►
From YouTube: Policies and Telemetry WG 2018-08-01
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
And
there
are
couple
items
on
the
agenda
is
the
calorie
team?
Here?
Yes,
did
you
guys
want
to
take
give
you
a
demo?
Should
we
start
with
that?
We.
B
B
E
F
B
B
B
I
actually
have
a
couple,
and
this
will
be
important
when
I
show
you.
The
grim.
I
have
a
couple
virtual
services,
for
example,
on
the
detail
service
I,
have
a
fault
injection
in
front
of
the
detail
service
that
you
can
see
here
and
I
have
another
fault
injection
here,
where
it's
going
to
or
50%
at
a
time,
but
I'm
also
routing
between
the
three
reviews:
services
evenly
33,
33
and
34
percent.
We
another
interesting
thing
about
the
UI
here.
B
Just
real
quick
I
did
it
I
change
and
updated
this,
and
what
I'm
trying
to
show
here
is
we're
doing
some
validation
on
the
yamo
that
gets
created
so
I
created
this
outside
of
the
SPO
CTL
I
just
did
I'm
on
open
shift,
so
I
just
used
the
OC
tool
and
I
created
it.
So
it
allowed
me
to
do
it
now.
I
know
the
gallery.
Stuff
is
going
to
start
doing
some
of
this
validation.
B
B
How
many
do
you
know
to
control,
get
to
look
at
the
animal
in
the
service
link
here
now
we're
actually
going
to
add
another
one
over
here,
we're
not
quite
sure
how
we're
going
to
do
it,
we're
thinking
of
adding
a
workload
item
on
the
left
here,
but
right
now
we've
got
services,
and
here
you
can
see.
Obviously
the
book
info
service
like
clearly
as
you
can
see
all
my
other
ones.
B
B
This
is
why
I
say
we're
probably
gonna,
add
a
workload
item
over
here
because
recall
that
key
ally
was
running
previously
on
Oh
a
doe,
and
we
didn't
really
have
a
notion
of
workload
back
then,
but
now
that
the
telemetry
at
least
is
exposing
workload
more.
The
first
class
concept
than
service
we're
going
to
start
doing
more
of
the
same,
we're
going
to
start
showing
workload
as
a
more
of
a
first
class
entity
here.
B
D
D
B
Ok,
so
here
you
can
see,
we
got
some
error
rates.
This
is
because
of
my
fault
injection
that
I
showed
you
that
I've
already
added
and
they'll
make
more
it'll
be
more
clear
when
I
look
at
the
graph,
but
again
I
just
wanted
to
show
you
this
stuff,
so
I
click
on
the
product
page,
and
here
you
can
actually
look
at
what
the
pods
are.
The
deployment
virtual
services.
Now,
if
I
go
to
virtual
services
here
and
I,
look
here,
I
get
a
more
of
a
graphical
representation
of
what
the
yamo
is
right.
B
Go
back
we're
still
contemplating
what
we're
going
to
change
here,
there's
going
to
be
workloads
over
here
or
getting
rid
of
source
services.
Probably
again,
this
is
we've
been
talking
to
our
user
interface
folks
about
this,
and
you
know
this
is
a
good
plug
here.
If
anybody
in
the
community
wants
to
contribute
to
what
they
think,
the
key
lui
should
look
like
should
how
it
work
go
over
to
the
key
ally,
github
or
our
IRC
or
mailing
lists,
or
Google
Groups
and
start
chiming
in
what
you
want
to
see
all
right.
B
B
B
Four
deployments
actually
I
should
say
four
apps
and
there
are
two
workloads
per
app
version,
one
and
version
two
we're
in
one
version.
Two,
then
you
can
see
as
I
high
as
I
hover
over
or
if
I
even
select,
I,
pretty
I,
focus
and
I
highlight
just
the
edges
and
the
nodes
that
are
related
to
this.
But
this
is
kind
of
noisy
right.
This
is
I'm,
seeing
every
workload
in
every
app
of
this
namespace.
B
If
I
were
to
change
this
and
go
back
to
work
load,
your
goodness
we're
going
to
go
back
to
what
we
saw
here,
but
the
labels
will
be
different.
So
we
have
that
app
graph
version
and
then
app
graph
version
less.
Then
we
have
a
workload
graph
and
that's
going
to
show.
You
know
this
is
no
boxes
because
there's
no
app
version
here.
It's
just
workloads
and
you
can
see
the
note
the
node
labels
are.
B
Actually
these
are
my
deployment
names
I'm
in
vertebra
Nettie's
right
into
the
appointment,
means
they're
not
has
nothing
to
do
with
the
app
labels
here.
So
we're
still
trying
to
figure
out
how
best
to
provide
these
different
views.
But
the
fact
is
that
we
have
these
views
now,
we're
just
trying
to
now
figure
out
a
good
way
to
so.
B
B
A
A
Ya
know
the
reason
I
was
asking
is
because,
for
some
of
the
other
integrations
there
was
an
introduction
of
an
edge
to
which
would
cement
the
Captur
edges
in
your
work.
Clothes
are
in
your
mesh,
but
that
obviously
doesn't
go
into
a
Prometheus
back
in
and
so
just
wondering
if
there's
a
place,
maybe
in
the
future
that
could
be
consumed
by
Kelly
as
well.
A
B
All
right,
so
here's
the
good
here's,
the
part
that
we're
not
entirely
happy
with.
So
if
anybody
wants
to
chime
in
or
again
even
if
you
want
to
just
do
it
later
off
off
the
call
here,
you
can
obviously
see
our
product
page
diversion
apps
version,
1,
2,
3,
the
version
2
&,
3,
good
or
ratings.
We
all
know
this
from
the
book
info
demo.
Right
now,
I
have
like
I,
said
fault,
injection
controls
in
place
in
front
of
details.
B
Half
of
my
requests
from
product
page
of
details
are
going
to
fail
same
thing
with
reviews.
That's
why
you
know,
what's
up
with
these
nodes
here
in
my
book
info
demo,
that
I
deployed
I
actually
changed
slightly
the
ISTE
Oh
demo,
and
what
I
changed
was
divert
with
the
value
of
the
app
labels.
If
you
remember
this
service
name
and
the
app
labels
are
all
the
same
string
in
book
info.
So
the
details,
app
label
is
called
details
and
the
details
service
is
called
details,
but
technically
they
don't
have
to
be
the
same
right.
B
Your
app
label
could
be
whatever
you
want
different
from
your
service
name,
so
I
need
my
app
label
different
than
the
service
name
to
see
what
what
does
the
telemetry
look
like
in
a
graph
format?
If
so
many
service
names
does
not,
is
not
the
same
as
the
app
value,
and
this
is
what
you
see
so
in
the
telemetry,
when
a
client-side
fault
occurs,
and
in
this
example,
product
page
is
trying
to
go
to
details
right.
B
B
Now,
that's
different,
then
the
details
act
because
notice.
This
is
this
request
successfully
made
it
so
from
product
page
to
details
that
successfully
made
it
I
got
a
200
response.
The
telemetry
tells
me
the
service
name
is
details,
but
the
destination
app
is
details
app
and
it's
the
same
thing
with
reviews
right.
These
are
all
successful
product
page
went
there
won't
be
one
b2
or
b3
or
reviews,
which
means
the
telemetry
is
telling
me
these
requests
made
it
to
review
Zack,
so
I
can
grant
it
that
way.
B
However,
when
a
client-side
error
happens
and
product
page
tries
to
send
it
to
reviews
but
sales,
the
telemetry
knows
nothing
about
the
app.
So
it
knows
nothing
about
review
that,
but
it
only
can
tell
me
product
page
tried
to
go
to
reviews,
so
the
destination
service
name
is
reviews,
but
we
have
no
clue
that,
if
any,
where
related
to
the
reviews
app
workloads.
E
F
F
E
A
B
G
D
F
One
other
one,
other
alternative
is
that
we
can
actually
show
we
can
decompose
the
wall
node
into
two
right
and
then
show
that
yes,
the
first
thing
is
to
route
to
the
service,
which
is
virtually
routed
to
the
actual
workloads.
And
then
those
two
are
explicit
edges
and
then
you
know:
okay.
Well,
you
just
got
to
the
first
part,
so
you
know
very
wanted
to
go,
but
you
couldn't
reach
there
and
then,
if
you
could
reach
there,
then
there
will
be
another
arrow
from
the
service
to
the
work
or
yeah.
B
F
F
B
A
C
D
C
B
E
B
Yeah,
it's
analogous
to
this,
so
what
is
unknown
here
is,
if
you
think
of
the
way
the
telemetry
is
right.
I
I
actually
went
through
the
product
page
route
itself,
so
my
browser
directly
hit
product
page
I
didn't
go
through
ingress.
That's
what
this
edge
is
here.
This
edge
here
is
I
did
go
through
the
ingress
which
then
routed
the
product
engine.
Presumably
it's
going
to
be
the
same
thing
here.
We'll
see
you
know,
edges
going
out
either
to
unknown
I.
F
It's
somewhat
backwards
right
because
for
an
external
service,
what's
important
is
that
you're
actually
calling
that
external
service
and
then
you're,
calling
it
either
directly
or
via
egress
right,
so
it
almost
seems
like
egress
should
be
in
the
middle
of
the
color
and
the
service,
not
the
other
way
round.
It's
so
like
egress,
either
should
be
in
the
middle
in
some
other
node
form
or
not,
and
that
would
just
dictate
or
or
it
could
be,
an
edge
type.
B
Right
so
these
things
that
we're
experimenting
with
we're,
not
sure
if
we're
going
to
keep
them.
These
are
virtual
service
badges.
So
if
you
see
that
there's
a
little
get
their
little
git
branch
symbols
for
now,
you
can
see
up
here
has
virtual
service,
so
it
might
do
that.
It
just
tells
you
hey,
there's
a
virtual
service.
B
That's
you
know,
send
the
request
to
this
workload
or
this
app.
It's
just
a
little
reminder
to
show
you
that
you've
got
virtual
services
to
find
here
this
little
symbol
here
we're
going
to
change
this.
This
just
indicates
it's
there's
a
missing
sidecar.
Now
this
is
actually
a
bug,
because
the
ingress
gateway
is
nothing
but
a
proxy.
B
Address
we're
looking
for
the
sidecar,
annotation
and
ingress
has
a
connotation,
but
if
and
if
you
know,
for
whatever
reason,
I
deployed
a
deployment
in
my
mesh
and
I
forgot
to
manually
inject
a
sidecar
you'd
see
this
little
badge
here
and
you
go
I
need
to
put
a
sidecar
here,
there's
something
wrong
here.
That
was
the
purpose
of
it
again.
I,
don't
know
if
we're
going
to
keep
it
or
not,
but
these
badges
are
still
under
discussion.
E
B
B
F
F
B
Know
when
I
mean
we're
thinking?
Okay,
it's
not
it's
not
mom!
You
know,
okay,
we're
thinking!
You
know
weeks
at
most
okay,
so
you
guys
also
what
we
did
with
the
helm
chart
right.
We
had.
We
had
to
take
the
the
oh
five
out
our
profile
right,
because
it
just
would
have
come
in
completely
broken.
People
would
have
got
in
the
key
ally
and
things
just
wouldn't
look
right,
so
we
thought
it
was
best
just
to
have
the
helm
chart
fail
to
pull
the
image
you
know
anyway.
B
The
the
one
question
I
didn't
want
to
ask
before
I
hand
it
back
over
to
Doug
was
the
question
that
we
had.
The
Keala
team
had
today
was
the
notion
of
the
uniqueness
of
app
and
version
combinations.
And
what
can
we
assume
if
we
have
a
unique
app
version
combination
and
more
specifically,
if
I
have
an
app
to
version?
One
can
I
assume
that
that
is
a
one-to-one
relation
to
some
workload
or
can
I
have
multiple
workloads
with
apps
ooh
version.
One
labels
on
them.
A
B
B
F
B
A
B
D
A
Thanks
very
much
I
know
that
I
had
sort
of
been
asked
in
a
couple
different
settings
if
there
was
gonna,
be
a
demo
Ciotti.
So
this
was
perfect.
Thank
you.
Okay,
good
I,.
A
E
C
E
I'll
sit
next
to
them.
I
sent
it
locally
we'll
talk
about
week
ago,
and
the
problem
is
I'm
trying
to
so.
Here
is
how
do
we
let
the
Czech
adapters
to
give
back
routing
directives
the
proxy
and
that
led
me
to
thinking
how
we
can
actually
combine
different
decisions
across
different
rules
and
because
it's
not
clear
to
me
whether
we
associated
directly
with
the
whole,
with
some
high-level
policy,
so
I
wrote
a
book
about
how
do
we
combine
goals
in
general
and
found
that
we
have
a
very
strict
policy?
E
All
rules
that
we
have
in
mixer
and
that
does
know
it's
all
the
issues
with
example.
How
do
we
to
whitelisting
indicate
when
you
want
to
have
an
optional
only
have
to
hold,
and
you
want
to
select
based
on
one
of
them?
Not
both
and
that's
not
currently
possible,
so
I
decide
I
suppose
to
add
another
config
to
describe
how
we
combine
syllables
and-
and
that's
that's
what
the
talk
is
about
and
I
also
added
that
config
routing
directive,
so
we
operator
can
decide
which
adapter
contributes
to
the
directive.
E
E
So
I
check
policy
config
that
is
basically
combinational
a
set
of
rules
and
then
how
you
do
logical
combination
of
those
and
another
part
of
it
is
directive
template
a
template
is
just
a
string.
Vibration
of
directive
in
the
policy
where
each
string
is
an
expression
and
expression
can
be
friends,
attributes
a
tool
produced
using
certain
key
inches
like
you
just
put
dollar
sign
who'll
and
then
talk
about
a
beacon
that
news
page
so
that
Pelagian
so.
C
C
F
F
C
So
okay
I
do
think
the
crude
model
is
is
certainly
too
restrictive.
I'd
like
to
explain
why
and
if
we
can
so,
and
my
first
proposal
is
to
say
all
right.
Let's
say
we
just
want
to
add
extra
support,
it's
more
than
just
oh
and
we
support
or-
and
maybe
some
like
parentheses
kind
of
situation,
I
think
we
can
express
that
streak
expression
language
as
that
it
exists
today,
as
opposed
to
introducing
new
types
of
config
resources
where
you
have
to
list
to
list
the
rules
and
put
say
this
is
or
vs.
is
this
I.
E
E
C
C
E
C
C
So
I,
unfortunately,
I
didn't
have
time
to
read
your
whole
proposal,
yet,
probably
by
tomorrow,
I
I
do
think
the
value
is.
We
need
this
kind
of
functionality
within
the
mixer
context
and
separate
from
that.
If
the
long
term
roadmap
we
end
up
with
a
general
runtime
that
can
evaluate
these
things,
we
get
to
decide
the
language,
that's
above
that
runtime
right.
So
how
restrictive
it
is
how
general-purpose
it
is.
It's
it's
orthogonal
to
D
what
the
capabilities
that
are
runtime
on
it,
because
we've
been
talking
about
using
it
like
the
webassembly
runtime,
for
example.
C
E
Yeah
I
mean
that's
that's.
My
goal
is
to
basically
define
a
spec
for
this.
What
what
exactly
are
we
doing
in
this
batch
logic?
So
we
can
list
it
all
out
both
mixer
and
move
it
anywhere.
We
want,
but
currently
it's
a
bit
of
kind
of
hard-coded
and
in
the
code,
which
is
not
ideal,
which
you
can
make
a
decision,
what
it
should
be
like,
and
then
we
can
just
implement
this
in
mixer
that'd
be
easier
to
move
forward
to
moving
that
project
anywhere
upon
yeah.
E
E
F
E
E
These
two
ways
you
can
either
write
like
explicit
combination-
or
you
say
it's
all-
am
belittle
or
cuz
I
what
you
can
break
up
at
the
do
it
in
the
mall
right.
So
we
become
the
people,
all
the
rules
and
expect
to
end
it
it's
difficult
to
tell
them
to
just
misses,
number
and
work
if
you
had
another
one.
That
is,
if
you
had
this
policy
position,
hold
embedded
in
the
policy.
C
I
think
from
a
compatibility
state
for
it,
if
you
just
say,
look
if
there's
no
check
policy
resource,
then
it's
all
and
and
as
soon
as
you
add
the
check
policy
resource
from
that
point
on
that
namespace
you're
on
the
policy
plan
and
all
your
rules
must
be
listed
in
the
policy.
I
think
that's
reasonable.
F
About
stuff
in
this
kill
system,
or
from
that,
we
should
probably
start
talking
about
how
we've
migrated
away
from
that.
So
a
namespace
policy.
So
then
they
would
be
independently
namespace
code
when
this
to
system
would
have
a
document
like
this,
it
should
produce
an
answer
and
a
namespace
we'll
have
another
one
of
these,
that
population
answer
and
then
you
so
I
like.
E
F
E
E
C
H
I
ask
a
quick
question:
you
know:
did
you
guys
look
at
models
that
don't
require
other
resources,
I'm
I'm
worried
that
this
is
gonna,
be
like
in
practice,
unusable.
G
H
H
So,
as
opposed
to
say
in
the
existing
rules,
be
able
to
say,
like
oh
yeah,
this
is
you
know
this
isn't
allow,
rather
than
a
deny
default
or
whatever
right,
like
I,
think
there's
other
way
so
I
don't
require
a
separate
rapper
resource
that
then
now
you
have
to
manage
those
and
figure
out
how
those
all
get
connected
and
wire
it
up
in.
So
actually
that
we're
standing
this
system.
F
So
so
I
think
that
I
don't
think
waters
like
I,
don't
mean
what
you're
saying
is
equivalent
right.
You
cannot
express
things
that
the
rule
level
that
you
can
express
at
the
combination
level,
so
there
is
just
there
is
just
a
difference
of
that.
There
is
a
preacher
gap
between
them.
If
we
are
okay
with
that
feature
of
the
web
gap,
then
yes,
then
we
can
just
add,
like
more
properties
and
bhuvan's
and
things
like
that
to
the
rules
themselves
and
wait.
F
H
You
know
optional
right,
like
fall
through
verses,
not
fall
through.
Basically
right
tell
it,
you
can
say
whether
things
are
or
and
the
only
a
one
level
right.
You
don't
get
arbitrary
complexity,
but
I
think
I'm,
arguing
arbitrary
complexity
is
probably
bad
like
we
should
just
solve
the
immediate
missing
use
cases
and
not
design
a
programming
language
without
actually
letting
them
just
use
average.
E
H
C
You
need
an
order,
the
precedence
yeah.
You
need
some
sort
of
warning
in
precedence
yeah,
so
that
so
the
problem
is
right.
Now
the
rules
are
all
equivalent.
There
is
no
there's
fundamentally
by
definition,
no
order
in
them,
so
either
you
need
to
put
numbers
next
to
them
saying
this
is
one
two
three
four
five
at
which
point
I'd
argue.
Your
complexity
factor
is
not
really
reduced
right.
A
H
H
H
Just
adding
like
the
minimal
saying,
like
you
know,
being
on
the
mark
things
saying
one
of
these
need
to
pass,
and
actually,
if
you
needed
to,
you
could
do
named
groups,
and
then
you
get
the
naive
ordering
to,
but
maybe
that's
not
great.
My
meta
comment
is
just
I'd
like
to
see
more
of
a
discussion
of
alternative
ways
of
solving
these
problems.
Also.
C
And
one
alternative
is
to
say
what
we
have
right
now
is
good
enough,
and
the
next
step
after
that
is
supporting
a
real
programming
language
period
right
like
now.
Yes,
that
would
that
should
be
one
of
the
options
considered.
Cuz
cuz.
All
this
stuff
would
be
so
much
easier
to
just
write,
simple
functions
in
JavaScript
or
something
as
opposed
to
all
the
CMO.
E
F
As
collected
on
other
thing
is
right:
he
is
in
this
case
it's
expressing
a
combination
overrules,
the
other
alternative
which,
which
actually
I
think
it's
more
natural,
is
to
express
its
expressed
combination
over
result
of
adaptors.
Only
directly,
you
don't
care
how
you
reach
the
adapter
right.
There
may
be
multiple
paths,
all
of
which
dispatched
to
some
whitelist,
and
the
combination
that
you
want
ultimately,
is
what
what
did
the
actual
adapter
say,
not
how
you
got
to
the
adapter
okay.
C
F
F
So
no
you
don't
express
that
in
the
context
of
the
rule,
so
this
like
what
would
what
I'm
saying
requires
a
high
level
constant
that
says
dispatch
to
whatever
adapters
you
want,
and
rather
than
ending
the
results
here
is
the
actual
expression
that
are
that
I
want
you
to
use
to
combine
them.
So
it
doesn't
go
towards
simplification,
but
I
think
it's
closer
to
believe
the
way
I
think
about
combining
responses
from
magnets.
Oh
yeah,.
C
So
right
now,
in
a
rule,
you
specify
the
list
given
I,
given
this
rule,
I
want
to
send
these
instances
to
these
handlers.
So
if,
if
instead
of
just
a
list
or
if
you
just
added
another
stanza
that
says
okay
give
it
all
these
handers
I'm
calling
please
apply
this
resolving
expression,
and
it's
just,
and
so
it's
a
it's
an
expression
that
operates
over
the
results
of
those
handlers
that
be
a
single
additional
line
and
config
to
support
this.
You
still
don't
you
still
can't
combine
multiple
rules
results
together,
I'm,
not
sure
that
matters.
E
E
A
So
I
think
we're
about
the
time
that
I
know
there's
other
stuff
on
the
agenda.
Maybe
we
can
I
think
this
is
warrants
more
discussion
and
some
thought
on
the
milling
listen
to
the
doc.
So
let's
continue
to
do
that
and
maybe
continue
this
discussion
at
the
next
meeting.
Sure
yeah
and
you
I,
don't
know
that
we
have
time
for
streaming
protocols
wearing.
F
E
A
A
C
F
A
B
Use
bleach
has
anybody
use
blue
jeans,
no.
H
C
A
Okay
and
let
there
any
objections,
then
we'll
plan
on
getting
it
set
up.
Let's
figure
out
the
record
because
easy
to
do.