►
From YouTube: Policies and Telemetry WG Meeting - 2018-10-24
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
B
B
So
we
are
particularly
interested
in
in
sto
mixer,
adapters
and
telemetry,
so
we
are
trying
to
build
some
mixer
adapters
for
Netta
product
on
cloud
in
sight.
So
basically
they
joined
first
time
here
to
to
collect
some
information
and
threat
veterans
and
the
product
list.
You
itself
and
everything
else
great.
A
And
I
think
we
have
others
on
the
call.
I
see
videos
here
that
have
done
some
other
walkthroughs
recently
of
in
Scott
I,
see
you
as
well
front
of
process
adapter.
So
should
you
have
questions
I
think
we
have
a
pretty
decent
community
here
for
walking
through
that
process,
so
feel
free
to
reach
out.
Okay,.
C
A
A
Okay,
so
I
have
one
item
that
I
wanted
to
bring
up
a
while
back.
We
introduced
this
use
adapters
here,
d5
for
a
mixer
to
try
and
provide
a
wave
so
that
mixer
could
start
up
without
perhaps
checking
all
of
the
individual
CDs
for
each
individual
type
of
adapter,
but
instead
use
just
the
adapter
CRD
for
listening
and
configuration
and
we
set
it
default
like
so.
We
set
it
to
true,
so
that
was
the
existing
behavior
by
default.
A
But
some
of
the
people
who
have
been
experiencing
API
server
performance
issues
have
noted
that
when
they've
tried
it
with
accept
the
false
I've
seen
things
become
stable
and
all
of
a
sudden
start
working,
and
this
is
particularly
on
eks,
so
I
think
we,
my
suggestion
here,
is
that
we
update
the
default
settings
that
flag
to
be
false
for
one
one
and
I
wanted
to
run
that
by
the
group.
So.
A
G
E
E
F
A
H
F
D
E
D
D
D
F
E
E
E
D
D
D
C
A
D
So
for
for
tell'em
for
telemetry
in
a
large
cluster,
what
seems
to
have
worked
so
far
is
scrape
interval
we
reduced
down
to
30
seconds
or
another
increase
to
30
seconds,
and
the
Prometheus
adapter
supports
metric
expiry,
which
will
remove
metrics
that
are
not
used
after
10
minutes.
So
doing
both
those
reduces
the
amount
of
stuff
that
Prometheus
things
to
hold
on
to,
and
these
numbers
have
worked
in
large.
A
Much
Koster,
yes,
so
just
yet
we
cleared
the
Prometheus
adapters
holding
onto
metrics
forever
by
default,
and
this
is
a
problem
if
you
have
large
number
of
deployments
happening
over
a
day
or
anything
like
that.
So
I
think
so
is
it
recommendation
to
move
these
values
into
the
default
four
one?
One?
Yes
and.
F
A
I
Just
the
other
day,
so
one
of
the
I
have
this
a
working
tool
which
would
generate
the
boilerplate,
like
scaffolding,
Co
for
moving
company
adapter
to
auto
process,
adapter
and
still
history
working
progressed
and
we
set
like
a
definite
developer,
doesn't
have
to
write
extra
cold
extra
go
code
to
make
it
all
about
to
make
up
having
adapter
into
auto
process.
It
just
need
to
generate
need
doctor
file
or
the
helm,
template
all
those
stuff,
yeah,
but
I
know
we
want
to
help
them
to
do
that
as
well.
Like
Germany
example,
doctor
file
and.
F
D
K
D
E
D
And
actually
another
thing
to
go
out
there
for,
for
one
one
is
that
our
default
install
doesn't
get
install
all
the
instances.
Sorry,
all
that
all
the
templates
or
does
does
that
pair
only
do
that
all
the
templates
ones
are
you
thinking
are
missing.
Did
the
other
process
templates?
Oh
right,
because
I
have
pointed
people
multiple
times
to
artifacts
that
we
have
already
generated
and
saying:
okay
just
apply
these,
but
I
think
we
should
just
put
them
in
our
install
so
that
so
this
select
the
standard
templates
in
the
dynamic
form
or
already
available
I.
L
M
A
L
D
J
D
D
So
for
operations
that
have
a
straight
cube,
CTL
command,
I,
don't
think
we
should
have
an
STL
CTL
come
on
right.
It's.
It
is
possible
to
list
adapters
using
coop
CTL
today
and
look
at
their
configuration
so
having
to
is
like,
unless
there
is
actually
no
I
think
having
to
is,
is
just
bad
I.
Don't
think
there
is
any
good
reason
to
have
to
it's
yeah.
D
We
moved
away
from
that
right,
all
the
deputy
redundancy,
so
it
should
only
contain
like
net
new
function,
which
cannot
be
provided
by
cube
City,
our
validation
right,
nothing,
so
so
some
sort
of
validation
like
running
some
tests,
like
some
validation,
desk
acceptance,
tester
like
it,
has
to
be
on
that
flavor.
So
let
the
song
so
say
you
can
I
mean
I
would
say
that
if
you
come
up
with
just
a
proposal
or
a
write
up
first
sure
before
POC
right,
we
just
see
okay.
A
The
other
thing
I
mind:
I'm,
going
to
bring
up
as
the
agenda
was
the
fair
closed
versus
fell
open.
It
looks
like
at
some
point
right
before
100
we
switched
over
and
only
supported
failed
closed
for
policy
without
the
option
to
override
it
in
support
Kelvin,
so
they've
been
a
bunch
of
PRS
in
the
last
day,
I
think
that
have
been
added
to
master.
We
need
to
decide.
No,
we
are
to
one.
Oh,
let's
make
a.
A
F
F
F
D
F
D
When
it's
not
really
not
a
processor
after
thing
right,
even
for
him
present
after
that
communicates
with
outside,
you
have
all
the
same
issues
like
what
happens
when
that
thing
can't
communicate.
So
it's
not
really
an
out
of
processing.
It's
just
just
like
what
happens
when
external
services
are
not
available,
you
treat
it
as
a
success
or
failure
or
you'll
end
up
client
decide
like
right.
Now
we
don't
give
client
the
granularity
of
that
result.
Right
did
you
say
almost
the
result
of
this
call.
H
H
A
A
A
I
E
N
N
E
N
D
And
I
think
I
think
we
can
so
the
question
there
is
caught
in
anyone
else
impose
things
experimented
with
this.
Is
that
what
we
had
talked
about
even
in
the
initial
designs,
is
that
using
the
non
session
based
where
we
send
the
whole
country
with
with
every
call
it's
possible
for
the
adapter
to
optimize
it
by
just
having
a
pointer
right,
some
sort
of
pointer
in
the
config
itself,
and
then
that's
like
one
string
or
something
and
then
speed
after
saying,
oh
okay.
Yes,
this
thing
means
what
about
the
actual
configuration
is
and.
D
Is
already
you
can
fix
the
restraint
which
is
a
server
or
galleon
right
I
mean
you?
Can
you
can
always
say
that
look,
my
my
configuration
and
which
is
why
I
said
like
so
pointer
to
a
blob
store
or
some
API
server
location?
That
is
your
config
inside
the
actual
adapter,
so
yeah
you
get
that
send
every
time,
but
that's
very
small,
and
then
the
adapter
itself
is
looking
up
the
real
config
and
like
doing
that,
so
has
has
anyone
experimented
with
that
or
have
you
Scott
or.
N
N
E
N
N
D
E
D
N
D
Yeah
we
we
have
stayed
away
from
actually
providing
these
sort
of
services.
So
right
what
would
get
us
that's
something
to
consider,
though
I
mean
if
it
gets
over,
is
a
blobstore
of
some
time
right,
yeah,
Bob's
yeah.
So,
like
you
can
say,
look
my
configuration
resides
at
config
map,
whatever
blah
blah
blah
I
know
if
Sigma
is
much
loved,
but
something
else
like
that
and
then
config
maps
can
be
mounted
in
file
system,
so
that
offers
a
bit
of
decoupling
and
then
you
can
read
from
there.
D
D
It's
not
persistent
state,
but
what
I
did
did
yes,
the
require
state
management
and
a
like
for
for
things
that
completely
reside
out
out
of
the
mesh
I
fully
expect
those
adapters
to
the
things
that
I
described
anyway
right
way
when
you're
calling
from
mesh
to
a
completely
different
thing
like
something
or
the
internet,
I
expect
like
an
API
key
and
some
some
other
token
to
pull
up
all
the
other
configuration
from
the
other
side.
Right
like
that's.
N
E
D
D
F
D
L
A
L
But,
for
example,
for
the
book
and
for
demo
I
do
see
the
sources
are
known
even
for
the
requests
that
go
to
the
ratings
service
say
so
in
the
book
info
demo
we
have
the
rating
service
right
and
yes,
the
requested
that
goes,
we
reduce.
So
in
that
case
it
shouldn't
be
unknown
right
right.
You
should
not
be
seeing
unknowns
for
that,
but
I
think
you
do
see
it
as
unknown.