►
From YouTube: Policies and Telemetry WG - 2019-12-04
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).
C
We
have
spent
today
what
was
that
meant
to
mean?
Oh,
he
was
going
to
come.
B
A
D
So
we
gave
a
talk
about
Bluetooth,
telemetry
and
I
mean
all
the
questions
that
we
got
were
mostly
related
to
like
wasum
and
how
that
how
that
fits
into
the
whole
system
we
did.
We
did
cover,
was
on
on
a
few
slides,
but
like
that,
wasn't
the
the
title
of
the
talk,
though
we
did.
D
We
definitely
conveyed
the
fact
that
the
the
extended
team
is
is
delivering
that
so,
like
some
of
the
other
interesting
stuff
from
what
telemetry
standpoint
was
psyllium
released,
Hubble,
which
was
which
is
a
l-3
telemetry
using
EBP
F,
and
then
they
go
back
up
to
envoy
for
error
seven
and
would
be
good
to
investigate
it's
kind
of
one
level.
Deeper
Roman
has
already
looked
at
it
like
some,
but
I
think
we
should
be
able
to
do
something
more
there
just
to
see
if
we
can
get
some
l3
telemetry
for
free.
D
So
at
least
the
people
I
talked
with
that
they
were
all
pretty
excited
about
the
fact
that
it
is
gonna
be
extensible
and
they're
going
to
be
able
to
like
add
this
extra
behavior
and
for
telemetry,
specifically
yeah.
They
were
very
excited
that
they
can
get
to
the
material
for
less
far
fewer
resources
resources.
Then
they
they
do
today.
I
think
both
those
positives,
okay,.
D
G
D
One
one
feedback
that
I
got,
which
should
feed
into
how
we
decide
to
do.
The
mapping
was
the
mixer
mixer
language
was
too
difficult
to
pick
up,
which
and
and
and
this
was
kind
of
a
meta
feedback
right.
This
was
by
someone
who
works
with
like
they
need
to
explain
to
their
customers,
how
to
use
mixer
and.
D
D
And
then
there
was
also
like
related
comment
about
metrics
consume
ability,
so
SMI
metrics
give
people
a
standard
way
to
consume,
telemetry
not
just
reduce
it
so
that
they
can
get
tooling
and
whatnot
on.
On
top
of
that,
and
our
answer
is
being
used
to
meteors
yeah
is
that
standard,
but
regardless
that
did
come
up
several
times.
D
H
A
F
If
people
would
remember
the
early
days
of
sto,
we
used
to
have
like
source
service
way
back
when
well.
The
problems
with
that
is,
we
didn't
derive
that
very
usefully.
It
wasn't
actually
consistent
between
the
destination
and
the
source
and
had
much
problems.
We
end
up
taking
it
out,
so
this
is
kind
of
bringing
that
back,
but
in
a
way
that's
a
little
bit
more
standardized
and
consistent
on
both
client
and
server
side.
F
So
both
client
and
server
will
have
a
canonical
service
and
it
will
be
consistent
and
one
of
them
so
actually
there's
a
couple
of
well
background.
Let's
get
the
background.
Ok,
let's
skip
this,
so
the
goals
are
probably
they
just
part
here.
So
one
goal
is
that
there's
exclusivity
so
I'm
skipping
the
first
one
so
that
each
so
each
and
every
workload
will
have
exactly
one
and
only
one
canonical
service
and
everyone
will
have
some
canonical
service,
so
it
will
basically
always
exist.
It'll
always
be
exactly
one,
and
that
gives
us
exclusivity.
F
F
So
this
proposal
has
a
bunch
of
heuristics
to
try
to
generate
a
name
in
case
the
user
doesn't
provide
one
but
allows
overriding
it
if
there
is
and
the
existence
that
they
want
to
use,
and
then
that
then,
once
this
name
exists,
we
take
sport,
expose
it
as
askew
attributes
in
the
extra
battery
vocabulary
on
both
source
destination
and
then
use
those
in
some
job,
and
this
proposal
explicitly
does
not
include
using
this
as
a
policy
attachment
point
sort
of
explicitly
or
for
enforcement.
That's
in
the
future
work
potential.
F
It
also
explicitly
does
not
support
these
canonical
services
spanning
across
multiple
in
spaces,
so
they'd
only
be
within
the
namespace,
so
yeah.
So
those
are
the
goals.
The
proposal
is
basically:
we
introduce
new
standard
label.
The
straw
man
was
arrested.
This
denial,
canonical
name
I,
don't
I'm
not
tied
to
that
at
all.
I,
don't
really
care
what
the
service
name
is
or
the
thing.
F
So,
if
you
guys
have
a
better
name,
great
and
then
I
think
the
important
part
is,
we
would
add
code
to
the
injector
that
would
automatically
fill
that
label
out,
so
that
label
would
basically
never
not
be
full-time
and
farther
down
in
the
dock.
There's
a
list
of
heuristics.
So
first
we
would
look
for
that.
So
if
someone
already
filled
out
the
label
exit,
don't
change
it.
F
If
not,
we
would
let
the
user
specify
alternative
label
names
and
we
would
have
some
defaults,
which
are
the
Canaries
Apple,
a
Apple
able
or
the
application,
C
or
D
label
that
they
came
up
with
after
companies,
dot
IO
name
and
that
I
think
may
be
more
controversial
would
be
because
this
one
I'm
not
sure
about
number
five
there.
If
there
is
exactly
one
service
pointing
at
something
we
could
use
that,
but
actually
I
don't
know.
If
there's
problems
with
that,
I
think
I'm
worried
about
there
is
if
that
changes,
so
I
could
flip
around.
F
F
F
D
So
so
five
only
five
seems
like
not
not
correct
yeah,
but
for
my
register
I,
don't
know
what
would
happen
if
we
did,
that
I
mean
I.
Think
six
six
is
six
is
always
gonna
be
available
and
it
seems
that
that
default
is
I,
cannot
think
of
a
situation
where
five
would
be
better
better.
Well,
exactly
yeah
I
mean
if
you
have
a
service
name,
that's
named
more
meaningfully
than
the
name
of
the
workload
yeah.
A
D
F
F
B
F
Yeah
so
that
I
think
that
main
so
actually
originally
there
was
a
whole
bunch
of
discussion
about
different
possible
names
for
this
concept,
but
then
Doug
mentioned
canonical
service
and
that
seemed
great,
so
they're
still
listed
in
different
names.
If
people
have
other
ideas
but
I
actually
like
canonical
service,
so
I'm
happy
with
that,
but
yeah,
that's
that's!
Basically
it
pretty
simple.
E
I
F
So
I
think
it
it's
not
quite
a
replacement
for
Apple
emotes,
but
it
sort
of
is
in
that
it's
meant
to
be
consistent
across
an
entire
mesh
and
across
like
so
yeah
we
could
say
actually
the
a
palatal
is
the
thing
so
that
that's
an
alternative,
as
you
say,
we're
not
introducing
a
new
label,
we're
gonna
use
that
label,
but
we're
an
auto-fill
it
if
it's
not
specified
so
I
think
that
would
be
an
alternate
proposal.
Yeah.
H
H
I
For
example,
you
get
the
destination
app
label.
Is
they
in
request?
Total
right
I
mean?
Would
there
be
any
reason
to
even
keep
that
in
telemetry
other
than
for
back
compatibility
or
something,
or
we
want
to
then
migrate
to
this
idea
of
the
canonical
service
which,
which
might
be
the
app
but
could
also
be
other
things?
Whatever
strategy
you're
trying
to
use
I
think.
A
D
Mean
they
are
they're
serving
the
same
function,
basically
right
like
if,
if
a
play,
ball
was
already
used
consistently
a
program
and
it
was
already
standard,
then
we
could
use
that.
But
this
is.
This
is
formalizing
that
concept
and
while
for
most
logic,
so
if
I
think,
if
the
applicable
name,
the
name
itself
was
already
acceptable
to
everyone,
we
probably
would
have
just
used
that
and
then
we
would
have
just
adopted
what
we
are
using.
But
it
is
simply
not
very
acceptable
to
the
rest
of
the
community
pause
now.
D
F
J
F
I
I
saw
that
it
was
but
I
mean
it's.
It
definitely
seems
like
a
good
thing.
You
know
so,
for
example,
if
people
are
using
app
labeling
or
if
they
want
to
use
the
kubernetes,
you
know
formal
dot,
the
name
label
or
what
or
whatever
approach
they
want
to
use.
They
end
up
with
at
least
something
given
this
list
of
heuristics
you
know,
so
that's
good
and
Doug
is
totally
right.
I
I
F
That's
that's
reasonable
to
call
out
in
here
because
there
is
a
discussion
in
the
like
future
work
section
of
that
this
would
be
useful.
So
once
this
is
a
standard
itis,
then
it's
actually
useful
as
a
best
practice
to
specify
it
and
then
use
it
in
as
a
label
closure,
and
we
could
even
get
to
the
point
where
it's
not
even
a
label
selector.
It's
just
so
like
on
a
let's
say,
authorization
policy.
F
Rather
than
saying
you
know:
here's
the
label,
selector
of
workloads
that
this
attaches
to
he
just
say
here's
the
comment
service
that
attaches
to
like
we
can
do
it
at.
We
can
actually
pull
that
up
as
a
field
and
say
smugger
canonical
service,
but
that's
yeah.
We
want
to
leave
that
out
of
the
proposal,
but
I
think
to
do
that.
F
Yes,
we'd
have
to
yeah,
you
don't
want
to
have
it
be
changing,
because
then
your
policies
would
change
yeah,
which
is
another
reason
actually
that
the
service
having
it
based
on
like
but
checking
for
well
one,
and
only
one
service
is
probably
bad
teachers,
because
you
don't
want
it
to
change.
It
should
only
change
if
they're
actually
changing
they're
like
if
the
user
is
changing
their
pod,
so
it's
based
on
Apple
able,
if
they
change
the
Apple,
a
bolt
right.
It
changes.
That
seems
fine
for
policy.
F
F
G
F
F
F
D
F
I
Guess
the
only
other
thing
I
would
say
is
that
it
would
seem
like
maybe
a
natural
extension
to
also
get
into
the
version.
I
think
it
was
a
sidebar
comment
in
the
doc.
You
know
it
seems
like
an
advantage
here
is
that
this
would
kind
of
cover
the
ISTE
Oh
app
and
version
or
well
the
sto
app
approach,
but
then
also
the
kubernetes
name
label.
It
kind
of
now
funnels
them
all
into
one
thing,
whereas
you've
got
the
same
kind
of
situation
with
version,
because
the
kubernetes
labels
not
version
its
whatever.
That
other
name
is.
I
So
it's
kind
of
a
similar
little
issue.
There
means
I
have
to
putt
version
yeah.
Well,
you
guys
have
happen
version
labels
and
it's
do
whereas
kubernetes
uses
different
standardized,
we'll
ables
for
those
two
things
Angus,
it's
there
with
a
canonical
version.
Potentially
that
funnelled
different
approaches
for
version
naming
into
one
place,
maybe
would
be
useful
at
I'm,
not
sure
yeah.
E
Hey
I
actually
agree
and
funnily
enough,
we're
looking
at
a
highlighted
paragraph
that
I
added
in
response
to
that
thread.
Calling
that
out
is
an
opportunity
for
future
work,
I
think
I,
think
getting
into
versioning
and
revisions
or
whatever
we
want
to
is
a
very
necessary
thing
to
do.
A
lot
of
the
user
journeys
that
that
people
using
is
to
you
are
gonna
want
to
deal
with
so
I
I.
D
Okay,
it's
so
these
are
the
things
that
people
do
today
with
this
job
right,
app
and
person.
People
are
are
used
today.
So
just
by
that
logic,
it
would
seem
like.
If
we
have
the
desire
to
address
both,
then
this
is
the
this
is
the
place
to
do
it.
It
is,
it
doesn't
seem
like
it's
whatever
too
much
to
bite
off
or
something
like
that.
It
you
see,
we
should
just
add
it
to
this
yeah,
but
it
it
seems.
We
should
add
it
to
this
proposal
because
it
is
like
we
actually,
even
in
our
documentation.
I
F
Yeah
I
think
we
would
want
to
fill
it
out
with
something
just
always
because
there's
a
lot
of
benefits.
That's
actually
part
of
why
this
pros
of
this
cuz
there's
a
lot
of
benefits
if
something
is
always
present,
but
then
it's
reliable
for
all
the
while
I
can
rely
on
it.
Yes,
I
know
this
will
always
be
present
and
always
I
didn't
have
a
bunch
of
special
cases
surprised
up
there.
Hey
don't.
D
From
a
from
a
labeling
standpoint,
there
is
already
like
I
think
we
are
going
to
work
on
it
in
1/5
anyway,
but
we
already
have
a
lot
of
duplication
of
labels
and
information.
That's
encoded
twice
in
like
different
forms
choose.
So
if
we
decide
to
add
a
new
key,
a
new
label
with
canonical
sellers
and
canonical
service
person
one,
and
we
also
have
app
and
person
so.
D
A
C
A
F
A
A
D
A
D
B
D
A
M
L
D
I
G
B
G
Original
the
whole
stripe
dot
I,
don't
know
either
way
depends.
Yes,
this
is
sort
of
restoring
the
function
of
origin
right.
Oh
the
the
origin,
dot
attributes,
it's
not
origin.
It's
is
more
what
was
asked
for
it
if
you
sent
to
service
a
and
then
networking
API
changes
to
service
B,
whose
service
a
intern.
I
A
I
A
D
G
C
A
Sure
you
had
mentioned
earlier
wanting
to,
and
we've
talked
about
it
I
feel
like
the
last
three
releases,
but
never
done
it.
Do
we
want
to
eat
codify
the
metric
schema
as
an
API?
Do
we
want
to
make
changes
to
it
to
address
the
way
it
is
you
wanna
have
a
rethink
of
it?
That's
not
something
we
should
do.
D
A
They
don't
see
huge
feature,
I'm
Melissa,
sort
of
trying
to
look
through
the
github
backlog
right
now.
What
do
we
want
to
try
and
accomplish
in
one
five
for
legacy
mixer
and
out
of
process
adapter
support,
although
all
the
docs
I
sort
of
fallen
in
a
bit
rot.
D
B
A
B
B
D
B
A
B
A
G
A
Know
someone
had
requested,
we
have
both
client
and
server
metrics
for
the
gateways
and
I.
Don't
know
if
that's
feasible
for
one-five,
but
something
we
could
think
about
it's
probably
easier
to
do.
It's
definitely
easier
in
v2.
G
D
H
A
M
Don't
know
we
have
another
discussion
since
you
know
the
original
position
was
trying
to
keep
the
installation
in
this
case
as
simple
as
possible,
and
if
you
wanted
to
do
an
extern
or
more
complicated
external
installation,
then
you'd
use
the
operator.
I
mean
I
have
seen
that
there
is
up
in
this
integration
with
a
Prometheus
operator.