►
From YouTube: Policies and Telemetry WG Meeting - 2020-06-03
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
Just
do
some
introductions:
justin
is
here
and
justin
justin.
We
it's
gonna,
help
us
out
this
summer,
so
I
just
wanted
to
introduce
them
to
everyone
working
on
policies
until
maturity
and
give
give
him
a
second,
maybe
to
say
hello,
I'll,
probably
be
showing
up
and
asking
you
lots
of
questions
as
he
works
on
some
of
this
someone's
suggestion
to
join
us
at
hi.
A
B
A
A
Epics
and
tracking
bugs
for
all
of
our
issues
on
the
roadmap.
So
now
we
should
be
able
to
track
all
the
work
entities
so
if
you've
seen
a
bunch
of
activity
recently.
For
me,
that's
what
that's
about
so
please
actually
working
through
things
refer
to
those
issues,
provide
updates
on
those
issues
and
I
can
use
that,
and
the
working
group
needs
meetings
and
other
meetings
to
help
us
track
progress
throughout
the
course
that
make
sense,
as
I
run.
A
A
A
B
A
B
You
explain
that
again,
what
did
you
say
like
Watson
mix-up
plug-in
and
why
is
it
special.
D
A
I
following
your
comments,
I
propose
that
a
new
Doc's
folder
in
the
mix
of
sub
package
and
then
cherry-pick
that
back
a
couple
branches
with
all
those
pages:
okay
and
just
remove
them
from
the
wiki.
Although
I'm
not
sure
I,
don't
have
permissions
to
remove
pages
from
other
people,
but
I
think
we
can
figure
out
how
to
do
that
and
that
would
just
clean
up
the
wiki.
There
wouldn't
be
a
whole
mixer
right.
A
B
D
Think
that
so,
okay,
so
I
think
that
the
short
answer
needed
is
that
yes,
I
think
we
do
neither
communications
plan.
It's
not!
We.
We
need
to
send
out
the
communication
that
this
is
happening,
maybe
a
month
before
it
actually
happens,
so
that
would
be
kind
of
around
the
time
when
one
seven
releases
or
slightly
before
we
should
say
okay
right
afterwards.
This
is
going
to
happen
so
just
job
just
an
FYI.
So
so
maybe
maybe
we
should
just
add
that
section.
Two
to
this
talk.
Okay,.
A
F
B
A
G
A
The
next
item
I
had
added
to
the
agenda,
was
someone
from
the
community
raised
the
issue
that
our
request,
total
doesn't
include,
requests
in
flight
and
that's
sort
of
by
design
at
the
moment,
but
I
was
wondering
what
we
think
about
supporting
that
use
case
and
how
we
think
we
might
do
that.
I
just
want
to
freeze
that
first.
A
B
D
B
B
B
A
B
A
A
B
D
D
B
B
E
D
D
H
C
I'm
just
wanted
to
talk
about
this
because
I
so
I
I'm,
looking
at
the
extension
that
has
been
proposed
specifically
for
was
extensions
and
I'm
not
entirely
happy
with
the
API
zone.
If
I
got
it
right-
and
there
was
also
it
was
discussed,
it
will
be
moving
forward.
Referencing,
instead
of
backward
referencing
did
I
get
that
right.
D
C
C
Doesn't
work
anyway,
so
I
was
just
looking
at
the
whole
lifecycle.
So
basically,
after
you've
developed
the
forces
for
custom
filters,
you
develop
a
filter,
you
publish
it
somewhere
or
college
nutrition
and
we
configure
it
on
the
workers
educational
to
use
it
with.
So
this
was
just
a
basic
idea
because
that's
something
that
webassembly
actually
covers
from
and
to
end
drive
survey.
There
is
important
place
where
you
can
actually
push
your
filter
that
can
also
offer
you
estimates
to
andhaka
images
to
repeat
visitor
centers.
So
the
cover
this
will
recycle
and
I.
Think
for
us.
C
If
the
coding
design
does
is
it
focuses
on
the
the
configuration
part
right?
If
actually
it
doesn't,
it
doesn't
look
at
how
filters
will
be
provided
to
the
cluster,
but
instead
it
just
allows
you
to
configure
any
kind
of
the
Envoy
filter
with
any
kind
of
configuration.
So
you
also
have
to
run
configure.
C
C
To
configure
these
filters
and
compete
against
you
workloads,
so
what
that
looks
like
is
it's
like
this?
Somehow
the
filter
diplomacy,
our
Guinea,
that
way
you
can
basically
specify
which
deployment
you
want
to
include
this
filter
with
some
basic.
Do
the
math
is
no
match
for
the
site,
cars
Mozilla,
a
specific
deployment,
and
then
you
can
pivot
and
a
convict.
That's
not
the
ball
good
boy
conflict,
that's
actually
the
configuration
strength
and
it's
passed
into
the
West
module.
C
D
D
What
it
does
not
do
right
and
and
and
that's
what
I
don't
I,
don't
know
how
much
so
what
this
does
not
do.
Is
it
doesn't?
Let
you
order
filters,
it
doesn't
take
care
of
any
dependencies
right.
So
so
this
tells
you
nothing
about
where,
where
it's
deployed,
what
what
exactly
it's
doing
so
I
think
I
I
agree
with
your
point
that
we
can
make
it.
D
We
can
kind
of
abstract
away
some
of
the
excess
boilerplate
stuff
that
someone
who
wants
to
configure
a
bit
assembly
filter
or
needs
to
do.
However,
at
least
the
way
I
look
at
it
is
that
it
it
will
be
good
to
have
a
uniform
VI
that
can
configure,
wasn't,
filter,
X,
top
filter,
lower
filter
and
anything
else
that
a
customer
wants
to
configure,
because
the.
D
So
we're
trying
to
make
on
water
filter
better
right,
like
so,
if
you,
if
you
go
through
the
dark,
yes,
so
today
that
is
on
what
filter
and
one
of
the
earlier
proposals
was
to
have
to
have
a
new
API,
but
it's
it
finally
compiles
down
to
a
water
filter.
But
there
are
many
reasons
why
we
don't
want
to
use
or
water
filter
directly
because
it
lets
you
do
too
many
things,
so
this
filter
API,
so
so
on.
D
C
D
D
D
But
but
but
and
then-
and
so,
but
what
happens
is
once
you
start,
adding
all
the
things
that
you
need,
then
you
have
like
two
distinct
things
like
one
is:
once
you
select
the
sidecar
or
once
you
select
the
Gateway,
how
do
you
order
things
and
where
do
you
put
things
that
that's
kind
of
one
concern
and
then
how
do
you
select
a
sidecar
or
a
gateway?
That's
other
concern,
so
the
first,
so
the
second
concern
is
selection
of
a
gateway
or
sidecar
can
be
done
through
forward
or
reverse
references.
D
And
now
we
are
kind
of
moving
towards
forward
references,
because
that's
how
that's?
How
other
things
are
people
moving
as
well
right,
so
what
we
can,
but
we
can
treat
that
orthogonal
be
so
those
mechanics,
I,
think
making
it
easier.
Still
for
webassembly
users
is
still
possible
and
and
and
we
can
begin
most-
you
can
most
certainly
there
any
PR
on
top
of
that
extension
in
which-
which
so
either
layer
an
API
or
offer
tooling.
C
My
question
is:
just:
is
our
goal
to
be
able
was
filter,
ecosystem
or
is
our
goal
riding
on
both
sides
of
the
tube
and
do
we
have
to
do
both
interceptor
and
as
I
think
we
don't
have
to
I
mean
if
there
wasn't
filter
can
be
more
specific
version
of
what
we
would.
What
comes
over.
This
is
amended.
I
have
totally
agree
that
we
should
be
forward
referencing
and
ask
the
option
project
you're,
the
only
ones
who
can
do
it.
We
can
change
our
sanities
tuition
reference,
a
filter,
but
the
landscape
with
that.
C
But
I
also
say
we
shouldn't
try
to
make
it
too
generic,
because
I
mean
that
was
supported
with
my
filter
and
while
that's
great
a
blessin,
a
boat,
for
example,
webassembly
up
to
two
built
on
top
of
the
steel
and
the
baby's
Weber
filter
to
Jake
Muzzin
filters
and
survived
me
today:
right.
That's
because
I've
also
to
a
surgeon.
Bergen
can
do
anything
even
if
this
doesn't
been
followed,
but
you
save
all
of
us,
but
do
we
really
have
to
do
both
at
the
same
time?
That's
the
only
thing
about
you.
B
Can
I
ask
a
good
question:
how
so
I
get
what
you're
saying
about
the
webassembly
CRD
it?
It
is
very
simple
in
terms
of
selection
and
it
is
focused
more
around
what
filter
you
want
to
instantiate
rather
than
everything
about
sto,
but
when
they
are
creating
envoy
filters
to
sequence,
the
filters
correctly,
how
do
they
know
where
to
start
it.
E
C
The
designer
Center,
simple
ideas
about
that
and
the
same
should
check
out
the
design
I'm
saying
we
should
make
it
as
easy:
a
state
if
I
were
to
install
the
single
cell
so
and
if
I
have
to
do
ordering.
You
should
remember
that,
but
I
don't
think
we
should
available
people
to
use
any
own
voice.
Also
do
this
because
that's-
and
it
is
the
exception
in
my
mind-
is
custom
ones
that
config
don't
necessarily
need
to
be
not
be
high
for
and
he
style
I
don't
I.
B
Don't
quite
agree
with
the
last
part
of
the
statement
totally
agree
with
the
first
part
of
the
statement,
which
is,
let's
make
the
simple
thing
very
easy
to
do
so.
Most
of
the
users
don't
have
to
deal
with
all
the
complicated
nuances.
The
current
on
where
filter
API
is
thoroughly
and
completely
broken,
however
majestic
it
might
be.
C
Think
well,
I,
just
think
what
makes
them
two
different
goals
and
maybe
I
I
think
we
don't
have
to
write
a
generic
API
that
can
do
everything
if
what
we
really
want
to
do
is
enable
people
to
use
wasn't
felt
as
easily
I
mean
I
can
I
can
live
with
having
a
new
generic
epi
I.
Think
it's
great,
but
it's
not
necessarily
required
for
people
who
want
to
use,
wasn't
filters.
So
what
I'm
saying
said,
I
already
put
this
in
the
design
directions.
I
can
I
can
absolutely
live
with
this
design.
C
If
we
look
forward
referencing,
that's
even
better,
and
but
what
I
would
like
on
top
maybe
will
then
be
a
CID
that
is
specifically
one
wasn't
filters
and
that
you
know
where
you
don't
have
to
specify
the
GDP
a
name
of
the
filter.
You
don't
have
to
specify
the
filter
conflict
map,
but
that
actually
is
time
safe
and
it's
four
thousand
filters
right
and
then,
if
you
start
doing
that,
you
can
later
extend
it
to
you
know
if
we
have
more
photos.
D
I
I
think
that's
that's,
like
that's
fair,
so
my
question
I
think
on
that
comment
was
how
many,
how
many
things
need
to
be
specialized
and
the
only
thing
that
I
could
think
of
that
needs
to
be
specialized?
Is
the
UDP
a
name
and
creating
a
whole
new,
a
whole
new
filter
for
UDP,
a
name
seem
so
actually
let
me,
let
me
back
up
I.
Think
I,
think
that
that
in
in
principle,
yes,
it
does
make
sense,
do
to
have
to
have
something
something
simpler
and
something
that
that's
that's
more
focused.
D
We
can
share
it
on
top
of
the
extension
API.
Is
that
because
we
need
the
extension
API
to
configure
even
our
own
telemetry
and
we
need,
if
we
so
the
way
customers
use
it
is
that
they
let
some
people
use
like
X
dot
if
they
let
some
people
use,
lower
and
webassembly
extensions
and
for
them
having
a
single
way
to
do
all
that
actually
does
make
sense.
So
that
is
also
motivation
in
having
just
one
unified
way
to
configure
all
these
things.
So
I
agree
we
can.
We
can
certainly
layer
it
and
Daniel
is.
C
No,
it's
not
about
complicated
its
own,
so
I
just
think
as
soon
as
we
leaked
the
Envoy
infect
them
we're
giving
up
on
this
chance
to
actually
have
these
things
as
first-class
citizens.
If
we,
if
we
want
to
push
for
wasn't
your
first
class
extension
API,
but
then
we
say
to
people,
look
you
just
have
to.
You
have
to
be
able
both
mister
HTTP,
resin,
filter
and
average.
If
it
wasn't
filtered,
then
you
just
invoke
the
Chasen
to
configure
it
and
then
why
you
know
why
don't
we
make
a
first
class
citizen
$0.10?
C
What
I
will
beg
wasn't
user?
If
we
were
trying
to
push
that
and
she
we
should
do
that
I
think
and
well.
I
think
they
they
could
share
code.
As
you
said,
it's
just
a
layer
on
top.
If
we
have
to
demand
solution,
we
can
make
that
wasn't,
filter
and
specialized
instance
of
that,
but
we
could
also
develop
that
imperative
and
share
the
coordinator.
E
C
C
C
B
I
think
layering
makes
sense
in
my
mind
at
least
and
like
we
all
agree,
and
if
we
have
to
layer
it,
it
does
mean
I
think
there
is
some
implicit
dependency.
Sadly,
or
we
go
the
route
that
Daniel
you
are
suggesting,
which
is
don't
think
of
them
as
dependent
things
and
start
development.
At
the
same
time,
I
will
leave
up
to
Duggan
Mandar,
but
I
still
prefer
layering.
In
this
case,
yeah
I
think.
A
You
like
we're
still
so
early
in
this
trying
to
get
the
extensions
api
sorted
and
then
figuring
out
where
it's
shortcomings
are
makes
some
sense
to
me.
I
don't
know
Daniel.
If,
if
that's
not
the
answer,
you're
hoping
for
you,
think
that's
the
wrong
approach,
but
I'm
worried
that
we
might
design
380
eyes
that
overlap
or
don't
don't
cover
the
right
bits
all
at
the
same
time
as
we
try
and
figure
this
stuff
out.
Well,.
C
I
think
the
design
of
these
should
go
hand-in-hand
and
you
come
to
know
designed.
There
wasn't
any
exceptions
if
you
guys
figured
out
that
the
only
place
to
change,
and
then
you
have
to
change
the
other.
If
you
I
think
that's
correct
and
also
you
know,
I
just
was
letting
just
a
second
time.
You
pretending
that's
working,
but
not
trying
to
tell
you
what
to
do
but
yeah
I'm,
just
looking
at
there
wasn't
filter
game
and
I
really
liked
to
help
also
implement
it.
And
that's
just
you
know
what
I
advise
my
thoughts.
C
D
B
D
D
B
B
D
D
B
D
There
is
a
there
is
a
specific
thing
which
I
would
like
to
drive.
Reubens
attention
on
is
that
the
need
for
having
clusters
that
are
matched
against
an
extension
right
is
is
one
of
the
reasons
why,
on
what
filter,
API
kind
of
co-evolved,
with
being
able
to
define
a
cluster
and
the
extension
that
is
not
being
handled
by
the
you
care
today,
so
that
that
remains
that
there
are
several
ways
to
think
about
it.
D
If
API
now
it
just
takes
what
you
want
to
call
as
a
direct
input
so
that
you
don't
need
to
define
a
cluster
and
then
refer
refer
to
it
by
name
and
that's
certainly
an
option
of
asking
the
extensions
do
not
use
cluster,
but
just
just
get
the
input
of
what
what
they
want
to
have,
and
then
that
said,
otherwise
we
will
have
to
have
some
other
way
in
sto
of
defining
a
cluster.
So
that's
a
very.
C
D
So
we
we
actually
have
code
where
we
define
jaribg
service
in
line
as
well
and
I
believe
that
in
stack
tower,
we
are
already
using
that.
So
it's
so
it's
definitely
an
option
like
cluster
or
no
store,
but
then
we
may
have
to
add
an
API
or
I
mean
maybe
service
entry
is
that
API,
but
that,
like
that
part,
remains
to
be
fully
fully
explored.
B
D
These
are
two
separate
streams
of
work,
but
the
cluster
is
important
and
if
we
need
to
support
it,
then
again
we
can
either
continue
using
onwards.
You
to
API
to
add,
add
a
cluster
or
have
just
a
better
API
to
add
a
cluster
right,
not
nothing!
That's
reasonable
right,
so
I'm,
so
I
think
it's
it's
fine
separating
it,
but
I
I
wanted
to
make
sure
everyone
understands
that.
That
is
something
that
we
need
to
solve.
Well,.
B
D
D
D
It
is
like
better
in
in
some
respects
and
someone
posted
a
message
on
on
the
channel
with
a
with
a
fairly
detailed
kind
of
POC,
of
what
it
would
take
and
what
are
the
advantages
and
it
seemed
seemed
great,
and
then
we
also
saw
that
there
was
a
the
book
that
Doug
is
just
fixing,
which
is
another
encoding
issue
with
the
Zipkin
client
implementation.
You
know
employer,
so
all
that
is
pointing
towards.
We
should
just
support
open
sensors
Zipkin
properly.
A
D
I
I
D
D
A
D
A
D
D
No,
it's
well
I
mean
at
least
in
what
I'm
describing
no,
that
was
nice
agent.
Well,
it's
just
purely
between
between
the
the
proxy
and
the
control
plane.
It's
like
I
have
this
yeah
I.
Have
this
IP
I,
don't
know
what
it
is.
I
need
to
reverse
map
it
and
who
all
can
the
worst
puppet?
Who
has
this
information
so
control
plane?
Does
all
an
extra
server
does
so
I
mean
those
are
the
two
broad
options
right
either
stand
up
a
new
metadata
server
or
use
the
control
plane,
I
guess.
B
B
Think
the
solutions
that
both
of
you
are
saying
us
essentially
the
same-
you
can
move
that
information
to
whichever
entities
the
architecture
and
once
you
move
it
to
pilot,
you
can
have
a
cache
in
the
agent
or
you
can
have
the
agent
do
all
the
work
and
have
the
cash
somewhere
else
if
needed.
Right
so
I
mean
I,
guess
that
that
we
can
discuss
implementation
if
this
is
actually
something
that
we,
what
it
is
I'm
still
not
convinced.
This
is
something
that
we
should
do
then
go
around
go
with
this
approach.
D
So
having
there
is
a
slight
advantage
in
having
like
a
look
aside,
one
right,
which
is
not
the
control
plane
as
well,
because
that
thing
is
actually
like
dead,
simple
right
it.
It
really
doesn't
do
much
so
making
any
synchronous
call
to
resolve
a
few
things
or
traverses.
All
few
things
is
not
like
not
that
hard.
Now
we
have
a
DNS
proxy.
D
We
didn't.
We
add
some
some
DNS
thing
to
pilot
agent
recently
a
vacay
yeah,
so
we
can
do
like
DNS,
reverse,
lookups
and
and
and
stuff
like
that,
and
then
use
the
extended
DNS
format.
That's
also
an
option,
but
again
it's
I
I,
don't
I,
don't
think
it's
just
a
different
type
of
transport,
I,
I!
Suppose
right
I
mean
yeah
I
want
you
can
just
make
an
HTTP
call
and
get
the
information.
So
there
is
no
good
value
in
using
DNS
to
get
the
same.
I
agree.