►
From YouTube: Istio Community Meeting 8-9-18
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
Can
everybody
still
hear
me?
Okay,
awesome!
Thank
you.
Cool
all
right,
well,
I
put
a
link
to
the
working
doc
and
the
chat.
If
you
have
questions
or
want
to
add
anything
to
the
agenda,
I'm
gonna
be
obnoxious
and
promote
the
call
for
presentations
that
are
coming
up.
There's
two
that
I
think
are,
of
course,
on
the
big
radar
for
anyone
with
SEO.
The
first
is
the
cube:
can
flash
cloud
native
con
and
it
closes
this
someday
I
know
that
means
that
pretty
much
everyone's
going
to
wait
until
Saturday
to
start
thinking
about
it.
A
There's
an
Events
Channel
and
that's
a
good
spot
to
you
know,
try
to
recruit
a
co
speaker
or
anything
else
like
that.
There
will
be,
of
course,
a
lot
of
competition
around
the
talks
for
coop
con.
So
don't
let
that
discourage
you
from
submitting.
Even
if
you
you
know
end
up
with
a
talk
that
doesn't
get
accepted.
Cfps
are
always
a
really
great
way
to
kind
of
kick
off
like
what
would
be
a
good
plug
post
and
things
like
that.
A
So
I
definitely
recommend
going
through
the
process
and
then
there
is
also
envoy
con,
which
is
happening
the
day
before
coop
con.
So
it's
you
know
pretty
much
same
week,
same
location,
I,
think
it's
the
same
location,
there's
a
whole
lot
going
on
the
day
before,
but
it's
one
of
the
co-located
events
that
is
happening
with
CNCs
and
on
week
on,
of
course,
I
think
also
has
a
lot
of
opportunity
for
folks
in
the
SEO
community,
since
we
know
and
know
and
love
envoy.
A
So
I
wanted
to
call
that
out
as
well
see
if
paper
that
one
closes
on
August
24.
So
you
have
a
little
bit
more
time.
But
perhaps
you
have
a
talk
that
you
know
you
you
want
to
propose
that
you
think
would
be
a
good
fit
for
both
submit
it
to
both.
So
just
wanted
to
call
that
out
that
that
is
happening
and
let's
see
a
Spencer
on
the
line.
No.
C
D
D
To
celebrate
the
1.0
release,
we
wanted
to
do
a
twist
stream,
that's
something
we've
been
doing
for
a
while.
Now
it's
making
these
kind
of
all-day
twitch
streams,
it's
sort
of
like
what
they
do
when
they
launch
a
video
game
these
days.
Is
they
just
have
a
bunch
of
experts
over
for
about
a
day?
So
I
wanted
to
drum
up
some
interest
in
that
I'm
always
looking,
we
have
about
a
half
full
schedule
right
now.
D
Maybe
three-quarters
full
schedule
right
now,
so
if
you
would
like
to
be
remote
or
in
San
Francisco
on
August,
17th
and
kibbutz
about
sto
or
show
a
ten
minute
demo,
or
something
like
that,
we're
more
than
happy
to
make
that
happen.
For
those
of
you
don't
know,
twitch
is
an
online
streaming
platform.
It's
like
the
YouTube
of
live
streaming,
it's
very
much
tilted
towards
gamers,
but
there's
a
lot
of
developer
activity
on
there
too,
so
like
AWS,
is
a
pretty
prolific.
D
So
there's
some
questions
like
what
does
that
look
like
this
is
kind
of
what
it
looked
like
at
docker
con.
We
had
a
couple
people
I,
think
some
of
you
know
Dan
Berg
from
IBM,
and
then
the
the
graphic
designers
made
us
this
nice
sto
overlay.
So
we
kind
of
put
your
Twitter
handle
here
and
whatever
demo
is
going
on
some
videos
down
here.
We
have
another
version
of
this
for
like
if
you
are
on
a
couch
or
something
just
chatting,
and
we
got
a
nice
piece
of
real
estate
down
there
in
San
Francisco.
D
D
You
know
some
of
the
other
stuff
in
that
we've
got
the
data
die
people
coming
on
and
got
some
Google
people
coming
on,
gets
that
coming
on,
so
always
looking
for
more
people
and
more
interest,
we
don't
have
balloons
yet,
but
we
do
have
a
cake
and
I
guess.
The
last
thing
I'd
say
is
like
we
have
a
lot
of
tech
experience.
D
E
D
There's
a
couple
things
to
call
out
here.
So
if
you
would
like
to
present,
you
have
to
coordinate
with
me,
there
isn't
an
official
invitation
for
that,
but
you
know
I'm
officially
inviting
you
to
reach
out
we'll
get
you
on
the
schedule.
If
you
just
want
to
watch,
that's
really
easy.
It's
just
on
twitch
TV!
If
you
would
like
a
calendar,
invite
or
working
that
out
I
can
send
you
a
link
to
a
calendar.
D
A
A
Spencer
Kevin
made
a
comment
that
Brian
Harrington
from
Red
Hat
is
interested
in
attending
as
well
and
he's
local
to
us
so
great
connect
all
those
dots
you.
A
A
G
H
G
So
yes,
I
want
to
click
them
of
how
we're
using
Sto
so
we're
watching
to
sell
them
and
we'll
do
machine
learning
deployment
for
enterprise,
so
just
want
to
give
a
few
slides
just
describing
what
we
do
and
how
we
integrate
with
this
Joe
and
then
I'll
do
a
little
demo
of
what
we're
doing
I'm
good,
to
get
some
feedback
of
how
we
could
use
this
show
some
more
so
yeah
so
yeah.
G
What
do
we
do
exactly
so,
if
you
view
this
is
the
stages
in
a
machine
learning
pipeline
from
beta
ingestion
data
transformation,
validation,
creating
models
then
pushing
them
out
and
serving
them.
What
we
do
is
a
seldom
is
the
last
stage,
so
we
have
open
source
projects,
oldham
core,
on
which
you
can
find
on
github,
where
we
manage
to
roll
out
survey
monitoring
of
machine
many
diplomas
on
kubernetes.
So
we
allow
you
to
build
your
models
using
any
toolkit.
Tensorflow
skycat
learn
our
java
based
models
and
we
just
managed
the
one-time
influence.
That
of
that.
G
Allow
you
to
do
it
up
powerful
graphs
as
a
service
graph,
and
may
you
deploy
a
scaler
so
part
of
it
was
a
wider
ecosystem
of
people
doing
that,
so
you
might
poet
up
other
projects
like
cute
flow,
so
a
part
of
that
is
where
you
can
choose
Selden
to
serve
models
and
also
have
a
push
like
IBM
a
fiddle
framework
for
deep
learning.
So
you
can
you
can
build
your
deep
learning
models
and
then
serve
them
one-celled
in
there.
So
you
have
a
part
of
these
projects.
G
It's
just
as
described
you
know
get
into
the
stuff
that
makes
of
how
we
fit
into
SEO.
So
in
terms
of
this
course,
Selden
graphs
that
we
deploy,
you
might
want
to
start
with
a
single
model,
a
machine
learning
model
which
you
deployed.
You
know
to
get
predictions
from
when
we
expose
an
API.
Do
you
have
to
see
West
for
that?
But
then
we
want
to
allow
people
to
build
up
more
complex
service
graphs.
G
Well
there
at
one
time,
so
you
might
want
to
turn
it
into
an
a/b
test,
with
two
machine
learning
models
and
test
that
out.
It's
only
into
something
more
complex,
like
a
multi-armed
bandit,
so
we
give
opportunities
to
do
that.
So
you
can
push
traffic
to
the
best,
serve
a
model
in
real
time
and
then
you
can
add
in
a
ver
components.
G
Each
of
these
components
will
be
separate
and
docket
images,
which
you
then
put
it
into
a
graph
and
run
in
several
deployments
and
pods
on
on
kubernetes,
and
so
you
can
add
in
that
extra
things
like
our
lie:
detection
and
explanation,
and
then
we
manage
that
whole.
The
question
response
flow
through
that
through
that
graph,
so
each
so
the
data
scientist
can
just
work
on
the
individual
parts
of
that
graph
and
then
you
can
join
together
like
sort
of
Lego
blocks,
to
create
everything.
G
So
you
describe
those
graphs
at
you
know
some
extension
of
connectors
at
some
custom
resource.
We
describe
the
graph
each
each
of
the
parts
of
how
you
made
them
build
up
the
actual
images
in
terms
of
you
know
and
that's
just
a
standard
pod
template
spec,
and
then
you
build
all
together
with
the
replicas,
and
you
can
have
a
separate
series
of
these
graphs
to
actually
deploy.
So
how
does
it
fit
in
with
this
job?
So
obviously
we
have
the
graph
there
at
the
top
that
I
described.
G
So
you
can
then
and
say
add
different
versions
of
models
have
prepared
your
deployment
menus
sto
to
manage
the
traffic
into
those
different
versions
of
each
part,
but
have
our
own
service
orchestration.
Managing
the
whole
will
question
responsible,
oh
I'm,
so
yes
I'm
going
to
give
a
little
little
demo
of
this
just
in
terms
of
a
simple
canary,
but
hopefully
you
can
see
from
the
previous
stuff.
You
could
have
much
more
complex
graphs
and
manage
traffic
between
all
parts
of
small
parts
of
your
graph,
so
it's
going
to
use
the
classic
M.
G
This
example,
where
you're
trying
to
pacify
handwritten
digits
and
then
and
we'll,
have
a
sky
kittler
model
and
then
a
canary
rollout
of
a
new
model
which
is
tensorflow
model
and
then
see
which
is
best
and
choose
the
tensorflow
model
so
yeah.
So
this
notebooks
on
our
selves
encore
get
hubs
you
can,
you
can
run
for
it,
so
basically
I
mean
just
going
through
someone
go
all
of
it
also
want
to
go
quickly.
I'm
so
upset
that
list
you
and
I've
got
someone
for
the
installation
here.
G
Many
of
that
was
because
part
of
them
what
we've
done.
We
need
to
access
the
internet
to
give
em
this
digits
down.
So
I
had
to
just
add
in
some
custom
settings
there
for
the
include
IP
ranges,
so
we
can
get
out
to
the
Internet,
and
apart
from
that
is
this
data.
This
geo
1.0
install
that's
all
installed
and
then
you
can
install
and
seldom
which
you
can
install
by
helm
and
the
install
that.
So
that's
all!
G
It's
also
sending
the
feedback
on
whether
the
model
got
it
right
or
not.
So
then
we
can
go
to
the
sto
control
pain
and
we
see
here
so
got
about
40
requests
going
through
there
per
second
just
to
test
it
out
and
we've
got
our
skycam
learn
model
there
with
the
request
going
through,
and
we
also
have
our
self-service
Orchestrator,
which
is
not
so
needed
in
this
case
it
was
weird
when
it
got
a
single
model.
G
Obviously,
obviously,
when
more
complex
graphs
would
have
a
more
complex
thing,
so
people
people
can
see
that
and
it's
great
to
have
that
dashboard,
and
we
have
also
have
our
own
dashboard,
which
is
just
a
simple
one,
that
part
of
the
open
source
one
difference
and
what
I've
talked
about
here
is
you
can
see
the
actual
reward,
which
is
the
actual
accuracy.
This
is
using.
We
have
two
a
P,
I,
endpoints
one-foot
predicts
that
you
get
your
predictions
obviously,
but
you
also
have
a
like
have
feedback.
G
You
can
say:
okay,
the
model
got
this
right
or
the
model
got
this
wrong
and
you
can
send
that
back
with
the
graph,
which
means
you
can
do
interesting.
Things
like
with
multi
on
bandage
to
say:
okay,
which
model
is
actually
working
best,
will
get
accuracies
stats
like
you've
got
here.
So
in
this
case,
this
guy
kit
learns
running
about
90%,
so
accuracy
in
this
case.
G
So
going
back
to
the
notebook
selecting
this
assume,
our
data
scientists
equated
a
new
version,
so
we
have
initially
we
set
up
a
sto,
I'm
networking
Walter
to
make
sure
we
just
keep
pushing
in
his
gesture
version,
one
of
our
model
and
we've
labeled
it
in
version
one
in
our
llamo
Jason
there
before,
and
so
we
push
that
out
and
make
sure
that's
happening.
I've
done
that
already,
and
so
then
we
have
a
version,
two
of
our
model,
what
we
send
a
new
Seldon
deploy
custom
resource
where
we've
got
our
existing
skycat
loan.
G
I'll
just
send
this
off
actually
to
make
sure
we
couldn't
get
save
time
as
it
were,
and
then
in
this
part
of
that's,
we've
got
a
skycat
loan
model
and
then
I
they
decide
to
have
quite
a
tentative
lo
model,
hopefully
which
they
think
is
better
and
we're
going
to
set
that
up
in
a
power-law
graph
and
then
we'll
try
that
out
as
a
canary
test.
So
let's
just
check:
where
are
we
so
yeah?
G
So
here
the
models
being
created
and
so
I
think
they
just
think
so,
each
of
those
pods
it's
got
obviously
the
sto
and
voi
there,
and
then
we've
got
one
deployment
with
the
tensor
flow
em
list
and
one
the
point
with
our
service
Orchestrator
inside
it
also
running
there.
So
that's
running
so
now.
We've
got
a
graph
like
this
with
a
the
rest
and
point
for
the
sky,
Kaitlin
and
one
for
the
tensor
flow
model,
and
we
can
now
well.
Obviously
we
can
keep
checking
other
things.
G
Working
is
fine
and
we
can
set
up
then
the
canary
routines,
because
at
the
moment,
obviously,
what
we
should
have
is
all
the
traffic
still
going
to
just
the
just
the
sky.
Kid
learn
model,
so
we
know
traffic
for
the
tensor
flow
model.
So
we
can
check
on
the
mister
dashboard,
so
yeah
no
traffic
for
there
and
so
obviously
will
now
go
back
to
the
canary
routing
and
will
then
transfer
it.
So
we've
got
30
percent
traffic
we'll
send
that
to
the
we'll,
send
that
to
the
tensor
flow
model,
so
we
can
test
it
out.
G
So
hopefully
that
should
start
to
shoot
up
yeah
there
you
go
so
now
we're
sending
about
so
10
requests
per
second,
although
about
store
a
tensor
flow
model
and
obviously
we've
got
the
sky
kilo
model,
which
would
be
decreasing
a
bit
in
traffic,
and
this
so
the
same
thing
now
on
our
dashboard
in
Selden.
They
should
be
like
a
new
model
here,
which
is
just
starting
up,
so
we
can
measure
the
traffic
there
and
we
can
measure
the
accuracy
doing
a
simple
AV
test.
G
Obviously,
in
this
or
machine
learning,
world
be
able
to
have
much
more
complex
graphs
and
what
I
showed
you
here,
but
to
allow
people
to
build
up
these
complex
graphs
and
then
maybe
add
different
versions
of
different
parts
of
the
graph
and
then
put
traffic
and
very
you
know,
doing
fine-grained
way
to
different
parts
and
manage
their
there
and
how
they
actually
going
to
deploy
the
machine.
Learning
and
Inter
production
is
really
key
and
so
yeah
well,
we
would
think
it's
Joey,
definitely
part
of
that,
and
so
yeah
thanks.
That's
my
very
quick
demo.
G
Absolutely
it's:
it's
all
open
source
I'm
like
Apache.
You
can
find
it
on
github
under
I'm,
Seldon
core
and
we've
got
this
and
I'm
no
bug
I
showed
you
that,
hopefully,
we'll
have
more
complex
notebooks
with
this.
Do
you
know,
like
green
blue
deployments,
number
things
that
fit
in
and
the
more
times
I've
been
discussing.
A
A
A
I
Hey
hey
so
my
question
was
I,
got
sto,
1.0,
running
and
actually
I
guess.
I
could
comment
on
my
previous
question
from
last
week,
so
it
what
happened
was
I
I,
guess
I
hadn't
really
looked
in
the
best
practices,
because
I
was
running
envoi
inside
on
SEO,
instead
of
using
the
new
gateway
architecture.
So
once
once
I
switched
over
to
to
Gateway
instead
of
kind
of
duplicating
Android
proxy
inside
HD.
Over
again,
it
seemed
to
solve
most
of
my
problems
because
every
time
I
kind
of
try
and
duplicate
the
architecture.
I
It
runs
into
this
permissions
issue
with
with
with
the
root
and
the
namespace
thing,
so
that
that
simply
resolved
what
I
was
trying
to
figure
out
last
week.
But
then
this
week,
when
I
switch
over
to
using
the
the
Gateway
structure,
the
for
some
reason,
the
the
I'm
using
it
to
host
an
external
domain
like
example.com.
I
But
it's
it's
not
doing
the
TLS
redirect
for
the
domain.
So
some
people
go
to
like
example.com.
It
doesn't
redirect
them
over
to
HTTPS,
and
so
the
redirects
are
just
failing.
There
was
a
previous
github
issue
on
this
and
it's
just
kind
of
sitting
there,
so
I'm
wondering
what
the
status
is,
because
right
now
essentially
to
use,
is
do
if
you're
hosting
a
domain.
You
kind
of
can't
I
guess,
because
it's
people
don't
type
into
HTTP.
It
doesn't
upgrade
them.
I
I
E
I
I
F
F
A
A
F
C
J
F
A
A
Awesome
I
don't
know
if
anyone
noticed
but
github
did
release
a
new
verification
process
similar
to
like
Twitter
and
all
of
that
and
sto
is
now
officially
verified.
So
I
think
that's
a
good
as
we
get
bigger
in
the
community
grows.
Folks
know
for
sure
that
they're
at
the
right
spots
and
that
that
is
the
place
to
download
from
so
we
have
that
cool
new
little
badge.
A
K
Hyah
produces
Thiago
from
redhead
right.
We
we
are
we're.
Gonna,
have
the
death
of
the
first
stabbed
I've
confessed.
A
very
famous
conference
in
Europe
happens
in
in
bruno
czech
republic
every
year,
and
this
year
is
gonna,
be
the
first
time
in
Boston.
It's
gonna,
be
in
next
week
and
I'm
gonna
put
in
the
chat
the
link.
It's
it's.
A
free
conference
is
sponsored
by
right
here
and
we're
gonna
have
two
talks
on
Easter
there.
K
J
K
A
A
F
F
B
H
B
B
I
did
I
did
a
con
tutorial
about
setting
up,
sto
and
monitoring
it
about
a
month
ago,
or
so.
Some
folks
had
were
able
to
get
mini
cube
up
and
running.
Others
were
using
Google
cloud
platform,
we're
using
Amazon
overall
I
think
it
was
pretty
good,
but
like
getting
getting
the
base
environment
up
and
running
was
definitely
like.
I
told
people
hey
come
to
this
tutorial
with
this
running.
Nobody
did
so
it's
a
little
bit
challenging
thing,
but
a
lot
of
a
lot
of
folks
came
away
from
that.
Like
yeah,
this
is
cool
stuff,
yeah.
F
A
J
A
A
If
anyone
has,
you
know
once
once
feedback
or
help
kind
of
you
know
working
on
the
writing
or
anything
like
that.
You're
welcome
to
reach
out
I'm
I
am
still
behind
on
emails,
but
it's
getting
better
so
feel
free
to
send
me
an
email
and
I'll
be
keeping
an
eye
of
keeping
an
eye
on
it
over
the
weekend.
So
if
you
have
questions
or
just
want,
some
help
feel
free
to
reach
out
and
if
no
one
has
any
good
person.
I
Say
the
of
the
transition
to
one
point
that
was
really
smooth
good
yeah,
but
really
amazing.
Like
the
way
the
dog
switched
over
we're
like
using
that
this
idea,
with
the
preliminary
where
you
can
see
the
knob
coming,
one
is
now
like
I
I
guess
one
point
one
is
is
on
the
up
coming
in
one
point:
oh
it's!
It's
really
smooth
because
you
can
see
the
changes
and
you
can
kind
of
prep
and
just
switch
everything
over.
So
it's
a
very
smooth
setup
and
what
some
say.
A
Yeah,
thank
you
for
that
feedback
I'll,
be
sure
to
pass
that
along.
That's
Martin
has
been
leading
a
lot
of
that
effort
and
I
I.
Personally,
thank
you
know
for
for
a
project
that
still
is
relatively
young
as
it
is.
The
docs
I
mean
the
the
website.
Experience
is
great
for
it,
and
so
he's
done
a
fantastic
job
with
that
and
will
definitely
pass
that
along
yeah.
I
A
A
E
F
So
if
you
look
at
so
pre
P,
so
this
is
a
little
bit
easier
because
for
me,
the
ax
scrapes
right.
So
really
it's
just
a
matter
of
programming.
Your
current
Prometheus
instances
to
go
scrape
the
right
stuff
in
the
issue.
So
when
you
install
isseo,
there
is
a
Prometheus
config
map
that
gets
created
in
the
issue
of
system
namespace.
That
config
map
has
all
of
the
Prometheus
scraped
configuration
for
the
various
SEO
components.
So
you
should
just
be
able
to
add
that
to
your
existing
previous
configuration
and
and
get
metrics
and
in
those
servers.
E
Okay:
okay,
the
only
thing
is
this
I
guess.
The
only
thing
which
needs
to
be
changed
is
because
of
the
Prometheus
collects
all
of
this
from
the
sto
mixer
right,
so
I
guess
the
scraping
is
done
from
the
Prometheus
to
reach
out
to
the
mixer.
You
need
to
make
sure
that
the
mixer
is
exposed
externally
routable,
so
that
which
I
believe
is.
It
is
already
so
it's
then
I
may
have
to
do
some
kind
of
a
ingress
controller
or
a
gateway.
Now
it
is
changed.
So
yes,.
I
A
I
A
All
right
thanks
everyone
for
joining
today,
if
you've
got
anything
that
comes
up
before
our
next
call,
please
feel
free
to
reach
out
or
add
to
the
working
dock.
Think
of
cool
stuff
too.
It's
a
demo
for
us
in
two
weeks.
Let's
see
the
exciting
things.
Y'all
are
working
on,
thank
you
again
and
if
you
need
anything
just
reach
out,
but
otherwise
we'll
call
it
for
this
one.
Thank
you.
Thanks,
bye,
everybody.