
could
suggest
commenting
on
the
single
pr
might
be
a
pretty
narrow.
B
I
think
we
lost
paul
first,
but
I
I
I
think
the
pr
is
us.
It
seems
like
it's
useful
in
general
for
other
things,
it
just
doesn't
solve
all
of
the
use
cases
that
we
are
that
I
think
we
need
to
focus
on
in
the
sig
just
some
of
them,
but
you
know
I'll
I'll,
probably
follow
up
with
a
comment
to
that
effect.
E
An
implementer
of
the
mcs
controller
would
use
this
leader
election
package
and
establish
some
key
and
key
comparison
potentially
to
solve
for
either
their
cross-region
situation
or
multi-tenant.
Or
this
gateways
thing
and
that's
where
the
scope
of
this
pr
intersects,
with
the
scope
of
our
general
issues
right.
B
Right
yeah,
I
think
I,
like
paul's
suggestion
of
kind
of
starting
with
categorizing,
without
that
I
think
it's
hard
to
make
statements
about
like
this
is.
This
is
how
you
should
configure
your
key
comparison
function.
E
B
So
maybe
start
by
gathering
gathering
some
use
cases
from
people
and
and
I'll
get
the
pr
authors
to
come
talk.
B
Maybe
at
our
next
meeting
in
two
weeks,
oh
yeah,
another
reminder
for
everyone:
we're
gonna
switch
to
bi-weekly.
We
talked
about
that
last
week,
so
there
will
not
be
a
meeting
next
week,
we'll
starting
this
week,
we'll
go
to
bi-weekly.
E
All
right
cool
definitely
see
if
we
can
talk
to
them
and
my
in
my
vision
I
brought
I
brought
this
one
this
page
up,
because
I
feel
like
this.
I
know
paul's
gone
so
he
can't
comment,
but
this
kind
of
overlaps
with
what
he
was
talking
about.
If
there
was
a
version
of
something
like
this
page
but
for
our
mcs
use
cases
that
would
have
a
lot
of
portability
in
this
conversation
everywhere
we
go,
we
might
go
so
that's
kind
of
what
I
envisioned
yeah.
B
Awesome
well,
thank
you
laura
for
both
topics
this
week.
I
think
that
is
our
agenda.
Unless
anyone
has
anything
else,
they
want
to
add.