►
From YouTube: Istio Community Meeting 8-23-18
Description
Chat transcript at https://gist.github.com/thisisnotapril/563fa2a4eddcc9fc3e67c1e6f8c6d247
A
All
right
good
morning,
everyone
can
you
hear
me:
okay,
awesome!
Thank
you
all
right.
We've
got
a
good
crew
here
already,
which
is
fantastic
and
one
sure
we'll
have
some
folks
join
in.
As
you
know,
people
wrap
up
earlier
meetings
and
things
like
that,
I
just
have
a
few
general
announcements
that
I
wanted
to
share.
First
is
that
envoy.
Con
is
happening
the
day
before
coop
conciencia
con,
which
is
in
Seattle
in
December,
and
because
of
course,
you
know
envoy
is
a
big
part
of
this.
A
A
It's
got
all
the
information
about
the
event
itself
and
then
it
has
actually
I
think
that
link
got
changed
so
I'll
check
errect,
it's
got
all
the
information
there
and
then
it
also
has
a
link
to
the
CFP
highly
recommend,
checking
that
out
and
submitting
to
on
boy
con,
especially
if
you
had
something
that
you
had
maybe
proposed
for
the
larger
coop
con
that
related
to
envoy
as
well.
The
other
thing
I
wanted
to
call
out
is
this
awesome,
sketch
I,
don't
know
if
you
saw
it
on
Twitter
but
Denise
you
from
pivotal.
A
Did
this
amazing
like
what
is
it's?
Do
sketch
I've
been
waiting
for
us
to
have
something?
You
know
cute
like
this?
Someone
with
the
designer
skills
kubernetes
had
like
the
children's
illustrated
guide
to
kubernetes,
which
is
a
great
kind
of
intro.
Explanation
and
I'm
super
excited
that
we're
starting
to
get
some
design
folks
in
the
community
that
can
create
some
cool
stuff
for
us,
so
do
check
that
out.
I'm
also
who's
chatting
with
her
about
getting
some
postcards
made.
Getting
it
actually
printed
on
something
that
you
know,
people
could
like
hand
out.
A
You
could
hang
in
your
cube
or
hand
out
it
meetups
and
things
like
that.
So
I
think
it's
a
really
cool
kind
of
high-level
look
at
what
sto
is,
which
is
a
question.
Should
we
all
get
a
lot,
and
now
you
can
answer
it
a
little
bit
easier,
so
do
check
that
out
and
and
let
Denise
know
your
thoughts
on
that.
Does
anyone
else
have
any
announcements
or
anything
else
they
want
to
share
before
we
jump
into
questions
and
the
upcoming
meetups
or
conferences
that
folks
are
speaking
at.
B
A
C
A
Cool
I'll
make
sure
we
have
that
on
the
agenda,
I
mean
on
the
calendar.
Sorry,
the
community
calendar
you
anybody
else
have
anything
they
want
to
call
out.
I
did
see
some
stuff
on
the
Twitter's,
with
some
meetups
that
folks
have
planned
so
I'll
be
sure
to
add
those
to
the
community
calendar
as
well,
and
if
you've
got
stuff
that
you
want
to
add
to
the
community
calendar,
please
let
me
know,
excuse
me
and
we'll
get
that
in
I'm,
happy
to
give
you
privileges
to
add
things
yourself
to
calendar
it.
A
Just
please
don't
delete
anything
else,
actually
trying
to
get
a
special
permissions.
That
added
for
that
Christopher
put
a
link
and
there
would
be
an
issue
birds
of
a
feather
session
at
the
Cloud
Foundry
summit
EU
in
October.
So
that's
awesome.
A
lot
of
good
stuff
coming
up
I
feel
like
we're
hitting
another
wave
of
conferences.
We
had
our
earlier
wave
and.
A
All
right,
if
no
one's
got
anything
else
they
want
to
share,
we
can
jump
into
questions.
Julian
had
a
question
some
of
it
had
served
on
the
mailing
list,
but
I
encouraged
him
to
you
know.
Let's,
let's
talk
about
it
here.
I
think
this
is
a
good
opportunity
to
kind
of
have
this
discussion,
but
what
your
development
flow
looks
like
Julian.
Do
you
want
to
take
it
from
here.
A
C
A
I
was
gonna,
say
nothing
else.
You
know
if
thoughts
on
that,
do
you
check
out
the
discussion,
and
hopefully
we
did.
The
sound
issues
worked
out
later
so
Vanille
had
a
question
from
VMware.
What
sort
of
improvements
are
being
planned
in
the
mixer
adapter
space?
Do
you
want
to
elaborate
on
that
anyway?
Oh.
E
Yeah
sure
so
I
was
trying
to
out
of
Cree.
So
it
feels
like
the
processes.
I
mean
kind
of
there,
but
it
still
needs
a
little
improvement.
Like
I
was
trying
to
generate
the
config
like
the
proto
files,
and
it
was
giving
me
some
errors
because
there
are
some
complications
there.
So
that's.
Why
I
had
this
question
as
to
what
sort
of
improvements
are
going
to
be
there
like?
Is
there
anything
being
planned
in
this
area
or
and
how
can
we
help.
A
So
just
you're
you're
trying
to
build
an
adapter
is
that
right,
yeah,
okay,
I,
would
suggest
I,
don't
know
if
I
don't
know.
If
dancer
really
has
any
feedback
here
just
looking
through
who
all
joining
us
on
the
call
Dan
is
trying
to
answer,
but
you
can't
hear
me
know
so:
Dan
I'll
have
an
answer
for
you,
but
I
would
also
suggest
the
working
groups.
A
E
A
So
I
would
definitely
suggest
checking
in
with
the
policies
and
telemetry
working
group
and
then
Doug
and
Mandar
are
the
two
working
group
leads
there
and
they
would
definitely.
You
know,
want
that
feedback
regarding
your
experience
and
how
it's
been
and
challenges
you've
run
into
you
because
I
know,
that's
something
we
want
to
improve.
I,
see
dancer
really
now,
but
I
don't
hear
him
yet
said.
A
E
E
D
So
we
were
just
talking
about
things
to
work
on
next
and
one
of
them.
We
really
wanted
to
get
the
out
of
process
adapter
model
in
for
one
dot,
oh,
and
we
did
just
just
in
the
nick
of
time.
However,
we
I
think
the
way
it
was
written
on.
Our
to-do
list
yesterday
was
finish
out
of
process
or
adapter.
I
think
that's
the
way
it
was
phrased.
D
But
yes,
the
team
knows
that
there
are
that
there
are
issues
there
and
and-
and
we
do
want
to
in
you
know-
as
an
immediate
task
may
finish,
making
this
this
process
a
easy
and
be
well
documented.
So
I
think
that
that
everything
that
April
was
saying
before
I
got
my
microphone
work
was
working
is
right.
You
should
either
we
can
either
do
this
on
the
email
list
or
or
via
github
issue,
but
make
sure
that
that
team
is
aware.
D
D
For
anyone
to
write
an
adapter,
we
want
those
adapters
to
be
out
of
process,
so
it's
easy
for
people
to
start
using
them
right.
The
beauty
of
that
process.
Adaptors
is
you
don't
have
to
take
down
your
system
right,
someone
installs
it
and
then
they
put
some
configured
in
it.
So
it's
working
so
one
of
the
to
either
email
to
sto
April.
You
tell
me:
do
you
think
it's
do
Deborah's
to
your
users
to
that?
It's
your
dev!
For
that
also.
A
D
Is
a
great
idea
actually
but
yeah,
let's
make
sure
they're
where
the
issues,
because
they
they
do
know
there-
are
some
issues
I
want
to
make
sure
that
they
know
about
all
the
issues.
Someone
who's
actually
trying
it
can
I.
Ask
you
what
kind
of
adapter
you're
writing.
You
know
it's
a
metric
adapter
for
what
system
so.
D
E
D
A
A
G
G
Actually
before
I
talked
about
the
question
I
had
posted
here,
I
had
another
one
that
was
sparked
by
the
last
part,
so
I
met,
light
step
by
step
is
root,
cause
analysis
platform
big
in
the
distributed
tracing
space,
and
we
are
very
early
embarking
on
trying
to
do
some
contributions
to
sto
and
we're
planning
to
do
this
in
a
iterative
manner.
And
so
the
first
thing
that
we're
actually
hoping
to
do,
which
is
why
I
was
trying
to
get
my
desk
set
up
working,
is
that
my
step
is
directly
integrated
with
ongoing.
G
There's
like
specific
built-in,
like
step
configuration
options
in
envoy,
and
so
one
of
the
things
that
we
were
hoping
to
be
able
to
do
is
to
expose
the
like
to
configurations.
We
need
so
that
someone
using
sto
could
just
specify
those
and
then
have
the
envoys
actually
directly
communicate
with
essentially
our
products,
and
so
that's
kind
of
the
first
thing
that
we're
hoping
to
do
and
then,
as
you
might
know,
we
might
not
know
like
step-
is
also
a
big
proponent
of
open
tracing.
G
I
know,
there's
currently
a
effort
to
integrate
open
census
with
SEO,
which
is
awesome,
and
so
the
subsequent
iteration
of
what
we're
trying
to
do
is
actually
build
an
open
tracing
integration
as
well.
So
our
work
into
our
exploration
to
that
topic
is
still
early,
but
I
guess
now
having
Dan
here
is
great
and
I
guess.
Maybe
it's
something
I
could
bring
up.
I
guess
with
the
policy
and
telemetry
working
group
as
well,
but
I
guess
I
would
be
interested
in
understanding.
D
D
So
so,
first
of
all,
yes,
that
that
is
the
same
working
group,
the
the
policy
in
telemetry
working
group,
you
know
likelihood
a
mixer
adapter
is
the
right
way
to
do
this.
They
are.
They
have
thought
about
some
other
mechanisms,
but
but
mixer
is
the
place
to
start
and
if
it's
for
some
reason
isn't
that
would
still
be
the
right
working
group
meeting.
So
that's
definitely
a
place
to
go
and
thank
you
for
the
background.
B
D
So
so
maybe
this
is
just
turning
into
a
policy
and
telemetry
working
group
fashion,
but
but
that
the
thing
that's
good
about
those
working
group
sessions
is
the
the
people
there
or
where
you
can
ask
pretty
detailed
technical
questions
because
it
is,
it
is
literally.
The
each
working
group
is
run
by
a
maintainer
on
that
that
area
of
the
code
base
and
so
they're
the
people
who
know
this
the
best
and
there's
other
people
who
are
trying
to
do
the
same
thing.
You're
kind
of
do
right
and
then.
F
D
G
How
are
you
going
about
testing
it
and
also
type
of
stuff,
mostly
because
I
was
running,
I
found
I
was
running
on
my
machine
is
the
Mac
and
I
was
having
some
trouble
with
just
some
of
the
default
commands
that
are
listed
in
the
dev
guide.
I
ended
up
scouring
github
issues
and
found
things
that
I
needed
to
change
so
I'm,
hoping
to
actually
make
those
like
updates
to
the
dead
pad,
but
just
more
generally
like
what
does
the
flow
will
look
like
for
people.
F
A
little
bit
better
I'm
on
the
road,
yeah
I
went
on
and
I
just
used,
the
helm
chart
to
install
it
with
a
mutual
TLS
authentication
and
it
seems
to
be
kind
of
the
easiest
way.
I
think
on
last
call.
We
were
saying
that
there
are
some
issues
in
terms
of
the
configuration
setup
for
mini
cubes,
so
I,
just
when
I'm
working
with
you,
I,
try
and
abstract
away
from
the
kubernetes
details
that
way.
F
I
just
have
the
latest
version
of
kubernetes
and
I
kind
of
if
I'm,
working
based,
UI,
abstract
away
from
trying
to
resolve
the
kubernetes
items
with,
let's
say,
I'm
a
local
deployment,
so
that
that
helps
me
just
resolve
some
of
those
items
and
then
I
focus
on
really
getting
the
history
of
things
to
work.
And
then
the
main
thing
for
me
is
really
recently
has
been
getting
the
Gateway
items
to
work.
F
So
I
think,
if
you're
trying
to
ingress
things
from
an
external
service
into
Sto
having
it
in
some
kind
of
replica
of
your
development
environment
seems
to
be
the
best
approach
for
me,
because
then
I
don't
have
to
take
the
local
step
and
then
they're
like
the
pre-development
step.
I
can
just
be
like
okay.
This
is
what
it's
going
to
look
like
in
my
develop
environment.
So
then
I'm
just
kind
of
actually
trying
to
figure
all
the
configurations
as
close
as
possible
to
a
deployment.
F
F
So
when
I
go
back
in
the
cloud
shell
and
was
a
new
session
and
my
environment
variables
would
not
be
set
so
I
think
small
things
like
that,
just
making
sure
that's
set
and
making
sure
like
the
ACO
CDL
is,
is
in
your
bash
profile
and
stuff.
So
those
things
can
lead
to
minor
errors
which
then
you're
wondering
like.
Why
is
this
not
working?
F
F
H
D
And
I
see
on
chat,
Christopher
said
Christopher
uses
IBM
IKS
to
test,
and,
and
so
it
isn't
surprising
but
yeah,
it
seems
that
the
best
way
to
do
that
Christopher
says
mini
cube,
works
for
some
small
things,
but
drain
the
battery
very
quickly.
I
assume
that
this
will
probably
be
the
best
is,
is,
if
you
can,
if
you
have
access
to
an
ability
to
spin
up
a
cluster
on
an
IBM
or
or
Google,
that's
probably
going
to
be
the
easiest
way
to
do.
It
would
love
to
see
other
cloud
providers.
D
I
Like
you
know
what
IP
range
is
for
for
egress,
if
you
wanted
to
to
whitelist
external
so
that
only
proxied
internal
connections
and
I
had
to
reduce
the
pilot
memory
allocation
because
it
wanted
all
two
gates
that
doctor
had
default
and
I
I
should
probably
file
a
documentation
ticket,
but
I
want
to
get
some
other
developers
going
through
my
internal
box
before
I.
Do.
That
is.
I
F
D
I
D
I
A
K
A
Awesome
so
Julian
I
would
say.
Hopefully
some
of
this
was
helpful
and
then,
as
we
talked
about
the
other
questions
you
had
in
just
the
overall
development
process,
we've
got
a
link
to
that
thread
in
the
working
doc
and
I
would
say:
let's,
let's
start
adding
to
that
discussion,
and
then
that
can
also
form
a
good
framework
for
for
future
doc
enhancements
as
well.
So.
A
D
M
L
So
see,
I
I
think
I.
Think
that's
well.
First
of
all,
it's
awesome
that
that
you're
using
it
and
I'm
sorry
that
you're
having
trouble,
but
for
this
we
we
would
absolutely
would
like
to
know
and
help
you
get
that
out
and
then
certainly
improve
documentation
and
process.
So
please
do
send
like
an
email
on
the
on
the.
However,
we
have
been
communicating
with
tan
here,
but
I've
yeah.
We
would
certainly
want
to
help
yeah.
E
A
G
I
A
Well,
we're
glad
to
have
you
here,
we're
very
excited
and
then
I'm
super
excited
that
one
of
the
things
that
I
have
been
saying
in
a
you
know,
conference
talks
and
there
was
the
twitch
stream
last
week
was
that
I
was
waiting
for
that
one
point:
oh,
and
once
we
hit
that
I
knew
we
were
gonna,
get
more
folks
in
the
community.
That
would
start
building
mixer
adapters
and
you
know
start
just
playing
with
this-
do
overall
and
and
doing
blog
posts,
and
things
like
that,
so
it's
really
exciting
out.
A
D
By
the
way,
one
thing
that
we've
been
talking
about
is
adding
kind
of
community
pages
to
the
site
for
users
of
SEO,
as
well
as
for
anybody,
who's
done,
integrations,
mixer,
adapters
or
any
sort
of
integration,
and
what
we'd
like
to
do
is
put
up
pages
and
other
open
source
projects.
Do
that's
where
there's
logos
and
we
say,
hey
your
if
you're
using
it.
You
know-
and
you
want
to
tell
the
world
that
we'd
love
it
and
we'll
offer
you
a
little
space
on
a
page.
So
I
don't
know,
we
haven't
worked
out.
D
What
the
best
way
to
allow
that
to
happen
is
we'll
probably
put
up
a
page
with
some
logos
except
PRS,
but
we'll
probably
also
do
PR
for
somebody
if
they
just
send
us
the
the
logo
and
their
tagline.
So
if
that's
interesting
to
you
I,
don't
know,
maybe
I'll
start
a
thread
on
the
SEO
users
group
and
you
can
you
can
respond
if
you
think
you'd
like
to
have
your
your
logo
up
there.
A
Awesome
someone
I
can't
see
who
it
is
because
it
just
has
a
combination
of
letters
and
numbers,
but
someone
said
the
sto
twitch
was
awesome
and
it
was
awesome.
If
you
got
a
chance
to
watch
it
live.
Hopefully
you
enjoyed
it
as
well.
It
was
recorded
so
I
actually
have
an
action
item
for
myself
to
connect
up
with
Spencer
from
IBM,
who
let
everything
and
see
if
we
can
maybe
break
it
into
smaller
chunks
and
get
it
on
our
youtube
channel,
so
keep
an
eye
out
for
that.
A
A
C
A
We
love
each
other,
it's
hard
to
tell
so
Steve.
Also
excellent
tie-in
Steve
a
store.
The
Oz
con
sto
de
video
is
available.
So
yes,
but
it's
a
little
tricky
to
get
them
they
don't
they
put
them
in
their
Safari
online.
They
don't
like
put
them
up
on
YouTube,
but
they
give
us
the
speaker's
the
opportunity
to
download
and
put
them
on
YouTube,
so
I'm
trying
to
grab
everyone's
and
I
want
to
upload
them
to
the
co
YouTube
and
then
do
a
blog
post,
pointing
out
to
everything.
A
F
I'd
found
an
issue
regarding
just
how
to
use
this,
to
to
connect
to
create
a
gr,
PC
gateway
and
I
got
some
great
help
on
the
issue,
and
it
was.
The
resolution
was
I
kind
of
pointed
it
out
there
in
the
yellow
section.
It
was
just
about
going
into
the
value
section
of
the
llamó
and
just
creating
that
that
that
that's
that
porch
and
then
naming
it
HTTP,
which
which
then
creates
that
inside
the
default
gateway
the
the
default
is
your
ingress
gateway.
So
you
don't
even
have
to
create
the
custom
gateway.
F
So
it
was
good
to
have
this
resolved
and
then
I
pointed
out
suddenly
issues
that
are
still
kind
of
lingering
or
open
there,
but
they
were
dealing
with
the
same
thing
so
and
then
one
of
them,
seven,
nine,
oh
nine,
that
one
suggests
having
a
also
setting
that
the
node
port,
so
I
think
it
would
be
interesting,
just
maybe
throw
in
a
blog
post,
something
or
just
have
like
a
consolidator
way
to
write
up.
This
is
how
to
just
modify
the
basic
issue
of
ingress
gateway
to
use
G
RPC.
A
Is
something
that
a
lot
of
folks
have
had
questions
about
and
interest
about
and
I
would
love
to
get
some
documentation
around
that,
whether
it's
blog
post
that
leads
to
official
Docs
or
whatever
so
I,
don't
know
if
that's
something
that
you
want
to
do
often
or
if
you
want
to
open
it
up
for
other
folks
to
do
but
happy
to
help,
and
maybe
Riggs
could
help
us
as
well,
but
I
think
that's
an
important
thing.
We
need
to
get
some
documentation
on
sounds.
F
J
H
You
so
I
wanted
to
ask.
I
just
went
through
the
tutorial:
that's
on
your
documentation,
page
and
I,
deployed
sto
to
gke,
using
the
home
templates
and
I.
Have
it
automatically
deploying
the
sidecars,
so
I
deployed
the
book
binder,
app
and
I'm
noticing
that
the
sidecar
image
that's
being
used
is
sto
proxy
v2
I,
assume
that
comes
from
the
sto
proxy
github
repository
and
I'm
wondering
how
that's
related
to
the
Envoy
proxy
if
they're
related
at
all?
L
Yeah
yeah,
so
so
as
part
of
our
build
process,
we
add
a
few
filters
which
are
it's
just
specific
to
envoy,
and
then
we
create
a
build
of
envoy,
and
then
we
package
it
as
if
you're
/
proxy.
Be
that,
like
that's
the
name,
so
it
is
on
boy
with
some
extra
filters
that
is
already
running
now.
If
you
have
a
need
to
use
a
specific
middle
of
envoy
that
can
be
done
it,
it
will
be
a
little
bit
involved.
Essentially
what
it
would
mean
is
you
would
have
to
have
your
own
boy.
L
Add
filters,
then
package
it
up
again
and
use
that
image
in
terms
of
actually
using
an
alternate
image
that
that's
that's
somewhat
easy
right.
So,
for
example,
we
publish
multiple
images.
There
is
approximately
two
proxy
we
to
debug.
You
may
want
to
use
a
later
or
earlier
image
and
all
that
is
handled
through
the
site
garbage.
L
H
L
Yes,
so
you
can,
you
can
just
go
into
the
sto,
it's
a
it's
a
basil
build!
So
are
you
so
yeah,
so
you
if
you're,
not
you'll,
have
to
get
familiar
with
basil
a
little
bit?
Yes,
some
people
have
said
it's
non-trivial
to
get
familiarised
what
yeah
so
just
go
into
the
history
of
proxy
depository
and
there
is
a
top-level
make
file
even
and
then
eventually
there
is
a
basil
build
okay,.
L
That
actually
gets
the
so
middle.
Build
is
set
up
in
such
a
way
that
the
actual
Envoy
itself
is
a
dependency
of
that
field,
and
then
the
code
inside
sto
proxy
are
the
additional
filters
that
we
add,
and
you
can
click
and
look
at
more
details,
that
kind
of
starting
with
the
workspace
and
then
the
billa
file.
There
awesome.
A
D
Mike
Mike
asked
about
scalability
guides
and
and
Mike.
We
are
right
now
working
on
some
publishing
performance
numbers
that
include
expected,
expected
kind
of
scalability
guidelines,
hat
like
how
many
requests
per
second
and
ingress
should
be
able
to
handle
I,
don't
think
those
are
quite
ready
for
publishing,
but
we're
working
on
them
as
we
speak,
that's
great
news.
Thank
you.
F
Comments
actually
to
Jerry's.
One
of
my
kind
of
errors
early
on
was
trying
to
also
use
envoi
inside
a
sto,
and
then
I
found
out
that
there
was
kind
of
some
like
memory
collision.
That
was
occurring
as
a
result
of
that
and
then
I
just
realized
that
you
kind
of
probably
don't
want
to
use
and
way
in
there,
because
it
it
I,
put
so
much
a
while
ago
that
it
has
some
kind
of
root.
D
F
F
D
One
of
the
one
of
the
links
that
I
just
posted
into
the
chat
and
we'll
get
this
in
the
in
the
documents
as
well
was
a
really
good
description
of
the
networking
concepts
Sebastian
there.
It
looks
like
a
good
blog
post
and
a
video
that
accompany
it
rigs
actually,
who's
on
this
call
is
also
working
on
a
kind
of
conceptual
ducks.
D
You
know
improvements,
but
I
think
that
the
new
network
API
networking
api's,
are
complicated,
but
we
have
some
good
Doc's
to
explain
the
concepts
rather
than
just
reference.
Doc's
on
the
individual
types
will
help
so
I
hope
that'll
be
coming
pretty
soon
and
in
the
meantime,
that
guy
wrote
a
really
good
explanation.
F
Or
in
terms
of
the
other
scalability
god
I
just
posted
something
in
the
chat
there.
So
in
general,
when
people
are
deploying
with
helm,
should
we
be
resetting
those
those
so
to
kind
of
specify
those
to
our
containers
relative
to
what
you're
talking
about
with
the
stats
and
and
and
the
reference
points
coming
up.
A
F
Guess
I
guess
the
answer
is
kind
of
like
because
when
you're
deploying
there
are
some
very,
very
minor
limitations
there
like.
If
you
look
at
here's
another
line,
for
example,
just
if
you're
doing
a
just
a
basic
deployment-
and
you
just
run
the
helm
install
you
are,
you
are
creating
some
container
request
limitations
there.
So
I
was
just
looking
during
it's
kind
of
something
to
maybe
go
through
and
and
look
at
if,
if
you'd
like
to
customize
it
because
throughout
the
docs,
there
are
resource
limitations
that
are
tied
into
the
animal
files.
L
Typically,
when
we
use
for
performance
testing
we
allocate
so
to
just
as
like
as
an
example,
we
allocate
several
entire
CPUs
to
pilot,
so
three,
four
or
five
all
the
way
up
to
seven
and
and
then
at
that
scale.
Similarly,
would
at
least
allocate
100
milli,
CPUs
or
at
cart
proxy,
and
then
let
it
go
from
there.
L
So
this
so
some
of
this
stuff
is
actually
covered
in
the
performance,
Oh,
wiki
or
or
the
performance
section
of
our
Docs.
And
then
there
are
several
values:
files
that
are
that
are
available,
that
that
document,
how
we
do
performance
testing,
so
those
can
also
be
used
as
a
guideline,
and
then
we
are
coming
out
with
more
specific
numbers
on
several
different
axes.
So
if
you
have
a
thousand
sidecar
sneer
deployment,
how
many
pilots
would
you
need
to
hit
all
the
all
the
propagation
performance
metrics?
F
L
A
N
Obviously,
performance
is
one
thing
and
the
resources
that
we
allocate,
but
do
we
need
to
kind
of
you,
know,
customize
the
steel
deployment
specifically
for
making
sure
that
all
of
the
control
plane
components
are
highly
available
and
I
guess.
Another
question
in
the
back
of
my
mind,
is
which
of
the
components
are
you
know
the
most
critical,
which
components
can
we
essentially
lose
and
in
production
without
impacting
the
services
service
communication
within
the
class
right.
L
L
All
of
them
have
different
types
of
penalties
when
kind
of
in
terms
of
what
happens
when
they
go
down.
So,
for
example,
if
pilot
goes
down
for
a
brief
period,
nothing
actually
happens.
Everything
continues
to
work,
because
envoys
will
continue
to
use
the
configuration
that
they
have
and
they
will
search
traffic.
However,
if
pilot
goes
down,
then,
though,
new
sidecars
will
be
able
to
come
up,
which
means
that
you
cannot
bring
any
any
new
resources
on
mine
until
violet
is
restored,
which
means
at
just
a
normal
horizontal
pod
autoscaler
and
having
been
replicas
is
sufficient.
L
Eyelet
up
and
overall
availability
up
into
the
other
component
is
mixer.
So
if
you're
using
policies
and
you're
using
a
and
of
a
reject
on
failure
policy-
and
if
mixer
goes
down,
then
you
can
actually
have
outage
it
in
that
case
as
well,
just
a
normal
horizonte
pod
autoscaler
works
just
fine,
because
mixer
is
a
stateless
component.
L
D
Me
elaborate
a
little
bit
on
there.
If,
if
you
were
because
minor
I
said
this,
but
I
want
to
make
sure
it's
explicit,
he
said
if
you're
using
mixer
for
policy
and
if
you
have
it
in
essentially
a
fail,
closed
configuration,
then
mixer
going
down
would
result
in
requests
getting
rejected,
but
but
those
are
two
kind
of
big
IFS.
Some
people
use
mixi,
mixer
purely
for
telemetry
collection
and
they're,
not
having
mixer
enforced
policy,
in
which
case
mixer
going
down
would
mean
that
telemetry
isn't
getting
reported.
You
might
lose.
D
You
know
some
some
some
period
of
time.
In
your
say,
your
your
logs
or
in
your
telemetry.
However,
those
requests
would
be
served
and
the
second
is
even
if
you
are
using
policy
and
policy,
it
could
be
an
integration
with
the
back
end
policy
engine
or
something
like
quota.
Those
can
be
configured
in
either
fail
open
or
fail
closed
fashion
and,
depending
on
your
policy,
you,
you
might
choose
to
have
them.
You
know
here
at
Google
we
have,
depending
on
the
backend
policy
system,
we
decide.
F
N
L
Alright,
so
sidecar
injector,
going
down
would
also
similarly
prevent
new
things
from
coming
up,
which,
which
means
that
again,
normal
just
a
min
replicas
setting
on
the
deployment
is
fine.
If
it
goes
on,
it
could
have
been
going
back
up
and
that
I
children
matter
in
terms
of
ingress
gateways
and
egress
gateways.
L
They
can
also
be
handed
over
to
a
deployment
which
they
are
just
just
by
default,
that
that
that's,
how
we
deploy
them
and
again
a
simple
wind
replicas
was
a
horizontal
or
not
a
scalar.
There
also
works
just
fine,
because
even
those
components
are
stateless,
they
should
just
respond
to
load,
and
a
few
of
them
should
always
be
available
and
Citadel
is
so.
If
Citadel
is
not
not
available,
then
you
are
again
going
to
have
this
kind
of
boot
strapping
issue,
so
new
things
cannot
come
up.
Existing
things
continue
to
work.
D
For
and
that,
just
to
be
clear
there
two
things
first
Citadel
only
affects
you
if
you're
using
MPLS.
However,
there's
one
caveat
there
if
Citadel's
gone
for
a
long
time,
certificates
will
expire
and
so
I
think,
typically,
the
I
think
our
default
configurations
certificates
have
a
life
of
one
hour
and
get
rotated.
So
if
your
Citadel
is
is
down
for
a
length
of
time,
you
know
say
it's
down
for
30
minutes
or
60
within
30
minutes
after
your
certificates
would
expire
and
those
pods
which
were
running
would
not
be
able
to
obtain
new
certificates.
D
A
We
got
a
couple
well,
we've
got
two
more
questions
in
the
chat.
We
only
have
about
eight
minutes
left
in
the
call,
so
I
want
to
try
to
get
through
them
as
much
as
we
can
Amy
had
a
question
about.
Is
there
a
manual
or
recommended
steps
for
adding
a
new
micro
service
into
an
sto
enabled
service
match
I.
D
Think
that
the
the,
if
you
are
not
using
em
TLS,
adding
a
new
service
is,
is
easy
and
by
easy
I
mean,
if
you're
using
a
proxy,
auto
injection.
Then
when
someone
deploys
into
that
into
that
cluster
kubernetes
will
make
sure
that
the
a
the
the
proxy
gets
installed
as
a
sidecar
and
B
the
service
gets
registered
so
that
other
people
can
call
other
other
services
on
the
mesh
can
call
it.
So
in
that
case,
that's
I'll,
say
a
trivial.
If
you
are
using
MPLS,
then
I
would
investigate.
D
I
would
encourage
you
to
look
at
what
we
call
incremental
MPLS
incremental
MPLS
allows
you
to
put
a
service
into
permissive
mode
so
that
it
can
accept
calls
either
via
MPLS
or
without
em,
TLS
and,
and
that
will
allow
you
to
have
that
new
service
come
in
and
and
even
if
it
doesn't
have
a
sidecar
that
would
allow
it
to
communicate
via
MPLS.
It
would
allow
it
to
send
traffic
to
to
another
service.
D
O
Yeah
I
think
I
was
looking
for
something
in
the
line
of
like
a
checklist
right.
There
are
a
lot
of
questions
on
rocket
regarding
my
birth
virtual
service
is
not
working.
I
can't
reach
my
gateway,
so
it
would
be
kind
of
nice
to
see
a
checklist
of
how
to
like
SEO
101
I
mean
with
with
the
risk
of
like
me,
volunteering,
to
write
you.
You
just
said
the
docs
are
open-source,
but.
O
Yes,
III
think
it'll
be
very
useful
to
have
that
kind
of
documentation
to
do
a
checklist
so
that
people
can
go
back
and
troubleshoot
their
settings.
Yeah
I
understand.
What's
a
sidecar
enabled
yeah,
you
can
just
do
it
automatically.
That's
why
I
automatic
psych
heart
works,
but
yeah
yeah.
So
thank
you,
yeah.
That
really.
D
D
D
D
D
D
D
Know
you
know
what
would
be
it's
like:
yeah
I
think
what
would
be
good
actually-
and
this
is
not
gonna
happen
quickly-
is
a
page
on
local
debugging
and
just
kind
of
recommend
like
here's,
how
you
set
it
up.
Here's
things
to
watch
out
for
I.
Don't
think
that
will
happen
anytime
soon,
Hector,
but
but
I
can
I
can
see
from
from
several
the
questions
on
this
that
that
is
something
that
a
lot
of
people
are
doing
and
yeah
great
thing
for
the
community
to
to
chip
in
on.
O
D
K
O
Yeah
I
found
some
issues
with,
like
the
latest
Hjelm
chart,
like
part
of
the
helm
option
default
value
for
ingress
is
for
ingress,
is
disabled,
I
think
because
it's
deprecated,
however,
the
ingress
the
I
forgot.
The
term
is
not
it's
actually
referring
to
the
point
eight,
whereas
it
is
the
deprecated
as
so
part
of
it
is
trying
to
support
point
eight,
but
it's
that's
actually
not
you
know
so.
I
found
it
kind
of
confusing.
Someone
corrected
me
on
rocket
chat,
but
yeah
I
found
some
issues.
O
A
We
got
a
chance
to
at
least
tell
people
what,
if
we,
if
we
don't
have
the
answer
where
to
go,
get
it
yeah.
We
we
are
at
time.
I
know
that
there
are
some
comments
about
people
submitted
stuff,
john-boy
Khan,
which
is
awesome.
Please
do
submit
more
things
again.
It
closes
tomorrow
and
if
we
weren't
able
to
get
to
your
questions
or
anything
today,
please
do
join
us.