►
From YouTube: Policies and Telemetry WG 2018-08-29
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
Are
well
when
we
came
to
listen
but
also
have
a
number
of
questions,
we'd
love
to
ask
at
some
point,
probably
towards
the
end
of
this
meeting,
whenever
there's
a
some
time
but
excited
to
get
to
work
with
you
guys,
awesome.
B
B
B
D
B
D
D
B
Beautiful
yeah,
so
I
think
some
of
the
big
things
that
were
going
on
the
last
couple
weeks
are
just
following
up
on
some
of
the
performance
issues
of
the
mixer.
We've
been
looking
at
a
couple
of
different
dimensions,
including
sort
of
the
client
performance
inside
of
envoy
and
its
impact
on
envoy
scaling
and
some
of
the
issues
with
killing
the
cube.
The
API
server
with
too
much
load
that
start
up
so
I
know
that
there's
no
work
going
on
there.
B
I
don't
have
a
lot
of
final
results
to
report
and
when
DARS,
not
here,
you
can
tell
you
more
about
some
of
the
work
he's
doing
in
that
perspective,
so
yeah
I,
don't
have
anything,
is.
Is
there
anyone
that
wants
to
talk
about
anything
in
particular?
Should
we
just
go
straight
to
the
questions
and
sort
of
go
around?
A
Your
questions
sure
this
is
great
wow.
This
worked
perfectly
I
have
a
number
of
them,
so
if
someone
else
has
another
question
feel
free
to
interject
at
any
time.
I'm
happy
to
to
share
so
I
was
listening
to
a
previous
working
group
meeting
that
I
found
on
YouTube,
which
one
awesome
thank
you
for
uploading.
Those
it
actually
was
like
really
helpful.
I
think
I
was
the
one
of
six
viewers,
or
so
so.
A
I
was
proud,
where
there
is
a
discussion
about
the
reason
why
distributed
tracing
was
being
supported
both
directly
through
envoy
and
through
the
mixer,
and
there
was
a
discussion
about
how
some
people
want
to
be
able
use
sto
without
using
the
mixer
and
I
want
to
just
directly
get
traces,
the
Envoy
and
Wow.
At
the
same
time,
mixer
provides
additional
features
on
top
that
you
could
make
use
of.
The
first
question
I
had
was:
is
there
anything
in
place
that
prevents
double
reporting
in
the
event
that
you
miss
configured
to
accident?
B
B
No,
that
was
it
yeah,
so
there's
nothing
that
prevents
you
from
saying:
okay
envoys.
We
want
you
to
report
to
say
Zipkin
right
now
and
also
like
here's
the
configuration
for
a
trace
pan
adapter
for
mixer.
That's
gonna
report
to
Zipkin
as
well,
so
you
can
duplicate
the
results.
I
would
think
that
you
that's
I,
don't
know
how
you
detect
that
necessarily
we
could
probably
yeah
I.
We
don't
have
anything
in
place
that
could
detect
that
today.
B
Okay,
there's
like
a
third
piece
of
that,
it's
sort
of
hidden
right,
is
that
mixer
itself
generates
trace
data
about
things
that
flow
through
it
and
it's
not
pulling
that
configuration
from
the
same
way
that
the
envoys
are
getting
configured.
So
if
we
unify
that
we
might
have
a
way
of
detecting
in
process.
If
you
have
duplicate
reporting
going
on
right,
we
don't
have
anything
for
that
right
now.
Oh
okay,
thank.
B
A
It's
there's
like
built-in
functionality
for
that,
in
addition
to
configuration,
and
so
one
of
the
things
that
we
would
like
to
be
able
to
do
is
to
be
able
to
allow
people
that
are
using
SEO
to
also
use
light
step
as
tracing
backend,
and
so
one
of
the
ways
we
thought
would
be
easiest
was
if
we
exposed
the
light
step,
configuration
options
through
sto
and
essentially
we're
able
to
propagate
that
down
all
the
way
to
the
envoys
and
I
think
I
found
that
this
is
similar
to
what
we
do.
What
is
done
for
Zipkin
one?
B
E
D
E
The
right
mood
strap
config
for
for
envoy.
One
current
question
here
is:
there's
a
there's,
an
API
key
type
thing:
that's
you
and
whether
or
not
like
how
sensitive
we
want
to.
We
want
to
have
that.
I
talked
to
some
of
the
folks
that
lights
up
and
they
said
it's
not
a
very
sensitive
key
like
we
could
just
have
it
as
a
config
map,
with
the
rest
of
the
that
config,
so
I
think
that's
kind
of
the
open
question
there
and
then
just
to
have
that
have
that
done.
Yeah.
A
So
does
I
think
you
came
a
little
late.
I
am
one
of
the
likes
that
people,
not
the
one
that
had
been
talking
to
you,
I'm
a
new
comer
on
this
one
right,
yeah
so
actually
touched
on.
What
was
gonna
be.
My
follow-up
question
was
yes.
There
is
this
axe
again
file
that
generally
is
necessary,
I
needed
from
the
envoys,
and
what
I
was
thinking
was
to
just
provide
that
via
config
map,
but
I
wasn't
sure
if
that
was
something
that
would
have
the
support.
A
B
A
This
so
my
I
have
another
question
which
was
about
that
one.
What
we
were
hoping
to
do
was
to
provide
like
step
support
via
directly
through
the
on
voice,
because
envoy
already
has
the
ability
to
do
that
and
then
potentially
also
write
a
mixer
adapter,
which
I
now
realize
that
they're
in
process
and
out
of
process
adapters
and
I'm
not
totally
sure
what
the
direction
is
he
was
trying
to
head
with.
That
is
so
that
was
going
to
be
my
next
question.
Actually
so.
B
As
as
a
basically
100,
basically,
we
are
transitioning
all
new
development
to
out
of
process.
Okay,
so
I
was
asking
because
if
you
build
another
process,
adapter
right,
you
can
manage
that
token.
However,
you
wanted
to
manage
with
the
deployment
of
your
other
process
adapter,
and
we
wouldn't
have
to
have
an
sort
of
an
opinion
on
that.
So.
A
That's
that's
one
option
cool
yeah
that
would
actually
great
I
saw
someone
had
asked
a
similar
question.
One
of
the
Google,
Groups
and
I
think
got
that
answer
as
well.
Another
actually
now
we're
talking
about
out
of
process
adapters
actually
meet
pause.
If
somebody
else
has
a
separate
question,
please
you
can
interject
I
just
have
a
list
of
them,
so
it's
not
I'm
gonna
keep
going
down.
It
go
once
twice:
okay,
I'm
gonna
steamroll
through
okay,
one
question
I
had
for
the
out
of
process
adapters
is:
is
there
any
within
mixer
batching
supported
so
I'm?
A
B
A
A
We
have
these
components
called
satellites
which
run
in
our
customers,
V
PC,
and
essentially
they
configure
their
application
to
send
trace
fans
directly
to
this
component
and
then
that
component
of
smart
and
figures
out
which
should
actually
propagate
to
our
back-end.
So
this
enables
you
to
reduce
the
amount
of
stuff
that
you're
sending
over
network
reduces
a
lot
of
costs,
so
we
can
perform.
You
know
cool
stuff
with
it.
Basically,
one
thing
that
I
was
considering
as
I
was
briefly
reading
over
the
out
of
process.
A
Could
it
just
send
directly
I
think
if
we
did
that
we
would
probably
ignore
the
configuration
stuff
that
comes
along
with
it
just
because,
having
basically
that
components,
configuration
is
already
quite
complicated
and
I
think
it
would
be
a
separate
add-on
project,
maybe
to
have
that
be
configurable.
It's
a
misdeal,
but
does
that
sound
like
an
absolute
abomination?
Or
is
that,
like
a
reasonable
thing
to
do
potentially
no.
B
I
think
it's
a
shame
and
Dara's
in
here.
I
think
this
is
one
of
the
visions
for
out
of
process
right
is,
you
could
just
add
the
api's
to
your
existing
pieces
and
use
them
that
way?
Okay
and
then
all
like,
depending
on
where
we
go
with
the
next
generation
of
mixer
architecture.
Maybe
it's
embedded
in
Envoy
itself
right
the
call
to
just
go
straight
to
that
to
your
satellite
and
then
from
there
that
beat
in
a
bit
so
yeah
I
think
that's
not
an
abomination
say
that
cool.
A
C
B
B
In
and
batches
from,
the
the
proxy
and
batches
of
up
to
a
thousand
or
once
every
second
or
something
right,
and
so
that
sort
of
batching
is
passed
on
through.
So
we
I
think
there's
a
feature
request
there
to
maybe
do
something
more
intelligent
with
how
we
are
grouping
and
batching
and
just
in
dispatching
to
the
adapters
sure
yeah.
D
B
B
C
B
A
B
A
given
up
issue
is
a
great
way
to
start
and
raising
it
here
is
it
so
we
can
track
it
sort
of
week
over
week
here
and
see
whether
or
not
any
progress
is
being
made
or
what
our
thoughts
is
in
kids.
On
the
roadmap
I
mentioned,
there's
a
road
map
that
was
being
built,
I,
don't
know
what
the
status
fully
of
that
is,
but
that
might
be
something
we
want
to
like
add
on
to
the
road
map
and
relating
to
performance
and
scalability.
B
It's
just
how
to
add
a
process
at
that
there's
work
in
terms
of
tuning
the
performance
of
them,
so
that
would
be
another
way,
but
yeah
I'd
start
with
a
github
issue
and
maybe
follow
it
up
with
an
email
to
the
audience.
One
entry
mailing
list
and
just
say
hey:
this
is
something
we're
interested
in
and
that'll
get
double,
get
the
ball
rolling,
okay,
cool!
B
B
A
Have
another
kind
of
procedural
question
I
think
this
was
Jeff
had
mentioned.
This
I
saw
that
there's
this.
This
is
going
back
to
propagating
the
like
step
configuration
directly
to
the
envoys.
What
is
the
process?
I
saw
that
that
mesh,
config,
proto
I
think
is
one
thing
that
needs
to
get
passed
through,
that
lives
in
I,
think
sto,
API
and
like
what
is
the
process
for
juggling
that
dependency
for
like
making
an
update
there
and
then
updating
sto
proper,
like
what
it?
What
how
should
I
best
do
that.
B
B
D
F
D
Their
current
situation,
because
is
a
template
in
history
pot
that
basically
moves
data
from
that
Mexican
taking
to
the
N
boy,
tembu
strapped
food.
So
that's
that's
just
the
flow
and
the
injector
will
need
to
propagate
it
as
well.
We
got
it
reads
the
a
config
map
and
then
it
translates
it
into
flags
to.
D
A
Anything
else
on
the
list
of
sorry
I'm
glad
I
jumped
around
a
little
bit
so
now,
I
need
to
go
back
through
sorry.
I
feel
like
I,
completely
hijacked
this,
but
no.
B
No,
it's
fine,
yeah,
I.
Think
I
think
this
is
good
because
we
didn't
have
a
whole
lot
on
the
agenda.
I
did
want
to
mention
that
I
think
in
the
next
meeting.
There's
probably
a
bunch
of
Doc's
that
are
gonna
come
out
related
to
different
designs
and
things
that
we've
been
asked.
So
I
have
a
doc
on
how
to
extend
the
attribute
set
for
mixer
in
different
ways
that
should
be
coming
out.
B
They
think
there's
the
roadmap
doc,
that's
gonna,
be
out
as
well
and
I
expect
that
there's
going
to
be
some
other
stuff
on
monitoring
applications
themselves
in
terms
of
not
just
the
proxy
that
will
be
coming
out.
So
there
are
a
bunch
of
stuff
that
we
should
see
traffic
on
the
mailing
lists
this
week
and
next.
So
there
should
be
tons
of
good
discussion
at
our
next
meeting,
so
just
wanted
to
sort
of
make
that
announcement,
but
right
now
there's
nothing
I,
think
it's
all
pending.
B
That's
one
of
the
docs
that
hopefully
will
come
out
there's
meetings
yesterday,
even
about
this
issue
between
the
two
teams.
So
we
want
to
make
sure
you
know
they
were
very
much
interested
in
building
a
mixer
like
thing
for
telemetry
and
so
we're
trying
to
see
if
we
can
eliminate
having
two
teams
that
building
the
same
thing.
B
So
we're
trying
to
align
I've
been
trying
to
convert
some
of
the
issue
of
component
monitoring
over
to
open
senses
to
enable
flexibility
in
terms
of
maybe
export
through
this
telemetry
service
that
will
probably
share
between
the
two
teams,
so
yeah
I
hope
we
hope
to
have
more
and
that's
something
that
should
go
out
soon
on
the
mailing
list
about
that.
But
we're
trying
to
align
and
share
if
not
share
components,
share
libraries
so
that
we're
not
reinventing
the
wheel
for
to
limit
read
consumption
and
generation
across
different
open
efforts,
cool
I.
A
B
Yeah
and
we'll
probably
start
actually
moving
some
of
the
old
adapters
out
and
data
process
adapters,
as
we
have
time
so
dilemma
ad-hoc
basis,
but
yeah.
The
idea
is
to
move
basically
almost
everything
out,
except
for
maybe
things
like
the
standard,
I/o
logger
or
things
that
are
just
basic,
the
bug
stuff
but
yeah.
That's
that's
the
future.
That's
so
we're
headed!
Oh
hey,.
E
A
B
G
I
guess:
verification,
because
right
now
the
the
main
problem
I
get
with
people.
You
know
my
customers,
I
guess
is
I
keep
running
into
issues
where
the
install
didn't
fully
complete,
sto
right,
and
so
it's
it's
and
it's
it's
difficult
to
verify,
and
so
I
I
think.
The
resolution,
if
there
is
one,
is
that
it'd
be
really
cool.
If
there
was
a
tool
or
you
know
a
way
to
verify
that
sto
was
actually
installed
properly.
G
Maybe
there
was
an
error,
but
they
just
like
whoops
right
on
by
it,
and
so
it's
that's
that's
kind
of
that's
kind
of
the
issue
that
I've
run
into
is
just
you
know,
in
the
absence
of
that,
it
would
be
nice
of
a
tool
that
would
do
it.
It'd
be
nice
if
there
was
like
maybe
a
set
of
steps
or
something
to
walk
through
people
to
like
this
is
the
best
way
to
verify
that
you
actually
have
a
working
installation,
yeah.
B
I
think
that
would
make
a
great
contribution
to
the
ops
guy
stuff
that
exists
today.
I
think
we
have
a
lot
of
the
data
we
just
don't
have
any
unified
way
of
you
know.
So
we
mix
it
reports
on
the
attributes.
If
you
can
see
the
number
of
total
is
the
total
number
of
handlers,
configured
etc
alike.
So
there's
sort
of
abstract
information
there,
but
we
don't
have
a
unified
saying
all
right.
This
is
what
a
valid
snapshot
should
look
like
and
make.
G
F
B
G
By
the
way,
I
just
throw
this
out
to
I
have
been
working
with
Mandar
I'm,
getting
experiment
after
working
at
the
GR
pcs
and
with
the
latest
fixes
it
does
work.
I
mean
with
all
of
like
all
the
template
types,
including
custom
types,
so
I
actually
I
do
have
it
working
there's
a
little
pink
penis
and
the
configs
that
I'm
getting
oddly
getting
my
completely
two
separate
raw
config
bytecode
sets
that
resolve
to
the
same
thing,
which
is
really
bizarre
and
and
it
and
it
breaks,
then,
with
the
the
code
that
the
example
code.
G
You
know,
because
it's
just
doing
Rob
white
checks
and
saying
that
it's
the
same,
you
know
I
I,
don't
need
to
create
a
new
one.
I
got
around
it
by
Ed
and
I
go
ahead
and
unmarshal
it
and
check,
which
is
fine.
It's
not
that
big
a
deal
in
fact
I
can
I
can
do
better
checks
that
way
anyway.
But
but
you
know
something
very
bizarre,
maybe
on
the
check
side
and
one
is
on
the
report
side,
possibly
or
yeah,
I,
don't
know,
but
it's
like.
C
G
B
Interesting
yeah
that'd
be
interesting
to
track
down
I'm
happy
to
hear.
Do
you
got
things
working?
Would
you
be
interested
at
all
and
sort
of
typing
up
some
sort
of
experience
guide
or
like
this
is
how
I
got
it
working?
These
are
the
steps
I
had
to
do
kind
of
thing,
yeah.
G
Yeah
I
mean
that
I.
Can
you
throw
one
thing
out
there,
there's
a
ton
of
boilerplate
stuff
that
but
I
have
to
do,
and
so
I
I
have
like
this
this
whole
class,
because
what
I
did
is
I
took
what
I
already
had
did
that
go
ahead
and
wrapped
it?
You
know
for
a
RPC
and
there's
just
like
a
lot
of
I
need
to
convert
these
types
to
these
types,
and
you
know
all
this
stuff.
That
should
just
be
absolutely
should
be.