►
From YouTube: Kubernetes Office Hours (EU Edition) 20171115
Description
Livestream where we answer user questions about Kubernetes.
https://github.com/kubernetes/community/blob/master/community/office-hours.md
Some links from this session:
- Walid Shaari's thoughts on the Certified Kubernetes exam: https://medium.com/@walidshaari/kubernetes-certified-administrator-cka-43a25ca4c61c
- https://www.cncf.io/certification/expert/
- https://www.cncf.io/announcement/2017/11/13/cloud-native-computing-foundation-launches-certified-kubernetes-program-32-conformant-distributions-platforms/
- Kops: https://kubernetes.io/docs/getting-started-guides/kops/
- kubeadm: https://kubernetes.io/docs/setup/independent/create-cluster-kubeadm/
- Contour: https://github.com/heptio/contour
- Kured: https://www.weave.works/blog/announcing-kured-a-kubernetes-reboot-daemon
- OperatorKit: https://github.com/giantswarm/operatorkit
- Kubernetes on ARM: https://github.com/luxas/kubernetes-on-arm
B
And
I
think
my
mic
was
muted
the
whole
time.
So
nobody
heard
anything
I
say
anyway
welcome
everyone
who
the
kubernetes
officers
we're
gonna,
get
started,
here's
how
it
works.
We
hang
out
in
hash
office
hours
on
slack
on
the
kubernetes
slack,
that's
a
slack
k-8
SEO,
you
ask
questions,
and
then
we
have
experts
here
who
will
answer
them
live
for
the
live
for
the
audience.
So
with
that,
if
you
have
any
feedback
on
us
on
how
the
audio
is
sounding
I
can
adjust
the
volumes
and
whatnot
and
we
can
get
started.
B
And
we've
got
201
people
showed
up
today.
So
hopefully
a
good
amount
of
questions
are
coming.
Let's
just
make
sure
that
the
stream
is
live
and
healthy
and
good,
and
with
that
we
will
start
to
take
some
questions
and
says
we
sound
fine.
So
we're
gonna
begin
here
so
and
asks
what's
the
current
state
of
kubernetes
on
arm.
Is
there
one
single
place?
The
project
is
being
coordinated
or
it
is
a
scattered
about
and.
B
A
A
Most
parts
that
are
generic,
this
project
have
been
integrated
in
the
main
repo
I,
don't
think
like
patches
or
anything
like
that.
Kicking
Brown
and
I
do
believe
there
are
release
artifacts
as
well,
so
there
main
you
know
like
sort
of
distress,
pacific
installation
instructions,
but
a
general
support
for
for
arm
architecture.
History.
A
B
B
He
hangs
out
in
say
cluster
lifecycle
and
you
could
probably
just
toss
him
a
quick
question
and
he'll
give
you
the
rundown
but
I
do
know.
I've
seen
his
work.
There
I've
also
seen
talk
on
the
github
issues
about
getting
binaries
for
power
as
well.
A
little
Indian
power,
so
I
know
that
there's
definitely
work
happening
as
far
as
cross-platform.
B
If
things
go,
I
don't
have
the
exact
I,
don't
have
the
exact
status
of
all
the
cross-platform
binaries,
but
that's
a
good
question
and
then
and
if
you
could
do
us
a
favor
and
whatever
he
tells
you
just.
Let
us
know
in
the
channel
we'll
make
sure
when
we
have
the
next
office
hours,
we
just
give
it
a
shot
out
so
that
we
have
it
on
video
somewhere.
So
hopefully
sometime
in
the
future,
when
someone
sees
it
videos,
they
can
find
out
what
the
state
is,
and
we
can
just
continue
all
right.
B
A
Right
I
mean
I
can
I
can
have
a
stuff
at
that
sure
I'm
not
in
any
way
expert.
This
area
present
it's
been
the
area
which
which
I
wanted
to
experiment
and
but
what
doesn't
get
around
there
yet
so
yeah.
So
we
have
things
like
yeah
customers
versus
operators
and
API
aggregation,
say
I
mean
an
operator
doesn't
actually
have
to
have
a
custom
resource,
the
financial,
what
we
meant
you
could
have
an
operator
that
just
sort
of
acts
on
its
own
for
that,
without
a
particular
without
exposing
a
particular
API
type.
A
So
operators
are
sort
of
a
collective
term
for
things
that
do
things
inside
your
cluster
right.
They
may
optionally
expose
a
custom
resource
initial
thank
a
guy,
but
they
may
just
expose
that
an
API
thorough
and
if
they
feel
like
it,
you
sort
of
up
to
the
author
write
an
ad
I
agree:
Gatien
I!
Believe
that's!
That's!
A
That's
some!
That's
that's
a
particular
case
or
there's
a
special
case
of
a
custom
resource
implementation.
Right,
say
if
you
from
from
from
what
I
understand.
Essentially,
you
do
need
an
API
server.
If
you,
if
you
are
to
support
like
version
of
custom
exhaust
API
right,
so
if
you
want
to
do
conversions
and
such
things,
you
then
need
an
API
server
and
then
you
might
be.
You
have
to
say.
B
B
A
Is
that
sort
of
giving
you
a
better
idea,
Clive
or
or
you
would
rather
even
wait
for
anything
to
respond?
Then
I
may
have
asked
of
the
session,
so
it's
kind
of
like
you
know
it
depends
what
you
want
to
achieve.
Basically,
so
yeah
operator
is
kind
generally
a
patent
and
API
aggregation
is
a
special
case
for
for
some
of
the
customers
or
the
finishing
implementations.
That's
that's
at
least
my
understanding
and
understanding
of
it.
One
of
the
things
you
might
want
to
look
at
is
the
operator.
B
A
A
B
A
Yeah
defaulting
and
validation
right,
that's
nothing,
something
that
leaves
and
the
like,
API
server,
api
server,
builder,
okay,
say
sounds
like
you've
got
you've
done
fairly
that
homework
they're
already
Clyde,
so
yeah,
it's
probably
best.
If
you
join
the
the
next
office
hours
called
where
Andy
will
be
able
to
tell
you
more
about
this
right.
Joe
yeah.
B
Yeah,
so
this
is
just
an
opportunity
to
drop
this
link
because
I
think
it's
useful.
All
the
working
groups
in
kubernetes
have
their
own
meetings
and
mailing
lists
that
are
open
to
the
public.
So,
generally
speaking,
if
you
have
a
problem
and
you
you
kind
of
have
a
general
idea
what
bucket
it
sits
in
I
like
to
toss
this
link
up
those
that
link
has
the
list
of
all
the
SIG's
other
public
meetings
and
their
mailing
lists.
A
In
our
next
session-
and
one
more
thing
I
wanted
to
add
here-
is
that
I
think
cloud
that
thank
you
you're
not
using
go
for
Igor
CRD
operator
implementation.
He
will
be
in
a
very
interesting
one
for
for
folks
who
work
on
this
stuff
so
because
I
mean
most
of
the
people
that
most
of
the
people
in
the
ecosystem
and
up
you
can
go
and
other
languages.
They
didn't
see
as
much
love.
A
B
A
B
Actually,
you
know
what
I'm,
if
Clive,
if
you're
available
later
on
today
for
the
West
Coast
Edition,
because
I
believe
Philip
is
on
the
west
coast.
Maybe
we
can
get
him
on
last
minute,
I'll
see
what
he's
doing
today.
Maybe
he
can
join
us
and
help
answer
some
questions.
That
would
be
great.
So
sorry,
we
couldn't
fully
answer
the
question,
but
it
sounds
like
we're
getting
somewhere
with
that.
One,
let's
see
have
I
miss
any
questions,
feel
free
to
keep
queuing
them
up
in
the
channel.
B
A
A
So,
essentially,
what
we
have
is
we
went
two
ways
deployment
in
in
Amazon
and
then
what
what
does
happen
is
that
you
know
once
in
a
while
install
package
upgrades
and
some
of
those
packages
that
get
upgraded
are
the
kernel
images
and
once
you
have
a
new
kernel
image,
it's
most
likely
a
security
type
update
that
yucks.
Forgive
you
among
the
benefit
from
when
we
went
after
you
reboot
the
system,
so
we
wrote
so
called
people
needed
reboot.
A
That
was
the
link
to
the
blog
post.
Here
say
here.
Is
it
here
here
is
the
announcement
of
our
property,
this
ribbon
diamond
that
basically
allows
you
to
even
like
do
a
rolling
reboot
of
your
Fuhrer
money
when
I
bought
the
notes
have
like
upgrade
updated
packages
and
some
of
the
things
that
didn't
take
care
of.
As
you
know,
draining
a
node
before
just
like
calling
RIBA
making
sure
that
all
the
pods
are
gone
before
before
it
good
to
know.
B
For
those
of
you,
car
OS,
there's
a
core
OS
specific
one
called
clue.
Oh,
they
call
it
the
core
Linux
update
operator,
which
is
quite
handy
for
that
OS
all
right
and
has
another
question.
Excellent,
because
I'm
aware
of
cube,
sprague
cubic
horn
as
two
systems
are
quickly
spinning
up
kubernetes
clusters.
Are
there
other
projects
that
fill
a
similar
role?
Oh
boy,
are
there
lots
yeah
right
cops,
for
example,
yep
yep,
so
there
is
cops.
In
fact.
B
A
B
B
B
A
B
On
my
bare
metal,
I
I
like
to
use
cube
admin,
obviously
the
easy
one
got
you
there's
you
still.
If
you
want
multiple
masters,
you
still
have
to
kind
of
do
that
by
hand
which
is
about
that,
but
they're
working
on
that
for
this
cycle,
so
I've
linked
to
the
announcement
from
the
CN
CF
for
the
32
conformant
distribution.
So
one
thing
the
thing
I
like
about
this
program
is
kubernetes,
has
an
end-to-end
test
suite
and
these
these
here
pass
that.
B
So,
if
you
use
any
one
of
these,
you
should
be
good
to
go,
and
then
there
is
I.
If
I
had
a
nickel
for
every
ansible
playbook
on
installing
kubernetes
that
have
discovered
there
is
a
lot
of
stuff
out
there,
but
these
are
generally
like
the
major
ones
that
are
at
least
passing
conformance,
which
is
important
and
if
you're
into
that
thing,
a
full
disclosure.
A
B
B
A
Like
yeah
and
the
the
kind
of
your
choice
will
depend
on
the
use
case,
and
unless
you
you
have
any
particular
opinions
about
how
you
you
manage
things
if
they
do
convert
just
yes,
then
you
know
some
folks
prefer
to
use
like
cloud
formations
or
something
like
a
nova
stick
for
that,
or
some
forms
ever
have
some
kind
of
distro
policy
that
they
have
to
use
like
else
or
some
things
yeah.
So
it's
like.
It's
really
do.
B
And,
of
course,
everyone's
waiting
to
see
what
Amazon's
gonna
announce
as
far
as
in
that
space,
so
it's
very
exciting
times
in
cops
land.
If
you
go
back
into
this
YouTube
channel,
cig
AWS
actually
just
gave
an
update
during
the
community
meeting.
If
you
want
to
go
back
and
look
at
look
through
the
notes
there
to
get
more
information,
if
you're
deploying
on
AWS,
let's
see
moving
on
people
talking
about
cops,
see
n
go
asks,
a
question:
what
is
the
production
grade?
Ingress
controller
for
Cates
clusters,
nan
Jiang,
nan,
Jie
ke?
A
B
Yeah
so
another
we
just
released
a
hep
to
your
contour,
which
is
builds
on
top
of
envoy,
but
that
just
started
I
would
I
would
be
hesitant
to
call
that
production
ready
right
now.
Although
development
velocity
is
very
fast
on
it,
and
we're
and
Vlad
mentions
traffic
as
well.
I
know
people
that
use
traffic
and
love
it
I,
don't
really
have
an
opinion
on
on
production,
quality
of
ingress
controllers.
B
So,
let's,
let's
kind
of
see
what
this
guy
here
Rafi
ro
FF
II,
says:
nginx
ingress,
AJ,
proxy
ingress,
but
traffic
has
some
serious
limits
when
it
comes
to
TLS,
certs
and
ingress
rules,
it
cannot
handle
them
unless
you
mount
them
into
a
pod,
and
now
a
bunch
of
people
are
typing.
So
it
sounds
like
there's
a
lot
of
opinions
here.
B
B
B
Yeah
and
of
course,
ranger
ingress
requires
you
to
run
ranger,
which
kind
of
try
to
be
a
problem
for
some
people.
It
would
be
so
here's
what
I'm
gonna
do
I'm
going
to
write
down.
We
need
ingress
information
and
I'm
gonna
ask
around
to
see.
If
maybe
we
can
get
someone
who
is
an
expert
on
ingress
on
a
bunch
of
ingress
solutions
to
kind
of
give
us
the
lay
of
the
land.
Does
that
sound?
B
B
A
A
A
B
A
A
B
For
me,
you
know
when
choosing
a
solution:
it's
not
just
hey.
What
technically
works
out
for
me,
but
is
there?
Is
there
like
a
person
or
community
there?
That's
able
to
help
me
out
and-
and
you
know
in
that
regard,
lots
of
people
are
using
cops
and
they're
very
responsive
and
part
of
an
active
sig.
So
I
don't
think
you
can
go
wrong.
A
A
The
you
know,
problems
you're
facing
we've
got
a
few
more
questions
here.
I
thought,
you've
ADM
had
some
shortfalls
and
same
topic.
Great
I
thought
ADM
had
some
shortfalls
as
it
become
the
community
de
facto
installer.
That's
right.
Q
ADM
has
become
the
communities
in
fact
install
some
of
its
features
as
and
we
make
steel
bird
not
to
be
used
in
production
morning
which,
which
I
would
read
more
like
you
know,
use
it
at
your
own
risk
essentially
writes
in
some
some
features.
May
it
may
may
change
terror.
Certain
things
may
change
in
the
future.
A
B
B
A
B
B
Right,
like
you
have
to
manually,
do
it,
which
is
still
kind
of
sucks.
So
now
that
upgrades
have
landed
so
I've
actually
upgraded,
and
it's
nice
like
when
you
do.
You
do
cube
admin
upgrade
plan
and
it
tells
you
this
is
what
version
you're
on
these
are.
What
versions
are
available?
So
you
can?
You
can
say,
let's
say
you're
on
one:
seven:
you
can
go
from
171,
172
173
or
you.
B
It
gives
you
the
option,
it
says
or
I
could
take
you
right
to
one
the
1.8
series
and
it
tells
you
this
and
then
it's
a
two-step
process.
You
upgrade
the
master
plane
first
and
takes
like
a
few
minutes
and
after
that
you
basically
just
run
the
upgrade
on
all
the
nodes
for
all
the
packages,
and
that
was
it
like
it
was,
it
was
pretty.
Seamless
is
pretty
cool.
So
now,
really
the
only
thing
stopping
you
know.
B
What's
once
the
H
a
masters
set
up
is
automated
and
like
we
do
that
by
hand
for
people,
but
once
that's
automated,
that's
pretty
much
like
the
last
big
thing
that
people
that
people
ask
you
for
it.
It's
part.
As
far
as
someone
asks
as
far
as
will
it
replace
cops
that
kind
of
thing,
ICQ
Badman,
is
more
of
like
a
raw
tool
that
pokes
stuff
right.
B
Still
need
a
tool
like
cops
to
do
the
AWS
specific
bits
right
that
might
be
out
of
scope
for
cube
admin
right
yeah
like
there's
very
there's
very
segmented,
like
queue
Babbage
to
do
these
things,
and
only
these
things
and
then
let
a
tool
like
cops.
That's
cloud
specific
or
you
know,
to
kind
of
push
cube,
admins
buttons,
but
a
lot
of
that
will
be
will
be
invisible
to
most
users
knock
on
wood.
Hopefully
you.
B
A
So
I
think
that's
an
interesting
question,
so
there
are
two
parts
right.
So
there's
the
part
where
that
something
like
spark
support
is
an
important
priority
and-
and
you
know,
what's
what
about
general
spark
tons
of
flow
and
tech.
Things
like
that
right,
so
I
wonder
in
what
sense
you're
asking
about
were
these
important
priorities
that
you
would
like
to
receive
whether
you'd
like
to
see
this
improve
soon
or
you're.
A
Thinking
of
this
versus
something
else
that
you've
been
waiting
for,
you
know,
like
spark,
support
versus
some
other
feature
that
you,
you
really
want
to
see
supported
sooner
so
I
wonder
what
about
that
part
of
the
question,
but
in
general,
yes,
spark
spark
spark
has
landed
fairly
recently,
I
haven't
seen
it
myself,
but
I
do
know.
Actually
one
person
who
is
earlier
that
was
supplied
from
Sheldon
they
they
do
spark
intensive,
low
and
kubernetes
that
they
have.
A
A
B
B
B
Now
214
people
in
the
channel
I
had
I,
don't
know
who's
viewing
we
so
the
viewing
live.
Last
time.
I
check
was
not
that
many,
but
we
get
a
long
tail
of
people
who
I
believe
watched
the
episodes
after
probably
or
something
like
that,
Rafi
ass
says
I've
got
one:
do
you
know
any
planet
I'm?
Sorry
if
I'm
pronouncing
that
wrong
by
the
way
do
you
know
if
any
plans
to
enable
additional
data
planes
for
persistency
in
kubernetes,
I'm
thinking
about
Etsy
D
replacements,
for
example,
consoles
zookeeper
or
that
sort
of
thing
yeah.
C
C
It's
part
of
the
control
plane,
the
the
backing
data
storage
or
CP
storage
layer
has
like
deep
hooks
for
some
of
the
features
that
we
wanted
to
enable
so
exci
d3
offers
like
some
really
cool
things
like
three
phase
commit
for
multi
commit,
and
we
we
put
in
a
lot
of
work
to
abstract
out
the
interface
layer
so
that
we
can
have
boundaries
to
know
put
in
these
new
features.
There
is
an
interface
layer
and
people
had
tried
programming
against
console,
but
it
was
kind
of
dropped,
because
the
review
was
too
large.
C
It
was
such
a
massive
amount
of
change
and
a
second
problem
was
there's
no
easy
way
to
do
the
testing.
So
if
someone
wanted
to
add
that
level
of
integration,
that
would
require
a
massive
test
load
because
you'd
have
to
be
testing
for
both
at
CDs
versions
as
well
as
console
versions
for
all
the
tests,
because
that
is
a
large
test.
Metrics
upstream
has
decided
to
punt
on
that
I.
A
Must
be
wonder,
you
know
a
white
horse
asked
about
this.
Whether
this
is
the
case
of
folks
wanting
to
use
a
data
store.
They
already
have,
or
you
know
just
in
general,
like
sort
of,
why
should
be
creating
xv+
trip?
We
already
have
a
console
cluster
for
such
a
such
use
case
and
folks,
wanting
to
you
know,
does
study
separate
and
takes
on
this
I'd
be
late,
but
I'd
like
now.
Doctor
here
will
put
roses
s.
That
is
a
trophy.
B
A
What
they
they
have
to
say
about
that.
What
why
why
they
asking
this
question,
because
I
mean
there
are
different,
consistent
that
you
know
may
be
appropriate
for
hold
and
based
on
that,
you
know
like
is,
that
is
that
that's
eh-eh,
that
they
don't
want
to
set
up,
or
although
something
else,
they're
the.
C
Most
common
answer,
I've
seen
I,
know
Rafi.
It
stop
typing
right
now,
but
the
most
common
Esther
I've
seen
is
that
they
use
console
for
service
discovery,
so
absence
of
like,
and
they
already
have
like
a
legacy
install
that
they
depend
upon
for
broader
service
discovery
across
their
enterprise.
So,
like
a
lot
of
people
still
use
Eureka
and
that
they
have
all
kinds
of
weird
Eureka
integrations.
I
would
call
it
weird
they're
just
integrations
because
they
don't
want
it.
They
can't
offload
from
Eureka
and
order
zero
time.
Yeah
yeah.
We
just
got.
A
A
B
B
C
I'm
Tim
st.
Claire
I've
been
working
on
the
project
since
it
started
pretty
much
I've
been
I've
been
around
the
community
for
a
long
time.
I've
slightly
done
several
different
things
and
I'm
on
the
steering
committee
too,
as
well
I've,
been
doing
all
kinds
of
different
work
over
the
years
and
for
different
organizations.
Most
recently,
I
work
for
hep,
Bo
and
I
am
doing
a
lot
of
coop
ADM
upstream
work
as
well
as
sort
of
deployment
configuration
stuff
at
least
nowadays,
but
I
did
a
lot
of
the
early
scalability
work
for
kubernetes
awesome.
B
Awesome
yeah:
you
must
we.
We
were
talking
a
lot
about
cube
and
earlier,
but
we
do
have
some
questions.
I
I
want
to
ask.
I
want
to
go
back
to
client
for
this
question
real
quick
to
get
Tim's
Tim's
point
on
this,
so
I'm,
just
gonna,
repeat
it,
but
Clive
asked
earlier:
how
do
you
view
custom
resources
plus
operators,
an
API
aggregation
moving
forward
in
the
future,
the
pros
and
cons
of
each
and
general
feedback
for
extending
whoa?
B
Hey
sorry
yeah,
so
a
zoom
basically
was
like
the
hosted
to
show
up
goodbye
and
hung
up
on
me.
So
that's
that's
good
to
know
for
next
time,
but
the
stream
should
still
be
going.
How
do
we
sound
audience
if
you
can
give
us
a
quick,
sound
check,
that'd
be
great
yeah,
so
cloud
was
basically
asking
our
opinions
on
CRTs
and
operators
and
API
aggregation
moving
forward.
B
C
See
are
these
are
a
much
lighter
weight
solution.
So
if
you're
trying
to
do
simple
integrations
see,
our
DS
is
probably
the
way
to
go.
If
you're
looking
to
stand
up
your
own
API
that
you
want
to
support
like
an
extension
mechanism
like
I,
want
to
develop
and
deploy
and
create
this
service
across
every
kubernetes
cluster,
then
you
can
do
API
server
aggregation
where
you
create
a
separate.
You
know
reusable
module
that
you're
gonna
support
with
this
separate
API.
C
Now
people
can
go
API
server
aggregation,
they
could
do
either/or,
it's
just
much
more
of
a
heavyweight
solution
because
you
have
to
manage
your
@
çd
integration,
etc.
Okay,
CR
CR
DS
uses
the
existing
@
CD.
That's
there,
but
API
server
aggregation
currently
now
I
could
be
wrong
here.
Cuz
I
might
have
changed
it.
It
requires
the
standup
of
its
own.
Oh,
it's
separate
control,
plane.
A
C
B
Yeah,
we'll
see
we'll
see
what
he
says
and
while
we
wait
to
come
around
back
to
that
Tim's
to
pass
from
what
I
can
find
the
uptake
of
horizontal
pot
scaling
is
pretty
low.
I
can
hardly
find
examples.
Do
you
think
this
feature
is
too
complicated
or
generally
not
very
useful.
I
personally
think
it's
pretty
neat,
but
it's
hard
to
implement
reliably.
What
do
we
think
I?
Don't.
C
Like
it
never
liked
it,
but
that's
my
opinion,
well
it
because
it
becomes
thorny
and
people
kind
of
it's
easy
to
blow
your
foot
off
mm-hmm,
which
is
the
problem
that
people
face.
Sometimes
it's
not
that
I,
don't
it's
kind
of
like
a
it's,
a
powerful
tool,
that's
easy
to
misuse,
yeah
and
that's
the
reason
I!
Don't
like
it's,
not
that
it's
not
a
good
idea.
C
It's
super
easy
to
make
a
big
mistake
and
some
some
people
have
made
mistakes
where
they
auto
scale
down
to
zero
and
then
Allah
send
their
services
go
down
and
some
people
don't
realize
that
the
dimensions
of
scalability
and
nothing
to
do
with
the
current
metrics
that
they're
using
right,
mm-hmm
so
auto
scaling
currently
is
only
CPU.
And
you
know
people
were
adding
networking,
but
in
reality
you
want
to
have
a
separate
controller
that
can
introspect
your
service.
That
knows
what
your
service
is
doing
and
scale
on
the
dimensions
that
you
find
useful
right.
C
A
C
A
A
B
B
A
C
B
B
C
The
community
support
I,
always
because
I
have
work
on
code
admin
nowadays,
I
always
recommend
abandoned,
because
you
know
it's
a
it's
a
well
trotted
path
and
a
lot
of
people
are
working
on
it.
I
always
follow
community
is
usually
the
answer
for
like
deployment
tools
and
whatever
has
the
most
active
community.
That
is,
you
know.
Interpreting
often
and
following
upstream,
is
important.
B
So
if
that
is
important
to
you,
you
can
kind
of
start
with
that
as
the
short
list
for
my
bare
metal
at
home,
I
like
to
hobby
I,
call
it
hobby
kubernetes
when
I'm
not
like
on
AWS
or
something
I
use,
Q
Badman
and,
as
we
mentioned
earlier
before,
the
only
gotcha
there
is
to
do
the
H,
a
masters
which
you
still
kind
of
have
to
do
by
hand.
So
that
is
like
the
last.
The
last
great
limitation
of
Q
Batman
is
is
getting
worked
on
this
cycle.
We're.
C
Starting
to
work
on
it
and
to
be
honest,
every
solution
that
exists
for
each
a
it's
kind
of
this
there's
a
it's
a
fractal
question,
which
means
that
there
can
be
many
racers
and
I
think
what's
happening.
Now
is
we're
trying
to
distill
down
the
best
practices,
and
you
would
have
all
these
solutions
and
we
want
to
unify
them
into
or
codify
them
into
a
known
some
of
the
best
practices
that
we
want
a
version.
B
C
C
You
know
trivial
I
do
realize
that
we
do
have
this
warning
in
there
because,
to
be
honest,
everything
that
has
the
beta
tag
should
be
treated
that
way,
but
there's
a
of
people
in
the
community
who
are
supporting
this,
who
are
working
on
it,
I
think
the
general
good
rule
of
thumb
that
I
always
follow
and
I
tell
other
people
to
follow.
You
just
don't
install
the
latest
zero
right,
wait
for
like
a
dot
two
or
three
from
a
major
from
a
minor
release
revision,
because
that
way,
you're
more.
C
B
You
go
at
your
own
pace,
but
it's
also
nice
to
just
grab
the
point
release
fixes
without
having
to
commit
to
like
a
major
upgrade.
You
know.
So
that's
that's
really
cool.
Whichever
one
of
you
planned
cube
admin
plan
that
command
is
like
ninja,
I
love
it.
So,
let's
see
Rafi
wants
to
ask.
He
says
I'm
not
on
the
panel,
but
if
you're
gonna
go
for
a
tool,
use
cube
admin,
but
my
personal
recommendation
is
to
try
the
hard
way
to
learn
all
the
components.
B
You'll
have
a
ton
of
experience
to
debug
your
environment,
built
from
scratch.
There's
a
lot
of
problems
will
hit
you
early
on
I
can't
say
enough
about
this.
I
I
went
through
the
hard
way
once
and
I
kind
of
told
myself
I'm
gonna
only
do
it
once,
but
it
does
kind
of
give
you
a
nice
sense
of
accomplishment
to
kind
of
figure
how
everything
is
sort
of
like
the
first
time
you
compiled
your
own
kernel
and
then
after
that
you're
just
like
never
again,.
B
A
B
B
B
At
the
old
gay
guy,
I
was
involved
in
one
of
the
initial
in
the
initial
meeting.
I
think
we
all
went
out
to
the
Presidio,
which
is
really
cool
with
the
CNCs
folks
and
kind
of
started
to
hash
out,
but
I
haven't
been
tracking
it
lately,
while
he
says
I've
taken
it,
it's
a
bit
hard
as
it's
both
operations
and
dev,
so
I
dunno
how
they
take
like
the
way
they
take.
A
B
C
Right
so
the
the
standard
practice
for
people
is
easier
to
use
some
floating,
VIP
or
DNS
to
globally
resolve
some
load
balance
master,
and
you
know
they
have
engine
X
or
H
a
proxy
configurations
as
a
front-end
for
that
load,
balance
and
then
you're
going
to
switch
to
VIPs
from
H
a
proxy
but
going
forwards.
We
want
to
investigate
using
envoy
as
a
means
for
doing
the
you
know.
Adding
and
removing
masters
will
be
kind
of
trivial
in
that
type
of
environment,
and
because
everything
is
side
card
as
a
proxy
to
the
master
you
could.
C
B
B
So
Tim
someone
asked
earlier
opinions
on
ingress
what
people
are
using
in
production,
I
kind
of
mentioned,
how
a
hefty
a
launch
contour?
What
what
are
you
seeing
as
far
as
people
running
in
production
and
ingress
I?
Think
we
kind
of
talked
about
it.
A
little
bit
and
the
consensus
in
the
channel
generally
seem
to
be
nginx.
Ingress
is
still
kind
of
what
people
are
using
in
in
real
life.
I
think.
C
Today,
that's
probably
gonna
be
the
answer
for
most
people
and
that's
you
know,
there's
been
a
lot
of
bug
fixes
over
time.
So
it's
it's!
It's
okay
for
now,
but
I
think
going
forwards.
If
you
have
a
larger
cluster
and
you're
doing
and
services
are
coming
and
going
all
the
time
I
do
see
envoy
and
something
like
contour
is
being
a
better
fit
for
the
dynamic
nature
of
the
larger
cluster
environment,
mm-hmm
and.
B
We
are
running
out
of
time.
We've
got
basically
three
minutes
left,
so
I
think
we
have
time
for
one
or
two
more
questions.
If
you
want
to
ask
it,
I
see:
Chris
love
has
joined
us
in
chat.
Welcome,
Chris,
he's
one
of
the
cops
cops
people.
So
those
of
you
that
were
asking
questions
about
cops.
He
is
your
go-to
man.
C
That's
it,
it's
kind
of
a
contour
is
a
fds.
Contour
is
a
ingress
using
envoy.
You
can't
you
could
use
pure
envoy,
but
you
then
you'd
have
to
do
the
same
things
that
contour
does
it.
Steel
has
a
bunch
of
stuff
loaded
into
it,
so
it's
kind
of
like
if
you're
just
looking
for
an
ingress
solution.
Contour
is
probably
better
and
you
would
do
the
same
things,
but
if
you're
sto
I
think
is
gonna
have
a
long
tail
of
development.
C
I
think
it's
awesome,
but
I
think
it's
it's
still
very
nascent
at
this
phase
and
it
tries
to
do
a
lot
of
other
things,
which
is
basically
how
do
I
set
up
a
the
pure
mesh
network
with
sort
of
flow
control
and
other
monitoring
across
my
whole
environment,
and
it's
it's
cool
idea
and
I.
Think
it's
it's
gonna,
be
an
awesome
next
step
for
the
community,
but
I
think
it
still
it's
still
early
days
for
that
yeah.
B
B
So
that
will
be
quad
definitely
definitely
want
to
get
some
sto
folks
on
as
well.
To
answer
questions
that'd
be
cool
just
kind
of
so
people
can
ask
questions
about
that.
Thanks
for
the
questions
Vlad,
and
that
brings
us
to
the
top
of
the
hour-
we're
pretty
much
done
here,
I'd
like
to
thank
our
panel
Ilya
and
Timothy
for
taking
time
out
and
thanks
to
have
dealing
weave
for
allowing
their
engineers
to
come.
Hang
out
and
answer
your
questions.
We
do
this,
the
third
Wednesday
of
every
month.
We
have
two
sessions.
B
We
have
one
today,
we'll
have
one
in
about
six
hours
for
the
kind
of
the
timezone
shifts
so
we'll
try
to
we
get
different
panelists
on,
so
we
we
do
try
to
to
mix
it
up
a
little
bit
and
get
different
levels
of
expertise.
So
thank
you,
everybody
who
showed
up
today.
We
hope
you
show
up
again.
Please
share
this
video.
B
Let
people
know
the
idea
here
is
to
make
this
valuable
for
you,
so
that
if
you're
ever
stuck
at
work
or
something
like
that,
you
have
a
way
out
to
get
an
avenue
to
an
expert
without
getting
lost
in
the
noise
of
just
asking
one
huge
Channel
right.
So
thank
you
very
much
everyone
for
coming.
We
hope
you
enjoyed
it
and
we
will
see
everyone
in
a
month.
Thank
you.