►
From YouTube: Kubernetes Office Hours 20190619
Description
Office Hours is a live stream where we answer live questions about Kubernetes from users on the YouTube channel. Office hours are a regularly scheduled meeting where people can bring topics to discuss with the greater community. They are great for answering questions, getting feedback on how you’re using Kubernetes, or to just passively learn by following along.
For more info: https://github.com/kubernetes/community/blob/master/events/office-hours.md
Links from this show:
- https://github.com/kubernetes/community/pull/2148
- https://jaas.ai/canonical-kubernetes
- http://github.com/pusher/wave
- https://github.com/goharbor/harbor
- https://jfrog.com/artifactory/
- https://www.asykim.com/blog/deep-dive-into-kubernetes-external-traffic-policies
- https://github.com/kubernetes/enhancements/blob/master/keps/sig-apps/sidecarcontainers.md
- https://docs.google.com/presentation/d/1_q3I2grii7_E9KIxKZ3PMy7Zj2-NiUCXnkJKqRU0QR0/edit#slide=id.g401c104a3c_0_0
- https://openai.com/blog/scaling-kubernetes-to-2500-nodes/
- https://kubernetes.io/docs/tasks/inject-data-application/downward-api-volume-expose-pod-information/#store-pod-fields
- https://github.com/pusher/quack/
- https://www.openpolicyagent.org/
A
Is
ready-
and
here
we
go-
welcome
everybody-
it
is
a
third
Wednesday
of
the
month.
That
means
it's
time
for
your
monthly
kubernetes
office
hours.
This
is
the
live
stream
where
we
hop
onto
YouTube,
and
we
decide
to
try
to
answer
as
many
of
your
questions
as
possible
as
quickly
as
possible.
So
we
are
in
hash
office
hours
on
slack
top
kate's
dot,
IO
just
hop
on
there.
You'll
get
an
invite
and
you
can
join
us.
A
Then
you
can
hop
in
the
slack
channel
which
we
are
streaming
here
on
the
side
where's
that
it's
over
here.
You
should
see
that
in
the
livestream
video
and
we'll
try
to
answer
as
many
of
your
questions,
so
we're
gonna
quickly
introduce
our
panel
and
then
we
will
get
started.
Kick
it
off
for
me,
Bob
all.
B
Right,
my
name
is
Bob
Dylan
I
am
a
research
cloud
administrator
for
the
University
of
Michigan
in
kubernetes,
for
quite
a
while
now
sort
of
my
specialty
is
things
like
on-prem
deployments,
Federation
and
multi
cluster
and
all
sorts
of
sort
of
created,
crazy
stuff
in
between
I
also
work
with
with
Jeff.
If
you
can
tell.
C
Hello
I
may
as
well
go
next.
My
name
is
Jeffrey
Sica
I
also
work
at
the
University
of
Michigan
also
been
working
with
kubernetes
for
quite
some
time
and
container
technology
in
general.
My
main
thing
is
batch
jobs,
kind
of
on
purim
deployments
like
Bob,
and
lately
I've
been
doing
a
lot
with
sig
release
and
dashboard.
E
Hi
I'm
Jill
speed
Oh
if
a
pusher
London
office,
I'm,
adding
structural
engineer
and
I've,
been
working
with
Cuban
assets
for
about
two
years
now
done
a
lot
of
stuff
to
deal
with
like
auto
scaling,
authentication
more
recently
been
doing
blood
stuff
with
queue
builder
and
controller
one
time,
and
that
kind
of
stuff
doing
some
custom
controllers
and
Ciardi
stuff.
And
this
week
I
am
working
on
deploying
prowl
for
us,
which
is
from.
F
G
H
I
A
I'm,
your
host
George
Castro
I'm,
a
community
manager
at
VMware
that
gets
to
work
on
kubernetes
so
before
we
start
thanks
everyone
who's
joining
us
in
the
live
chat.
Please
feel
free
to
introduce
yourselves,
tell
us
where
you
from
what
you're
working
on.
We
always
love
to
see
that
kind
of
stuff
and
let's
just
go
over
some
quick
ground
rules
here,
we're
gonna
go
about
an
hour
today.
A
This
is
a
judgment-free
zone,
so
everyone
had
to
start
from
somewhere.
So
if
someone's
asking
questions
that
you
feel
like
to
basic
help
them
out,
don't
don't
be
a
big
bad
judge.
We
will
do
our
best
to
answer
your
questions.
I'm.
However,
the
panel
doesn't
have
access
to
your
cluster,
so
we're
not
going
to
like
try
to
live
debug
and
SSH
into
your
cluster
or
anything
like
that.
However,
what
we
will
do
is
do
our
best
to
get
you
moving
to
whatever
the
next
step.
A
You
need
to
go
to
find
out
kind
of
fix
your
problem,
so
maybe
where
to
find
your
logs,
how
to
debug
certain
things
and
kind
of
generally
at
least
try
to
point
you
in
the
in
the
right
direction.
Panelists
you're
encouraged
to
expand
on
your
answers
with
your
experiences,
your
pro
tips,
anything
you've
broken
it.
Action
is
always
helpful
same
with
all
of
the
those
of
you
listening
in
the
slack
channel.
A
What's
a
good
thing
to
use,
for
this
feel
free
to
just
whack
all
that
in
there
what
I
do
at
the
end,
as
I
grab
all
the
URLs
and
put
them
in
the
show
notes
to
publish
with
the
video
it
just
helps
out,
because
there's
so
many
things
out
there
in
the
kubernetes
ecosystem.
It's
always
great
when
the
rest
of
the
audience
helps
us
collect
as
much
information
as
possible.
A
You
can
post
your
questions
on,
discuss
that
kubernetes
that
I
know
if
you're,
not
on
slack.
We
watch
that
thread.
In
fact,
Bob
is
about
to
paste
the
link
to
this
discuss
thread
into
these
slack.
So
you
have
it
those
of
you
listening
in
the
stream.
That's
just
a
slack
channel
there
on
the
side,
there's
about
a
1/2
minute
delay.
So
sometimes
what
might
happen
is
we'll
ask
you
ask
your
question.
We
might
have
a
question
for
you
to
get
more
information
and
that
will
go
on
to
another
question.
A
While
you
gather
your
information
and
then
we'll
get
back
to
you,
if
at
any
time
you
keep
asking
you
a
question
and
for
some
reason
we
don't
see
it,
please
let
us
now
start
putting
emojis
use
all
caps
whatever
it
is
to
mean
we
don't
purposely
skip
questions.
We
try
to
address
them
all
as
we
can
so
all
right.
So,
let's
get
started
I'm
being
told
that
zoom
is
gonna,
kick
us
out
in
about
a
half
hour,
so
the
stream
will
do
a
quick
interruption.
10.
A
A
A
A
J
A
A
K
D
K
A
Okay
in
mom
was
his
name.
I
will
try
let's,
let's
get
the
questions
going
and
I'll
find
them
on
slack
and
sound
again
so
chat.
How
are
we
sounding
sorry
about
the
technical
technical
issue?
That
is
definitely
my
fault.
Cuz,
that's
two
times
in
a
row.
Alright
Matt
says
sound
is
good.
Are
we
ready
to
get
started?
Let's
get
the
heck
ND
open.
D
So
yeah
there
is
this
issue
that
is
fixed
by
the
way
in
helm,
3
we're
just
now
in
alpha
and
we'll
be
hopefully
finished
by
end
of
this
year
or
even
by
helm
summit,
which
is
in
September.
Until
then
secrets
at
some
point
latest
when
you
fly
them
with
tiller
will
will
stay
a
public.
There
is
no
differentiation
between
config
and
secrets
and
how
I'm.
D
Currently
there
is
no
real
workaround
either
in
we
have
some
work
around
the
transform,
but
in
the
end,
once
you
get
it
into
helm
and
to
tailor
you
need
a
single
various
file
and
there
is
no
way
to
to
encrypt
that.
What
you
could
do
is
basically
use
DHCD
encryption
not
only
for
secrets
but
also
for
conflict
maps,
so
at
least
you
get
it
encrypted
in
at
CD,
but
still
in
in
cash.
They
won't
be
encrypted,
I
think
so,
because
contract
maps
won't
won't
be
treated
the
same
as
secrets.
D
B
D
D
B
B
D
A
B
F
A
Okay,
anything
else
on
this
one,
all
right,
I,
you
should
Verma
thanks
for
joining
us.
Papapa
communication
is
disabled
in
the
latest
version
of
kubernetes,
and
I'm
sure
many
would
have
encountered
this
issue
while
running
the
latest
version.
I'm
not
able
to
find
a
workaround
for
AWS
I'm
using
cops
to
set
up
my
cluster,
and
then
they
sent
this
link
here
for
server
fault.
So
anyone
want
to
take
a
stab
at
this
one.
D
H
E
A
A
E
A
So
I
think
the
first
step,
then,
are
usually
is
check
to
see
if
you
have
a
CNI
installed
and
then
let
us
know,
and
then
we'll
come
back
to
this.
That's
on
okay,
all
right
all
right!
Moving
on
PRS
hello,
there
I
recently
started
looking
into
kubernetes
and
at
this
point,
I
have
a
basic
knowledge
of
how
it
works.
A
Is
this
use
case
something
that
ku
Burnett
suitable
for
even
recommended?
Do
you
know
of
other
similar
use
cases
where
kubernetes
was
used
so
I
think
we
can
start
off
with
the
differences,
I
guess
between
infrastructure-as-a-service
OpenStack
ish
and
what
kubernetes
provides
so
who
wants
to
take
a
stab
at
this
one?
I
can
so.
G
I
think
like
like
kubernetes,
has
a
concept
of
nodes
right
and
like
a
single
virtual
machine
would
be
be
kubernetes
node.
It
has
like
CPU,
it
has
memory,
it
has
like
everything
attached
and
you
have
like
this
one
node.
Now.
What
my
question
says
we
want
to
do
like
a
single
container
would
spin
up
a
node.
So
that's
like
a
bit
of
a
different
concept,
but
I
think
it
would
be
possible
to
implement
that
using
like
built
roll
cook.
Tublat
I
think
once.
D
You,
like
word,
let
would
be
one
of
those
solutions
which
of
cubelet
is
rather
to
have
cubelets
that
are.
Did
you
that
you
don't
see?
Bootlid
would
be
a
solution.
There
is
others
where
you
can
spin
up
a
vm.
Instead
of
a
container,
some
people
put
Kimo
tooling
in
a
docker
container
and
spin
up
a
VM
so
that
those
are
not
really
the
use
cases
for
communities
so
and
there
very
few
people
who
are
doing
that
as
far
as
I
know.
Definitely.
B
D
A
Yeah
also,
when
I
add
it
would
depend
on
what
what
the
workloads
look
like
from
the
people
who
want
to
consume
the
platform,
if
they're
already
all
containerized,
which
add
out,
because
that
would
be
too
perfect
right,
then
some
you
know
just
giving
them.
Kubernetes
is
fine,
but
if
you
have
a
mix
of
really
stateful
workloads,
some
people
move
into
containers.
Some,
not
you
know,
then
you
have
to
kind
of
do
the
judgment
call
there.
A
You
know,
and
it's
totally
perfectly
fine
I
know
some
people
don't
like
this
to
have
a
thing
that
does
VMs
and
have
a
kubernetes
side-by-side.
Instead
of
trying
to
do
a
thing
together,
I
know
a
lot
of
people
do
use
kubernetes
to
drive
up
a
second
the
other
way
around,
but
you
have
to
make
that
judgment.
Call
based
on
the
complexity
of
the
thing
that
you're
building
and
for
a
lot
of
people
kubernetes
is
hard.
Openstack
is
also
hard
doing.
C
Them
together,
I
think
I.
Think
one
of
the
big
selling
points
is,
if
you
can
have
you
know
that
ideal,
one
single
API
to
rule
them
all
in
that
API
is
through
kubernetes
I,
think
that
is
incredibly
compelling
and
what
is
driving
a
lot
of
people
to
try
and
do
this,
but
at
the
same
time
there
is
complexity,
maintaining
it
for
everything
and
I'm
kind
of
mirroring
your
point
just
giving
a
little
bit
more
into
why
it's
happening.
B
D
A
Matt
welcome
back.
First
of
all,
he
wants
to
say
thanks
for
pointing
me
to
harbor
last
session.
I
might
image
repo
up
and
running,
and
it
works
much
better
easier
than
the
docker
repo
awesome
I.
Think
you're,
like
the
first
person
to
come
back
and
actually
thank
us
or
maybe
the
second,
that's
always
nice
thanks.
We
appreciate
that
his
question
is
I've
been
testing
kubernetes
version
upgrades
with
cubes
fillet,
slash,
ansible
playbook,
everything
works,
fine
up
to
version
13.7,
but
anything
14
and
all
the
nodes
go
through
and
not
ready
cycle.
A
D
B
G
E
A
D
A
Okay,
so
let
us
know
how
that
works
out.
We
are
moving
on
met.
Bd
I
guess
is
how
you
say
that
name.
Our
docker
images
contain
some
some
data
ie
within
slash
data,
and
we
create
the
pod
with
persistent
volumes.
The
slash
data
content
is
dropped
by
kubernetes
same
concept
as
a
mount
command.
Some
of
you
know
how
we
bypass
this
feature
merge
the
existing
data
with
the
new
one
when
a
new
pot
is
created
thanks
in
advance
for
your
reply,
I
think.
C
D
You
mount
a
volume
to
that,
no
matter,
if
it's
persistent
or
not,
you
will
overwrite
that
that's
folder
in
in
the
image.
So
either
you
don't
either
you
just
don't
mount
a
volume
to
the
slash
data
or
you
move,
which
would
be
the
better
solution,
move
the
data
out
of
the
image
and
put
it
in
like
yeah
in
a
different
way
into
the
persistent
volume.
A
H
Know
specifically
the
bigger
to
see
their
configuration.
What
is
it
they
have
in
their
pods
back
on
and
see
what
they've
been
down
there
because
I
don't
think
I,
don't
think
coverage,
even
if
they
declare
it
and
what
the
daughter
file
looks
like
if
they
declared
volumes
large
data
a
doc?
Well,
kubernetes
wouldn't
do
anything
about
that
automatically
with
it
long.
H
G
A
This
gives
me
a
follow-up
question
from
my
casual
casualness.
I've,
always
been
taught
or
learned,
never
to
put
data
into
your
like
images
and
containers.
Can
you
all
expand
on
this
a
little
bit?
Are
there
cases
where
this
might
be
a
good
idea,
I'm
just
kind
of
curious
if
they
inherited
this
image
from
somewhere
else?
You
know
like
what
it
feels
like
just
separating
that
would
solve
a
lot
of
subtle
problems.
Maybe
you
might
see
in
the
future.
F
G
Yeah
also,
if
we
have
like
data
which
never
changes
then
like
rocket
file,
might
be
a
good
approach,
because
persistent
volumes
might
not
be
working,
for
example,
and
then
metal
cases
or
yeah,
and
basically
you
would
have
to
actually
mount
the
same
volume
which
would
like
be
our
NFS
or
like
Jefferson,
similar
system.
Then
it
would
actually
do
like
calls
to
that
system.
Instead
of
actually
just
having
the
data
in
the
docket
path.
D
From
the
olden
days
of
containers
without
kubernetes,
it
was
already
kind
of
the
reason
why
volumes
were
introduced
into
docker
was
because
data
shouldn't
live
inside
the
image.
There
is
very
few
use
cases
like
mentions
that
would
make
it
maybe
okay
and
then
maybe
you
need
to
separate
folders
and
have
some
of
your
image
data
in
slash
data
and
some
some
and
something
else
to
separate
between
the
ones
that
are
fixed
and
the
ones
that
will
change
and
that
will
yeah
will
be
persisted
only
on
an
infrastructure
level
and
not
on
a
container
level.
This.
B
H
Vd
I
think
yeah
under
yeah.
Similarly,
like
I
mean
I,
guess
I
get
said
that
this
does
something,
because
if
you
use
ports,
then,
if
you
declare
a
port
in
your
dock
file,
it's
kind
of
almost
completely
appointments
on
the
docker
exposes
that
automatically
right.
It
doesn't
care
about
it,
some
of
those
volumes
yeah.
So
for
maybe
things
that,
could
you
wanted
to
check
in
you
ain't?
H
A
Alright,
moving
on
to
our
next
question:
if
you're
just
joining
us,
this
is
the
kubernetes
office
hours.
We
are
in
hash
office
hours
on
the
kubernetes
slack.
Please
feel
free
to
ask
your
questions.
I,
see
lots
of
questions
queuing
up
in
my
document,
hopefully
Bob
we're
caught
up
to
date,
but
moving
on
to
our
next
question
from
Tarun
postal
potty,
a
student
from
India
welcome,
says
recently
I've
been
trying
to
create
a
CRD
operator.
I
have
a
CR
which
depends
on
another
users
should
be
able
to
reference
the
dependency
R
and
the
independence
ers
definition.
A
D
Don't
think
you
can
do
that,
you
can
either
have
object,
references
or
just
references
that
you
use
and
or
you
you
can
declare
it
like,
like
you
would
do
pots
back
in
a
in
a
deployment
resource,
but
doing
both
that
is
most
welding.
Ups,
not
not
possible,
or
at
least
I
I,
wouldn't
know
of
a
way
to
do
that,
and
it
would
also
be
strange
because
you
you
most
probably
you
had
a
reason
to
have
them
in
two
resources
and
then
declaring
two
resources
in
a
single
resource.
D
E
E
That's
the
kind
of
thing
I
would
probably
suggest
you
do
one
or
the
other,
though
it
depends
on
what
the
relationship
between
them
is,
because
those
are
two
quite
different
relationships,
one
of
them
the
first,
the
deployment
it's
creating
those
pods.
It
owns
those
pods.
It
manages
them,
whereas
the
second
relationship,
it's
more
just
using
them.
So
I
just
saw
posters
on
thread
just
saying
what
actually
of
these
cases
you
tell
us
a
little
bit
more.
E
D
Yeah,
what
we
also
do,
if
you
have
something
like
you-
need
the
issuer
to,
for
example,
to
create
dessert,
is
you
could
use
admission
control
to
validate
and
see
if
there
is
a
reference
missing
or
the
reference
is
wrong,
then
give
back
the
the
custom
resource
affect
the
request
and
say:
hey
you're
missing
the
issuer.
That's
what
we
do.
For
example,
this
was
open
policy
agent,
it's
very
easy
to
do.
You
pull
in
the
list
of
the
first
of
the
dependent
and
then
validate
the
independent
against
the
dependent.
A
J
D
H
D
H
D
H
That
depends
on
whether,
whether
you
care
about
the
day
turnout
right,
you
may
have
you,
may
you
may
have
theta
volumes,
but
problems
are
so
they
have
to
replicate
them.
You
may
have
separate
way
of
sealing
your
databases
or
doing
other
things
or
or
you.
You
may
just
use
a
host
of
databases
right
now.
D
K
D
B
Of
the
use
cases
I've
seen
have
been
less
about,
you
know,
let's
came
out
the
data
and,
more
so
just
copying
and
pasting
the
deployments
in
which
case
I
definitely
+1
the
whole
get
ops
model.
If
you
need
to
control
or
do
it
like
lots
of
times,
that's
when
I
would
say
you
start
looking
at
things
like
Federation
and
more
multi,
clustered
paperwork.
D
It's
also
good,
if,
like
your
first
cluster,
really
gets
burned
down,
and
there
is
no
way
you
have
a
recent
backup
being
able
to
restore
from
scratch
to
a
new
API.
Like
someone
gives
you
some
kubernetes
api,
no
matter
where
it's
running,
can
you
get
backup?
That's
that's
always
my
like
best
case
or
worst
case
scenario,
training
restores
almost
no
one
can
do
that.
Yeah.
A
A
Alright
Oscar
lo
says
late
arrival,
hi
al
started
looking
at
and
learning
about,
kubernetes
at
the
beginning
of
this
year
excited
to
learn
more
and
more
as
I
work
with
it.
My
team
and
I
are
currently
working
on
setting
up
our
stage
and
production
clusters
and
we're
planning
around
on
hosting
services
aimed
at
both
external
customers
and
company
internal
teams.
For
the
internal
services
we
have,
some
company
demands
on
to
what
internal
services
can
be
accessible,
internal
network
or
VPN
or
whatever,
which
are
different
from
the
external
ones.
A
My
question
is,
would
you
say
namespaces
and
other
built-in
separators
are
enough
to
set
up
internal
and
external
services
in
the
same
cluster
in
order
to
gain
the
advantages
it
brings
or
would
it
you
say
it
requires
a
use
of
one
entire
internal
cluster
and
one
entirely
external
cluster
to
keep
things
sufficiently
separated.
I
guess
here
comes
a
multi-tenancy
question.
I.
C
I
was
just
gonna
say,
that's
that's
to
me
pretty
clear-cut,
especially
as
spinning
up
and
managing
multiple
clusters
has
become
easier
and
easier,
and
it's
only
going
to
get
easier.
I
would
just
if,
if
you
are
not
comfortable
with
them
being
on
the
same
cluster,
just
split
it
up
have
two
clusters.
It
should
be
just
fine.
That's
that's
not
to
say
that
the
technology
isn't
there
to
have
it
all
be
in
one
cluster,
especially
with
service
meshes
and
network
policies,
and
all
the
like,
but
I
would
just
spin
up
two
clusters.
It's
gonna
be
safer.
D
A
lot
easier
to
manage
it,
it
can
be
a
lot
more
secure,
especially
if
you're
in
some
kind
of
enterprise
security
department.
Then
it's
just.
You
can
even
set
up
different
kinds
of
clusters,
because
one
doesn't
even
need
an
external
load.
Balancer
can
be
fully
internal,
you
don't
need
any
external
access,
and
that
way
you
have
a
lot
of
different,
completely
different
setup,
which
is
quite
nice.
That
said,
if
you're
starting
out
is
also
actually
quite
nice
to
have
more
clusters,
because
your
blast
radius
is
just
lower
and
wiping
your
cluster.
G
The
also
when
you
think
about
if
you're
like
customers
are
way
different
and
we
have
different
kind
of
workloads.
For
example,
one
could
be
like
mining
Bitcoin
or
something
it
could
be
impacting
the
other
neighbor
on
the
same
node.
So
it's
like
I
would
definitely
use
different
coasters
for,
for
that
and.
A
Operation
and
concerns
hospital
says
thanks.
Great
answers.
Cheers
thanks
for
joining
stick
around
everybody,
because
we
give
away
the
t-shirts
at
the
end,
so
you
have
to
stick
around
for
the
whole
show
so
moving
on,
we've
got
about
20
minutes
leg,
guilhom,
I,
hope,
I
got
that
rises,
hello.
Does
anyone
have
any
resource
articles
papers,
books,
videos
that
describes
patterns
around
dynamic
application,
configuration
management
and
kubernetes
I'm?
A
Looking
for
a
pattern
specifically
to
share
configuration
files,
a
bunch
of
JSON
files
with
pods
and
update
them
whenever
a
change
in
this
configuration
happens
and
then
gracefully
restart
the
pods
I
wonder
if
there's
something
out
like
this
better,
it's
better
opinions,
any
thoughts.
Thank
you.
So.
E
I
replied
to
this
thinking
a
project
of
mine
wave.
So
that's
it
was
designed
for
us
because
we
have
a
bunch
of
applications
that
can't
dynamically
reload
their
configuration.
So
the
config
is
still
in
config
maps.
Someone
updates
compute
map.
You
have
to
go
out
and
restart
the
pods,
otherwise
that
new
configuration
doesn't
get
loaded.
So
we
build
this
controller
wave
that
looks.
E
A
compliments
and
secrets
adds
a
hash
of
the
data
within
them
to
the
pot
spec
on
the
deployment
or
dema
sale
or
sample
set,
and
then,
whenever
they
change
these
zona
references
and
stuff
to
watch
for
changes,
updates
the
pods
back
immediately.
So
as
soon
as
you're
applying
updates
the
conflict
map
that
should
trigger
the
deployment
role
or
something
to
roll
it
out,
so
I
suggested
that
might
be
suitable
for
the
East
Coast.
D
Yeah
pretty
much,
and
currently
you
need
some
some
custom
controller
to
do
that
for
you,
but
basing
it
on
the
standard.
Config
map
is
definitely
a
good
idea,
because
at
least
you
you
get
something
there,
and
maybe
that
that
very
very
long
and
still
not
being
worked
on
issue
will,
at
some
point
being
worked
on,
and
people
have
native
config
map
role
our
functionality
somewhere.
D
A
D
G
A
And
then
we
have
a
new
release
like
today,
ish
and
then
should
be
today
should
be
today
so
good
to
know
all
right.
My
next
questions
gonna
be
from
Imran
and
with
that
we
are
caught
up
on
questions.
So
if
we
miss
your
question,
please
repost
it.
If
you
have
new
ones,
if
you
have
any
follow-ups,
I
believe
we
do
have
a
follow
up
on
the
guy
with
the
CNI,
a
yoosh
I
believe
it
looks
like
they
didn't.
A
A
Use
cases
to
Reese
okay,
so
my
use
case
is
to
restart
the
deployment
pod
every
night,
no
changes
in
spec
or
image
file.
How
would
I
go
about
doing
that?
Declaratively,
active
deadline,
seconds,
didn't
work,
I
believe
from
1.15
onwards
and
will
be
easy
with
cuddle
rollout,
restart
any
workarounds
for
clusters
prior
to
one
dot.
Tea
I
have.
D
Know
I
mean
active
deadline.
Seconds
is
I,
think
part
of
jobs.
So
if
it's,
if
it's
a
part
that
it
could
run
as
a
job
and
you
could
have
a
cron
job,
but
if
it
just
needs
to
run
and
just
get
restarted
because
of
some
reason,
then
currently
its
building,
your
own
pod
killer,
I,
think
there
is
some
configurable
ones
out
there.
G
K
D
D
A
B
A
Pop
that
pop
that
thread
in
there,
if
you've,
got
a
useful
tool
and
you
want
to
share
with
the
community.
We
have
a
long-running
thread
on
the
forums
thats,
basically
post
your
cool
tools
that
you're
finding
on
the
internet,
so
he's
gonna
post
that
link
in
there
by
all
means
feel
free
to
submit
a
few.
That's
really
great,
alright,
also
hi
Alex,
it's
good
to
see
you
Gil
herm,
says
beautiful.
Thank
you,
I'll,
look
into
it.
You
are
welcome.
A
H
Tough
yeah,
you
can
devise
a
plan
right,
but
essentially,
if
you
have
a
small
number
of
pods
you
can
you
can,
you
know,
delete
one
and
install
the
other
or
I
think
actually
well,
it
sort
of
depends
as
well
all
right
which,
which
already
did
but
the
safer
way
would
be
to
delete
the
first
one.
Are
the
new
network
make
sure
that
you
are
able
to
start
a
new
pod
and
the
drums
on
the
new
network?
H
You
can
work,
but
you
know
you
obviously
are
looking
at
deleting
all
the
pods
and
well
there's
I
mean
there's
no
way
to
do
it
without
any
downtime.
Essentially
they'll
be
down
time
and
both
both
communication.
Potentially
you
may
get
away
without
that.
If
say
your
boss,
both
of
your
networks
are
using,
let's
say,
AWS,
PPC
or
or
whatever
the
other
night
here
native
network
at
9:18
is
working.
Options
are
so
if
you
were
to
go
from,
say
flower
in
the
casino
to
no
probably
not
but
I
mean
in
some.
H
D
A
G
Yeah
also
I
have
like
like
make
sure
that
you
don't
change
pad
I
people.
Basically
it's
a
third,
because
if
you
like
use
flannel
by
default
and
calico
by
default,
they're
going
to
be
different,
a
people's,
then
you
have
to
change
like
controller
manager,
cue
proxy
and
everything.
So
it's
you're
gonna
be
downtime,
I!
Think
so!
Yeah,
it's
it's!
It's
gonna
be
like
really
hard.
So
I
would
go
that
out
of
like
creating
a
new
class
that
I'm
just
dating
working
and.
H
Even
if
it
is
a
year
sort
of
business,
critical
production
use
case,
I'm
sure
you
have
support
options
available,
where
you
can
contact
a
network
vendor
and
discuss
how
they
can
help
you,
not
all
network,
implement,
not
every
CNI
implementation
available.
Has
a
network
vendor
that
offers
support,
but
I
can
vouch,
probably
net
that
there
options
available
on
there
I
think
Calico
probably
has
an
option,
but
you
can
pay
for
and
get
support.
A
All
right
anything
else,
don't
see
a
nice
like
I
feel
like
if
I
had.
If
I
was
bare
metal
and
like
my
business
and
I
had
X
amount
of
hardware
right,
clone
it
into
a
cloud
provider
for
a
little
while
then
get
your
stuff
I'll
set
up
redo
it
all.
So
you
don't
have
the
service
downtime
and
then
move
it
move
it
back
down.
It
feels
you
know.
That's
kind
of
a
flexibility.
Kubernetes
gives
you
right,
I
mean
it's
still,
gonna
be
tough
right.
A
You
got
a
clone,
build
anything
and
then
clone
clone
back,
but
yeah.
This
sounds
tough.
Any
other
comments
on
this
one,
but
I'm
really
interested
I'm
really
interested
in
what
approach
you
take
and
if
you
have
any
notes
I
would
I
would
like
to
so.
Please
feel
free
to
get
back
with
us.
Okay,
let's
see
Oscar.
Lo,
we're
gonna
put
your
question
last
Martin
and
a31
821
asks:
is
it
possible
to
hint
services
to
refer
local
ie?
A
E
So
this
question
had
last
time
where
we
came
up
with
the
external
traffic
policy:
local
setting,
so
I,
don't
know
if
that's
IP
vs
compatible.
Maybe
it
is,
but
in
teasing
the
IP
tables
implementation.
What
it
does
is
it
makes
sure
that
if
a
if
traffic
comes
from
externally
into
the
node
and
the
pot
is
on
that
node,
they
it
will
serve
traffic.
If
there's
no
pod
on
that
node
for
the
service,
then
it
just
black
holes
the
traffic.
E
So
if
you
have
health
checks
on
your
mobile,
so
it
should
only
mark
nodes
which
actually
have
a
pod
for
that
service
as
active
and
therefore
any
really
traffic
to
them,
and
then
it
will
stay
local
I,
don't
know
for
pods
apart
what
the
implications
are
and
I
also
don't
know
if
that
works
like
PBS,
forgive
it
go.
I.
B
Believe
it
does
work
with
IP
vs,
but
it
should
only
impact
again
like
external
traffic.
To
my
knowledge,
the
internal
service
will
still
actually
function.
The
same
I
think
there
was
a
networking
thing
that
came
up
about
this
and
preferring
like
local
before
the
pods
are
running
on
the
same
host
prefer
going
to
a
pod
running
on
that
host.
I,
don't
remember
where
I
saw
it.
Yeah.
D
A
D
Ooh,
that's
political
I
would
say
it
depends
a
lot
on
your
use
case.
What
what
I'm?
Seeing
a
lot
I
mean?
We
run
the
nginx
interest
controller,
the
community
one
which
and
be
quite
happy
with
that.
We
have
people
running
traffic
they're,
also
happy
with
that
contour
I've
seen
tons
of
people
who
are
happy
with
that
cuz.
D
G
D
C
F
D
You
were
in
the
cloud,
sometimes
the
call
might
be
to
just
use
the
cloud
load
answer
for
some
things
like
we've
experienced.
For
example,
if
you
have
tons
of
open
WebSocket
connections,
I
get
five
million
concurrent
WebSocket
connections,
that's
almost
the
case
where
none
of
the
source,
software
load
balancers
without
extensive
tweaking
works
and
just
paying
a
few
bucks
for
even
a
classic
ELB
fixed
it
basically
right-
and
that
is
that
is
sometimes
an
option.
But
it's
only
an
option
on
on
the
clouds
and
not
fall
services.
I
guess.
A
Found
the
capital
you
yes
I
saw
I
saw
I
saw
you
toss
that
link
and
then
I
will
I
will
toss
the
notes,
the
links
and
all
that
stuff
on
the
show
notes
in
the
discuss
thread
for
this.
Okay,
with
that
it
is
time
for
the
raffle
so
Jeff
we
were
giving
away
two
shirts,
so
I
hope
you
can
roll
twice,
because
we
don't
have
an.
We
don't
have
a
westcoast
section
today.
A
We're
gonna
try
to
back
the
westcoast
addition
for
those
of
you
that
are
listening
in
we're
just
trying
to
find
a
time
that
fits
for
West
Coasters
that
isn't
right
after
lunch.
So
we're
looking
we're
looking
at
how
we
can
best
do
that,
but
in
the
meantime,
they'll
keep
on
rocking
with
the
EU
session.
So
I'd
like
to
thank
all
of
you
join
us.
We
had
90
people
join
us
today.
So
thank
you
very
much
and
we
will
toss
I
mean
who
are
winners.
Jeff.
A
So
Jame
Rita
and
Tarun
puffle
Patti,
hope
I
get
that
right.
So
how
does
one
enter
the
raffle?
You
just
have
to
ask
the
question
and
we
automatically
put
your
name
in
a
thing
and
then
Jeff
does
this
Dungeons
&
Dragons
role
and
we
figure
out
who
it
is
and
then
that's
good,
but
keep
on
coming
back,
we'll
always
will
always
role
will
always
do
any
raffle.
A
If
you
show
up
a
lot
and
I
notice
to
you,
I
might
just
send
you
a
t-shirt
out
of
the
blue
I've
done
it
before
so
with
that
I
will
p.m.
those
two
of
you
that
have
won
and
get
you
a
CNC
F
store
code,
and
with
that
any
any
last
words
panel
we'd,
like
to
thank
everyone
for
joining
us
next
session,
will
be
always
third
ones
they
have
a
month,
so
that
is
going
to
be
July.
A
17Th
and
I
will
be
in
Greece
that
day,
so
we
will
figure
something
out,
but
we'll
get
you
a
host
or
something
like
that
with
that
thanks
panels,
any
last
word:
panelists
keep
being
awesome.
Yep
thanks,
fatherless
your
first
time
in
Ryan
as
well.
We
really
appreciate
your
participation
if
you
want
to
be
on
this
panel.
This
is
all
volunteers
only
so
feel
free
to
ping
us
and
everyone
have
a
great
day.
Thank
you
later.