►
From YouTube: Istio Networking WG meeting 2019-02-28
Description
- Status of 1.1
- Istio 1.1.x priorities
- Testing strategy
- Istio 1.2 priorities
A
Okay,
all
right,
hello,
everybody
welcome
to
this.
Your
networking
work
group
meeting.
We
have
a
new
series
of
invites
so
I
hope
everybody
knows
about
this.
We
had
the
issues
with
that,
but
now
everything
looks
good.
Today
we
have
a
special
special
topics
around
roadmaps
for
one
one
one
and
beyond
I
guess
it's
probably
worth
discussing
a
bit
the
status
of
one
zero,
since
we
have
Josh
here,
I'm
going
to
again.
B
C
Which
is
called
RC
wine
was
created
a
so
anyone
whose
home
is
testing.
Please
repoint
your
test
to
that.
We
still
have
a
list
of
key
zeros
that
we
think
block
the
release
and
I'm
not
you're,
trying
to
complete
a
bug
scrub
right
now
to
actually
start
guessing
when
we
can
actually
release
one
one
that
I
don't
know
right
right
now
at
the
moment.
D
E
E
A
Some,
like
non
user,
visible
features.
We
can
call
them
I,
guess
that
would
go
into
one
one
one
and
some
bugs
as
well
right.
So
how
can
we
go
about
this?
We
have
a
document.
We
have
a
spreadsheet
with
one
two
priorities:
shall
we
create
a
spreadsheet
width
when,
when
one
or
just
list
them
in
the
doc,
okay,
we
can
use
a
spreadsheet
later
right,
so
I,
actually
edit
I'm,
not
sharing.
A
E
Since
it's
clear
we
want
to
have
bugs
or
kind
of
critical
features.
Only
in
one
onyx,
we
should
be
able
to
have
maybe
bi-weekly
or
weekly
trains.
That
again
would
take
some
features
when
they
are
ready,
instead
of
kind
of
waiting
until
the.
If
you,
mr.
Takagi
things
that
we
are
busy
a
train
base
model
like
a
lot
of
other
people,
are
using
to
kind
of
diffuse
a
pizza
pressure
on
big
release,
isn't
getting
untested
stuffing
yeah.
E
A
F
A
In
general
yeah,
it's
so
that
the
end
point
ingestion
will
actually
enable
like
different,
like
multi
cluster
cases.
So
we
will
need
them
really
ship
that
this
teacher
right
notice
the
feature
right.
It
will
have
to
have
solid
testing.
Yes
around
multi
Casa,
which
I
think
we
don't
have
right.
Now.
We
have
like
tasks
we
have
so.
F
E
Vms
are
stable.
We
are
not
running
into
anything
right
now.
I
mean
it's,
there
is
just
support
for
SDS
came
up
today,
but
that's
it
mm-hmm.
Other
cluster
beans
for
coalition
I
worked
on
I'm
working
on
one,
of
course,
okay
MGK
adapter,
but
that
doesn't
have
to
be.
You
know
it's
more
or
less
independent
religious.
F
E
F
F
So
CP
into
third
party.
E
E
F
H
F
E
E
A
G
A
J
J
K
L
E
M
E
I
figure,
if
I
dare
not
speak
for
himself,
but
we
just
heard
do
another
report
of
so
the
probably
that
some
some
clusters
somehow
get
ipv6
addresses
for
whatever
reasons,
including
VMS,
and
that
case
problems
with
me,
not
necessarily
to
get
the
visas
for
you,
but
not
to
crash.
If
you
have
an
ipv6,
alright,
that's
it.
Okay,.
F
F
B
E
Is
that
bugs
and
after
we
disability
tables,
all
kind
of
bugs
afflict
into
the
readiness
check
doesn't
have
the
proper
column?
It's
not
property
formatting.
So
we
had
it
working
at
some
point,
then
kind
of
we
got
regression
different
places
and
fixing
it
before
wonderpana.
And
why
would
we
not
see
God
beep
wasn't
working
got
broken
again,
is
not
a
new
feature.
It's
something
with
work
before
I
guess.
L
E
A
A
E
B
White
white
books
will
not
work
because
the
redirection
to
the
pilot
doesn't
work
so
basically
I
mean
like
if
you
don't
eyelet
or
without
sorry,
without
a
sidecar
proxy.
You
have
ipv6
connectivity,
but
I
mean
what
easier
then
has
to
do
with
this.
So
as
soon
as
you
start
trying
to
redirect
the
ipv6
traffic
to
their
side
car
proxy,
you
start
seeing
problems
and
that's
what.
L
C
E
I
L
L
F
N
A
C
C
A
A
Q
One
thing
that
they
might
just
think
of
the
last
few
weeks
is
basically,
you
would
commit
recommendations
close
up.
Doesn't
my
user
to
like
people
are
excited,
so
yeah
I
thought
Mike's
here
pretty
good,
but
maybe
I'm
wrong.
So
so,
basically,
we
documentation
separate
from
from
the
main
repo,
which
is
just
basically
discourages
people.
You
know
if
keeping
dogs
along
with
their
PRS,
they
might
practice
that
usually
works.
Very
well.
Q
C
I
think
we
have
a
lot
of
I
think
in
this
working
group.
We
need
to
focus
on
the
never
features.
I
do
think
some
disclaimer
on
lines.
We
have
some
usability
and
testing
concerns
that
may
may
have
to
take
priority.
We
made
to
do
first
before
a
lot
of
these
things,
but
for
now
we
should
just
focus.
Thank
you
all
right.
So.
F
L
F
The
hope
is
that
we
will
have
some
new
testing
infrastructure
and
some
better
CI
infrastructure
available
so
that
it's
easier,
more
efficient
to
write
tests,
and
then
we
will
start
looking
at
test
coverage
and
test
plans
for
major
feature
areas
and
be
a
little
bit
more
purposeful
about
how
we
do
that
and
look
at
identifying
major
testing
gaps.
You
me
just.
C
F
F
Is
you
know
we
will
switch
to
an
elevated
level
of
testing
and
reliability
with
a
better
CI
process
and
hopefully
better
tooling,
to
do
it,
and
then
you
know
we
would
start
to
see
features
be
gated
on
meeting
that
test
bar
now
we're
not.
We
can't
do
that
because
that
infrastructure
doesn't
exist
yet,
but
the
plan
is
to
make
that
switch,
and,
ideally
sometime
before
the
one
got
to
release
yes
like
I
would
like
to
start
as
soon
as
we
release
one
hi.
F
F
One
of
the
things
that
we
did
a
lot
of
in
1.1
was
making
sure
that
they
were
well-defined
api's
between
every
component.
What
that
implies,
though,
is
that
and
every
for
every
feature,
your
feature
is
tested
at
every
layer
of
the
API
dependency
tree
right.
There's
a
test
for
your
feature:
testing
the
propagation
of
your
API
through
gallery.
F
H
F
Q
Q
Q
Thing
like,
ideally,
if
there
is
a
feature,
let's
say,
I'm,
looking
right
now
at
multi
cluster
plus
the
gateways
feature
so
ideally,
I
would
like
to
have
a
script
which
which
I
just
run
and
it
gets
to
set
up
and
then
one
of
clogs
of
choice,
ideally
with
switchable
backends
something
I
started.
Looking
me,
but
yeah
I'm
gonna
do
testing.
Q
F
One
we
want
to
get
the
tests
as
stable
as
possible
with
as
much
automation
as
possible
in
a
way
that's
maintainable,
so
we
can
enforce
the
testing
standard
and
then
to
how
applicable
are
those
tests
to
many
environments
anytime,
you
have
a
cross
product
explosion
of
the
tested
environment
right,
particularly
increased
cement,
you're
gonna
cause
yourself,
flakiness
I.
Think
that
has
historically,
you
know,
have
hearings.
L
F
F
E
F
E
Two
points
to
make
here
when
very
related
I
mean
one
of
the
reasons
I
mean
people
do
care
about.
The
program
is
that
the
current
infrastructure
I
mean
it's
there.
You
have
to
debug
needs
you
you
need,
we
don't
have.
We
cannot
reproduce
it
if
you
cannot
reproduce
it
your
own.
You
just
look
at
looking
at
the
log.
It's
very
hard
to
fix.
One
of
these
probably
cannot
SSH
one,
sir.
You
can
SH,
but
that's
complicated.
E
F
E
Q
K
A
F
We
have
new
people
working
on
CI
we're
going
to
help
us
with
this
transition,
but
it's
not
ready
to
go
yet,
and
so
we
can't
ask
people
to
change
until
it's
not
fair
to
ask
people
to
change
their
development
patterns
without
giving
them
support
right
right.
So
one
of
the
reasons
why
we're
talking
about
things
going
into
1.1,
as
opposed
to
1.2,
is
to
at
least
in
one
not
one
require
an
elevated
level
of
quality
and
caution
on
people's
behalf.
If
there's
no
need
for
structural
support
to
enforce
it
yet
again,
I.
E
Have
a
slight
disagreement
here:
I
mean
we
do
have
a
framework
in
place,
I
mean
Stefan
has
been
running
since
0-8
was
quite
success.
We
got
fully
automated,
it
doesn't
mean
it
doesn't
exist
and
quite
a
few
people
managed
to
get
it
working
to
want
to
test
to
it.
I
mean
I.
Think
it's
not
such
and
I
think
having
people.
E
C
C
L
Was
dumping
I
know,
I
personally,
think
everybody
in
this
project
that
develops
software
for
this
project
should
understand
how
the
checks
and
Gates
work,
even
if
it's
hard
it's
the
hardest
part
of
software.
In
my
opinion,
it's
gonna
get
understanding
the
test
infrastructure,
but
it's
worth
the
investment
I
think
if
we
can
figure
out
how
to
do
that,
that
would
be
beneficial.
I,
don't
have
any
immediate
answers.
You
know
for
me,
I
just
experienced
it
over
and
over
and
over,
and
got
so
frustrated
with
the
bugs,
with
the
gate-crashing
that
I
fix
bugs.
L
C
Just
so-so
so
I'm
trying
to
kind
of
bridge
what
you
just
said
in
what
Causton
said
and
one
possibility
this
is
a
strongman
is
that
we
view
this
as
an
optimization
problem.
We
had
tools
right
now
that,
unfortunately,
not
enough
people
know
how
to
use,
but
they
do
work
and
they
do
give
us
some.
If
you
know
how
to
interpret
them,
they
actually
do
testings
relatively
well.
We
start
requiring
that
now
and
then
we
in
parallel,
build
tools
to
make
this
easier
and
require
less
expertise.
A
F
Q
Q
D
A
L
I
guess
my
only
my
only
complaint
about
when
I
was
learning
now
to
do
that
work.
The
gate
work
is,
there
is
no
onboard
documentation
at
all
anywhere
in
a
repo
or
at
like
at
all
about
how
to
do
development
in
those
in
the
tests
directory
or
in
the
proud
directory
I
mean
you
really
have
to
read
the
code
and
understand
how
it
works,
to
be
able
to
work
on
the
gate
and,
if
I
think,
if
we
can
do
some
onboarding
Docs,
that
might
help
we
enjoyed.
A
Q
There
is
I
mean
yeah
I.
Think
one
important
thing
here
would
be:
even
the
person
doesn't
know
how
to
test.
We
still
should
be
able
to
get
get
this
feedback,
this
broken
builder
or
some
word.
If
you
think
something,
no,
because
if
people
don't
don't
bother
to
learn
how
to
perform
the
semi
menial,
yeah,
semi,
automated
sorry,
the
thing
and
the.
C
G
F
With
giving
people
test
cookies
yeah,
so
they
they've
shown
right
I'm
the
goal
for
anything
local,
like
you
say,
run
presumably
the
amount
of
infrastructure
you
they
might
have
set
up.
You
need
to
do
to
run.
Presubmit
right,
ideally,
would
be
build
test
or
make
test
those
about
gender
barrier
to
entry.
It
might
be
hard
to
achieve,
and
one
of
the
issues
we
have
is
doing
that
component
by
component
to
wait
what
pilot
needs.
The
test
is
different
from
what
homeboy
party
needs
to
test
for
writing.
D
S
I
F
Q
There
are
easy
ways
there
and
I
completely
agree.
If
there
is
no
automated
test
that
we
can
trust,
the
features
should
be
considered
broken
and
that's
it.
It's
like
no
manual
testing.
It's
like
you
see
that
right
now
we're
basically
I'm
walking
through
a
bunch
of
like
features
that
are
implemented
quite
some
time
ago,
and
now
we're
asking
you
does
this
feature.
It
won't
work
now,
just
like
kubernetes
Engrish
work
goes
like
this
Multi
cluster
completely
still
working,
we
can
know.
Yes,
we
should
assume
as
long
as
there's
no
tests
that
we
can
frost
it.
E
And
and
we
have
a
spreadsheet,
with
step
of
1:1
tests
that
we
perform
and
each
bit
is
so
far,
we
have
performs
a
test,
and
we
know
that
when
we
ship
it
works.
Yes,
it's
manual
because
we
didn't
have
time
to
automate
and
only
a
few
people
actually
cutting
both
addresses.
If
we
increase
the
number
of
people
who
work
on
this
kind
of
testing,
they
will
have
time
to
actually
automate
it,
because
it's
just
basically
what
gel
is
doing,
adding
some
committee
with
scripts
that
are
not
so
hot
what
it's
just
that
again.
A
We
have
to
dial
back
a
bit
on
the
features
and
focus
on
that
because
I
for
me,
the
manual
thing
is
not
a
solution.
We
cannot
continue
like
that.
I
loved
it
SEO
testing
day,
I
love
to
run
manually
studies
like
doing
manual
stuff
allows
you
to
learn
to
understand
better
things
and
all
that.
But
it's
not
a
release
qualification
criteria
so.
Q
Q
The
way
we
should
wait
just
just
my
opinion,
of
course,
I
have
like
very
limited,
like
scope
of
whatever
scene,
but
I
think
the
first
class
first
was
like
the
visits
this
one
long
leaf
cluster,
where
we
deploy
like
everything,
definitely
helps
with
some
use
cases,
but
overall
agencies
in
a
long-term
story
for
ya
for
this
kind
of
testing,
because,
like
it's
very
easy,
the
cirrhotic
20
pull
requests
outstanding,
all
on
their
own
code
bases.
How
can
run
like
in
parallel,
20
tests
use
basically
right.
Q
Well,
what
if
I
want
to
look
like
you
know,
take
my
time
and
actually
run
verification.
You
know
all
scenarios
are
available
and
the
third
there's
one
like
very
naive,
like
simple
stupid
way
to
get
it
done.
That's
what
I'm
doing
the
world
testing
of
those
scenarios
just
run
as
creep
that,
like
on-demand
you
around
at
this
pins
up
in
you,
cluster
deploys
the
given
version
of
beastie
own
attack,
runs
a
bunch
of
checks
and
tears.
Sit
down,
we
can
run,
but
essentially
was
our
medic
system
with
it
yeah
at
a
small
scale.
Yeah.
T
A
C
F
L
F
F
A
F
F
A
F
My
expectation
and
what
I'm
gonna
do
is
I'm
gonna
reach
out
to
specific
people
in
each
of
the
major
component.
Here
is
asking
white
tests
tests
of
critical
kind
of
structural
use
cases,
but
then
other
people
can
build
them
and
then
we
will
evaluate
so
I,
don't
want
to
just
say
we're
gonna.
Do
it
and
then
find
out
it
doesn't
work
like
we're.
Gonna
have
to
go.
Do
this,
it's
gonna,
take
some
effort
and
then
we're
going
to
evaluate
and
see
what
we
need
to
course-correct
room.
Well,
that's
the
goal,
but.
A
F
E
E
A
F
N
No
disagreement
and
I
completely
agree.
We
need
to
see
more
testing
and
the
documentation
how
to
do
the
testing
I
confess
a
lot
of
tests
that
I
don't
know.
How
do
you
run
locally?
There's
only
a
very
small
handful
testing
I
know
how
to
run
locally,
which,
which
is
it's
very
hard
to
troubleshoot
in
today,
so.
F
F
F
Assume
that
they
are
investing
a
large
proportion
of
the
aggregate,
engineering
effort
and
testing
and
release
hoping
and
over
the
course,
the
next
thing,
and
we
would
like
the
one
that
to
release,
do
not
take
as
long
as
the
one
by
one
of
ways.
Well
pragmatically
my
he
wants
to
include
his
new
features
in
one
love
to.
F
Q
F
A
F
A
A
E
F
E
J
F
I
U
About
yeah
on
that
on
the
testing,
the
the
CNI
I've
just
fixed
the
the
CNI
repo
side
to
test
against
one
dot.
You
know
everything's
passing
now
with
1.1
and
master
the
prowl
job.
John
was
looking
at
that
I,
don't
know
if
he
pushed
posted
a
PR
for
that
or
not,
but
we
we
had
some
problems
with
getting
our
test
and
release
stuff,
actually,
tagging
a
container
and
pushing
it
to
the
release,
location.
So
I
think
that
was
the
blocker
there
for
the
hand
being
how.
F
About
was
most
tricky
features
per
custom.
The
the
user
facing
feature
site
well
to
have
user
facing
impact,
eliminating
the
requirement
of
container
port
Jews
are
facing.
It
was
a
regression,
it's
still
meaningful
work
that
has
to
be
organized
in
schedule
for
and
we
would
put
in
release,
notes
and
go
and
talk
to
customers.
F
E
E
E
F
F
A
C
A
N
A
N
E
A
A
F
A
Q
The
least
of
indonesia's
no
we're
just
a
little
weird
and
what
is
like
generally
the
process
of
like
deciding
and
Fisher.
Would
it
make
sense
to
really
just
just
maintain
yet
our
like
backlog,
triage
few
shoes
and
just
work
through
them
in
terms
of
like
triggers,
so
we
will
go
back
what
Josh
has
been
doing
and.
F
Q
A
E
F
N
A
A
F
A
team
here
that
that
list
is
not
all
gonna
happen.
Already:
that's,
okay!
It's
a
candidate
list!
That's
fine,
and
you
know,
if
we're
all.
This
is
not
a
discussion
of
start
bacteria,
but
features
that
are
opposing
me
turn
if
we,
if
we
focus
on
one
of
two
themes
being
let's
resolve
usability
issues
with
the
features
that
we
already
have,
which
is
kind
of
what's
showing
up
in
that
list
Franklin
and
his
alignment.