►
From YouTube: Istio API Review Meeting 3.22.19
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
D
C
C
So
one
thing
is
that
we'd,
like
have
more
of
a
feature,
focused
API
and
get
away
from
naming
individual
control,
plane,
components
and
that's
something
that
spend
started
as
a
sort
of
a
high-level
objective
or
for
install
sto
and
and
just
how
the
history
of
control
plane
would
be
structured.
So
we'd
like
it
to
be
more
feature
focused
and
in
addition,
we
have
quite
a
bit
of
kind
of
arbitrary
mapping
between
hound
parameters
and
kubernetes
resources,
and
that's
something
that
really
could
stands
cleanup
as
well,
because
you
know
it's
it's
fairly.
C
Inconsistent,
like
the
HP
ei
Matt
thinks,
for
example,
differ
by
components
and,
and
they
don't
all,
have
the
same
set
of
source
parameters
going
to
the
the
autoscaler
kubernetes
resources.
And
then
we
we
have
kind
of
kubernetes.
We
source
requests
kind
of
scatter
all
over
the
place,
so
it'll
be
nice
to
present
more
familiar.
C
So
there
is,
there
is
a
it's
it's
exhaustive
in
terms
of
options,
what
it
doesn't
currently
have
is
is
direct
support
for
the
add-ons
and
that's
something
else
that
there
is
a
goal
of
this
this
API
is
you
define
a
way
that
we
can
reference
external
without
without
having
to
maintain
the
the
corresponding
charts?
You
know
in
our
install
package,
okay,.
C
C
Initially,
yes,
because
the
underlying
implementation
and
the
operator
is
based
on
the
home
templates,
okay,
long
term
it
something
we
can
consider
getting
rid
of
so
I
think
the
first
step
is
to
have
an
API
that
we
can
migrate
users
to
and
the
the
least
painful
way
of
doing.
That
is
to
maintain
the
existing
templates.
Okay.
E
G
H
B
C
I
G
H
G
That's
oh
yeah,
it's
more
if
they
like,
if
they
don't
want
an
operator
running
in
the
cluster.
Instead,
they
want
to
do
yeah
the
helm,
template
mode
and
they
don't
want
to
use
the
operator
template
mode
for
some
reason,
but
yeah.
We
have
to
find
that
out.
The
goal
is
the
goal.
This
didn't
remove
the
need
for
a
help,
because,
ultimately,
you
can
also.
C
Exactly
so
so
there
are
some.
There
are
some
options
that
are
not
explicitly
in
the
API,
but
they're
still
reachable
through
using
overlays,
and
that's
that's
actually
a
feature
of
the
operator,
design
and
you'll
actually
see
so
maybe
I
can
just
explain
how
that
works
right
now.
There
is
a
field
in
each
feature
which
is
right
here.
So
I
hope
you
can.
You
can
see
my
highlighting
there.
Yes.
I
C
There's
there's
a
kubernetes
object,
override,
which
is,
which
is
basically
a
customized
style
overlay,
customized
being
the
the
with
the
case,
spelling
and
the
Z
spelling,
allowing
you
to
actually
overrides
kubernetes
level
resources.
So
some
of
the
parameters
that
we
currently
support
explicitly
as
configuration
parameters,
would
end
up
being
an
an
overlay.
H
C
H
H
Was
a
presentation
in
environments
few
weeks
ago,
where
workshop,
if
I
believe,
was
presenting
their
way
to
deploy
SEO,
and
so
is
they
are
doing?
It
is
a
chicken
template.
They
have
the
temporary
desire,
CIC
system,
so
I
think
one
requirement
is
to
be
able
to
generate
various
templates
that
can
be
checked
in
and
again
plug
into
what
custom
CHD
pipeline
right.
C
So
I
you
know
initially
home
template
is,
is
going
to
be
the
the
underlying
implementation
of
how
the
manifests
are
rendered
and
and
long
term
I
think
we
can
discuss
getting
rid
of
it.
Although
I
think
there
are,
there
are
some
attractive
things
about
the
helm,
template
commotion
of
how
you
know
tiller,
I
I
think
is
debated,
oh
how
good
that
is,
but
certainly
the
template
part
of
it
I
think,
has
some
nice
attributes.
C
G
So
I
think
there's
two
there's
two
orthogonal
things
here
right.
So
one
is
kind
of
the
input
to
like
in
the
CI
CD
system,
and
one
is
the
output
and
the
output
there's
two
options.
One
is
the
output
is
just
a
bunch
of
kubernetes
resources,
then
some
CI
system
goes
and
makes
occur
in
the
cluster
and
the
other
option
is
the
operator
is
running
in
the
cluster
and
it's
doing
that
right,
like
it's
going,
the
role
of
CS
uses
cacb
and
on
the
other
side
of
it.
H
H
H
H
G
E
K
H
J
H
J
E
E
G
G
E
E
G
H
E
H
K
K
Fact
that
is
the
kimono
in
front
of
the
meta
mean
from
that
I
think
that's!
That's
that's
where
that
actually
is
goes
to
the
other
side
of
the
line
right.
So
there
is
the
okay,
it
is
the
East,
your
control
plane
configuration
and
then
there
is
a
qualification
of
how
we
generate
artifacts.
Ryo
runs
and
configuration
for
the
system
right.
So,
for
example,
that
particular
one
I
think
falls
into
the
other
category.
We
can
keep
you
capture
it's
here.
I
think
it
will
be
right.
K
G
H
The
side
class,
but
if
we
want
to
categories
I,
think
we
should
make
a
distinction
between
features
that
we
expect
or
a
nice
to
implementation,
to
implement
like
your
service
and
so
forth,
and
implementation.
Specific
features
like
I
lot
happens
to
equal
to
support,
but
that
not
necessarily
require
to
have
a
confirm
at
least
implementation,
because
again
you
may
have
different
deployments.
People
may
turn
off.
Some
features
may
use
alternatives
to
try
for
oh
sorry,
three
scenes
and
different
options
to
ranch
component.
Some
of
those
are
specific
to
one
particular
implementation.
G
K
A
Google
analogy
right,
so
we
have
like
different
kind
of
configuration
stores
after
they
have
or
internally
which
is
like
similar
to
what
this
does
right.
You
have,
or
you
have
more
configuration,
then
you
have
like
components
that
actually
provide
configuration
at
the
runtime
where
you
can
listen
and
change
your
behavior
based
on
that
configuration,
and
then
there
is
the
the
your
regular
data
paths
for
us.
That
is
the
actual
East,
your
configuration
right.
So
that's
the
thing
I'm
trying
to
point
out
like
another
is
extra.
K
E
K
Lower-Level
yeah,
which
probably
means
like
those
things
change.
We
may
need
to
discard
a
lot
of
state
and
then
rebuild
it
in
you
know,
pilot
or
yeah
lead,
because
I
changed
like
some
DNS
thing,
for
example,
and
that
effects
up
a
bunch
of
things
right,
but
I
I
think
that
that's
something
that
would
be
expected
rather
than.
D
D
G
So
this
is
a
bouncy
man,
it's
her
yeah.
We
like
we
like
the
right
terms
with
a
spin,
so
I,
don't
think
thinking
about
component
versus
feature
is
nice
because
we
actually
do
want
to.
Actually
you
can
think
this
part
about
features,
but
this
is
about
whether
a
feature
is
enabled,
and
so
it
might,
by
enabled
kind
of
install
like
whether
it's
whether
it's
on
and
then
sort
of
the
basic
settings
for
that.
But
it's
like
the.
E
D
M
D
E
C
Help
it
does
have
I
mean
so,
for
example,
the
security
config.
If
we
look
at
that,
so
so
something
like
control,
plane,
MT
OS,
it's
not
really
component
based
yeah
first
domain
I
mean
other
things
are
currently
I've
put
these
as
features.
But
you
know
you
could
probably
restate
these
as
sorry
that
I
couldn't
currently
as
component
but
I
think
you
could
restate
these
as
as
just
feature
and
just
do
a
citadel
where's.
My
food.
G
H
Look
at
this
one,
it's
a
misty
or
what
components
are
installed
with
the
ones
are
running.
It's.
The
provision
is
a
platform
that
supports
a
number
of
AP.
Is
it
may
run
different
implementation
of
a
storage
each
piece,
a
missive,
came
five
determination
of
tracing.
There
will
be
two
in
addition
of
mix
right,
but.
H
G
I
think
what
I'm
trying
to
say
and
again
I'm
thinking
about
here
so
give
me
a
second
is
that
like,
if
we
had
the
mental
model
of
the
operator,
config
describes
kind
of
what
components
are
installed
and
enabled,
and
then
you
have
separate
series
to
describe
the
behavior
of
the
system
right
likely,
but
what
security,
what
are
the
global
security
settings
right?
What
are
the
global
settings
for
policy
etc
like
this?
Is
that
should
we
mix
those
in
one
place,
should
we
try
to
separate
them
out
yeah.
E
To
me
that
the
mental
model
is
the
line
is
pretty
simple
if
the
component,
so
let's
push
this
to
the
extreme,
you
could
say
all
see
are
these
in
is
theö
are
in
this
file
and
when
any
configuration
changes
are
made,
you
have
to
restart
the
components.
Cuz
they're,
not
dynamic
in
nature,
so
clearly
that's
stupid.
What
all
right
so
I
think
we
want
to
be
in
the
other
extreme
as
much
as
you
know,.
E
I
want
to
change.
There's
that
one
setting
I
care
about
I
want
to
change
that
I,
don't
need
to
run
helm
I,
don't
need
to
run
this
operator.
The
component
will
pick
it
up
and
deal
with
it.
This
really
should
be
just
about
the
stuff.
That's
again,
I
go
back
to
the
term
static
static
configuration
for
the
individual
components
that
are
running
well.
G
G
E
G
G
E
H
G
E
G
G
G
Like
just
as
an
exercise,
split
them
out,
I,
don't
know
just
in
a
dock
or
whatever,
and
just
look
at
look
at
them
in
isolation.
And
what
are
the
component
things
so
like
what
names
tries
to
install
something
in
what
replicas
count
whether
that
component
is
on
or
not
right
like
clearly,
those
are
all
component
configurations.
H
C
N
C
H
G
G
H
Think:
okay,
one
more
comment
here
before
we
forget,
because
that
that's
an
excellent
point,
keep
in
mind
that
we
are
trying
this
for
the
environments
where
you
look.
One
of
the
goal
is
to
move
to
a
gradual
upgrade
where
we
don't
have
in
global
msconfig.
We
have
a
stable
production
config.
We
have
multiple
environments
in
two
different
configuration.
We
gradually
shift
up
shift
from
from
one
to
another
so
and
we
also
want
to
have
a
space
coffees,
which
name
is
Patriot.
H
Everything
we're
talking
here
should
be
very
environment.
What
recording
Socrative
solar
meanings?
That
means
spaces
will
kind
of
point
to
a
particular
environment
or
be
able
to
override
specific
settings
in
some
way,
because
customers
want
to
shift
gradually
to
any
configurations.
They'd
want
to
automatically
kind
of
turn.
Security
won't
have
any
see
more
question.
H
Believe
me,
my
environment.
In
that
context,
environment
is
that
is
a
set
of
configuration
set
of
components.
You
have
personal
production
environment
that
is
running
etiological
zero
and
you
have
an
environmental
running.
Is
you
wander
to
one
another
environment?
Is
your
master
and
you
have
another
100
East,
your
master
with
MPLS,
so
we
have
different
preset
mesh
configuration
with
different
companies
different
whatever
and
at
runtime.
The
namespace
can
point
so.
D
K
Can
throw
each
other
everything
else
can
be
different.
Okay,
I
am
confused.
Now
I
told
you
were
talking
about
the
control
plane,
components
in
that
context.
Right
so
I
actually
deploy
control,
plane,
components
in
those
environments
separately,
but
the
taxonomy
doesn't
apply
to
the
control,
plane
components
and
it
is
just
my
staging
or
my
no
no,
no
budget
environment.
It's.
H
E
I
think
all
that's
pointing
out.
There's
there's
two
different
things:
there's
using
environments
for
different
versions
of
different
versions
of
the
CEO
or
configurations
of
Sto
yep,
as
opposed
to
the
applications
that
are
running
on
top.
This
is
my
staging
environment
versus
my
production
environment.
That's
a
different
I.
K
H
H
K
K
G
You
install
this
plus
an
operator
into
a
namespace
that
is
an
instance
of
the
sto
control
plan,
but
let's
just
call
her
now,
any
namespace
can,
within
a
mesh
of
clusters,
can
point
to
any
of
the
different
control
planes
running
in
that
cost
in
that
match,
but
you
gotta
support
it
on.
Where
are
this
is
where
the
runs.
So
my.
K
Point
is
essentially
I
have
a
master
file
that
describes
my
topology
of
all
of
my
mesh
right
just
for
the
sake
of
arguments
and
then
I
can
go
ahead
and
say:
okay,
I'm
going
to
deploy
to
the
namespace
food
and
the
namespace
food
section
of
this
master
deployment
plan
or
master
configuration
applies
to
that
control
plane.
Is
that
how
this
is
it's
supposed
to
be
set
up?
Yes
and
a
different.
H
K
G
K
C
Yeah
sure
so
the
I
started
off
with
403,
actually
and
and
what
I
realized
was
that
photo
three
uses
uses
for
scalars.
It
doesn't
use
pointers,
it
just
uses
the
the
actual
type
and
what
that
means
is
that
you,
you
can't
specify
unset,
it's
just
it's
just
not
possible
and
and
where
that,
where,
where
that
becomes
a
problem
is
where
you
have
some
set
of
defaults.
C
Let's
say
you
have
some
installation
profile
and
now
the
user
would
like
to
override
that
by
specifying
something
in
this
CR,
for
example,
if
something
is
by
default,
enabled,
let's
say
a
feature
is
enabled
in
in
the
default
profile
and
the
user
wants
to
to
disable
it,
so
they
they
would.
They
would
say
in
a
way,
folks
want.
D
So
the
more
semantics
here
are
kind
of
explicitly
I
mean
we
expect
people
to
use
more
semantics
here,
whereas
in
our
normal
series
the
merge
is
defined
at
the
API
level.
It's
not
like
it's
never
going
to
the
case
that
they're
gonna
have
to
virtual
services
being
physically
merged
into
one
and
then
interpreted
by
the
API,
whereas.
I
E
C
K
C
G
So
I
think
that's
not
quite
what
I
was
saying
saying
that
once
so,
once
someone
has
installed
is
do
right,
whatever
default,
they
got.
That
default
sticks
around
regardless
of
updates
right
that
we
don't
have
the
need
to
change
those
things,
because
that
would
actually
change
behavior
right.
The
breaking
change
well,.
E
I
D
D
C
G
E
G
D
I
D
K
P
C
E
G
I
E
D
E
M
M
E
Controlling
which
config
gets
installed
right,
which
Tod
is
populated
downstream,
all
right.
So
it's
it's
setting
the
defaults
for
the
runtime
parameters,
the
runtime
state
and
it's
the
same
thing
any
other
adapters
as
well.
So
it's
gonna
it
it's
getting
different
with
mixer
v2,
because
now
we
also
have
the
question
of.
If
you
not
only
do
you
want
to
enable
the
config,
but
you
actually
want
to
deploy
the
adapter
in
your
cluster
yeah.
G
C
D
G
D
M
G
E
But
so
where,
when,
if
I'm
running
the
operator-
and
they
say
I
want
to
turn
on
the
standard
I/o
stuff,
where
does
the
config
come
from
to
populate
that
the
CRV
definition?
Are
you
saying?
Where
does
it
happen
now
or
very
uncommon
in
this
world,
with
the
with
the
operator?
Where
does
that
see
the
state
come
from
that
it's
gonna
push
into
the
cluster,
so.
G
G
D
E
If,
if
I'm
messing
around
with
this
stuff,
so
that's
a
dynamic
change
as
it's
going
to
do
stuff
and
there's
an
actual
CD
once
the
system
is
running,
the
user
is
going
to
look
at
the
cube
and
VCR
D
to
make
changes
in
the
behavior
of
that
component.
Correct,
not
here
but
more
they
if
I
did
some
options,
and
here
it
might
actually
be
overriding.
My
see,
Arnie
that's
of
the
in
the
cluster.
Yes
yeah,
so
I
think
that's
kind.
D
G
G
I
E
D
So
yeah
I
I
agree,
so
this
is
actually
even
more
more
specific
than
that
right.
What
all
this
is
saying
is
that
what
all
do
you
want
to
be
enabled
by
default,
but
they
are
all
there
all
runtime
features,
so
this
is
only
going
to
decide
whether
to
install
the
stdio
CRE
or
not
once
it
is
installed,
the
operator
is
free
to
change
the
behavior
in
that
CR
T,
because
the
CIT,
the
CR
is
our
API,
not
necessarily.
G
E
That's
I
think
the
behavior
you
want
here
is
to
make
those
so
they're
just
enabled
boolean
period
everything
else
you
go
and
mess
around
with
the
Sierra
Leader
to
the
system
and-
and
this
probably
add
factor
Scrolls
another
flag
different
than
the
Sierra
D,
that's
kind
of
an
override
that
says
and
globally
we're
turning
this
feature
off
globally,
we're
turning
it
on.
There's
a
CR
knee
there
that
if
it's
on
this
is
a
CRT,
we're
gonna
use
for
it.
Yeah.
I
I
Simplified
at
home
today,
that's
the
other
thing
I'm
struggle
to
because
at
home
today,
it's
kind
of
also
you
can
fake
a
different
configuration
into
each
subject
for
a
pilot
or
proxy,
which
is
cycle,
injector
and
also
tonometry,
and
policy.
Now
we're
asking
user
to
deal
with
so
many
CR
DS
and
the
customer
resources.
How
is
this
qualify
a
little
hard
for
you,
yeah.
G
Because
I
think
the
alternative
right,
the
alternative
here
is,
you
say:
okay
Prometheus,
we
have
a
higher
level
API,
then
the
adapter
it
yeah
or
whatever
that
that
CR
is
and
it's
in
defined
in
here
somewhere,
and
you
only
use
that
and
you
don't
touch
the
other
one
I
think
that's
the
alternative
like
we
can't.
We
definitely
can't
have
a
please
touch
both
option
so.
E
It
turns
out
the
value
of
this
layer.
In
my
mind,
is
this
provides
a
mechanism
that
it's
it's
a
continuous
running
thing
and
it
provides
us
a
place
to
add
logic
to
deal
with
upgrades.
That's
the
main
value
I
think
in
this
yeah
right.
So
it
could
even
there's
nothing
in
here
about
Prometheus.
It
understands
that
I'm
going
from
v1
to
v2
and
I
need
to
go
little
any
Prometheus
resources
in
the
cluster
right.
I
E
H
Should
point
out
the
new
installer
we
kind
of
do
because
we
no
not
got
installed,
or
least
you
at
once.
We
install
each
component
individual
and
we're
all
changes
which
component
indicate
independently
and
each
component
is
started
running
different
namespace
like
you
cannot
do
it
and
have
the
camera
tossed
around
for
security
reasons.
I
said
one
of
the
changes
you
good
little
options.
H
Point
is
to
run
components
in
different
names
for
security
and
people.
Who'd
run
all
of
them
in
a
single
namespace.
If
they
don't
care
about
security,
that's
kind
of,
but
even
if
they
run
is
the
same
namespace.
They
are
independent
deployment.
So
you
deploy
mix
telemetry
in
event
of
this
mr.
policy,
independent
of
or
second.
G
G
G
From
an
install
and
upgrade
perspectives
and
how
we
want
to
do
things
like
enabling
Prometheus
versus
settings
for
Prometheus,
how
many
API
is,
do
we
offer?
What
does
that
look
like?
What
is
the
right
mechanism
for
us
to
make
progress
on
that
question
independent
of
getting
language?
Should
we
should
we
down
the
scope,
the
Installer
to
just
on
the
like
core
component
stuff
and
then
try
to
make
progress
there
or
like
how
do
we
want
to
go?
This
is.
E
E
G
D
C
I
C
I
think
that's
that's
also
something
that
should
be
looked
at
as
a
cleanup
action,
because
you
know
that
that
values,
though
llamó
is,
is
something
that
has
just
grown
over
organically
and
it's
it's
evolved
in
in
something
that's
kind
of
difficult
to
to
render
into
a
nice
API
right.
So
this
is
kind
of
an
attempt,
but
it's
it's
not
easy.