►
From YouTube: Kubernetes Office Hours 20191218
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
A
We
are
live,
I
think
we're
live
Channel.
How
are
we
doing?
Welcome
everybody?
It
is
the
third
Wednesday
of
the
month.
Is
that
when
we
do
this,
whatever
it
is
after
Keep
Calm
cloud
native
Khan,
everybody
is
tired
and
it
is
the
last
kubernetes
office
hours
of
the
year,
those
of
you
in
the
office
hours
Channel,
please
let
us
know
how
the
audio
is
sounding
and
all
that
good
stuff
welcome
everybody.
A
So
what
is
this
is
the
kubernetes
office
are
is
what
we
do
is
once
a
month
we
get
on
YouTube,
and
then
we
try
to
answer
as
many
of
your
user
questions
about
kubernetes
as
possible
with
a
panel
of
experts
before
we
get
started
panel.
Introduce
yourselves
I,
didn't
really
give
you
an
order.
So
let's
go
Joel
Bob,
Chris,
Jeff,
Mario,.
B
C
D
E
F
A
And
I'm
your
host
Jorge
Castro
I
work
at
VMware
and
I,
help
organize
and
put
all
this
together
before.
We
start
just
a
quick
note
out
to
the
audience
that
this
panel
is
all
volunteers.
If
you
want
to
get
back
to
the
community
and
sit
on
this
panel,
it's
a
great
way
to
get
started
in
kubernetes,
I'm
gonna
pop
a
link
into
slack
there.
A
That's
chris
did
his
very
first
talk
at
cube
con,
so
we're
really
happy
when
people
start
contributing
to
kubernetes
and
the
next
thing
you
know,
you're
speaking
at
Q,
cube
con
and
today
we're
congratulating
Joe
who
starts
at
Red
Hat
in
the
new
year
and
he's
going
to
be
working
on
actual
kubernetes
itself.
So,
if
you're
looking
for
a
kubernetes
career,
giving
back
to
office
hours
is
a
good
place
to
start
okay.
So
let's
get
started
here.
Let
me
give
everybody
the
quick
ground
rules
before
we
get
started.
A
Those
of
you
that
are
in
the
channel
that
have
been
around
before
go
ahead
and
start
typing
your
question
and
what
we
do
is
we
put
in
a
little
document.
We
queue
them
up
and
then
we'll
get
through
as
many
of
them
as
possible.
So
before
we
start,
this
is
a
judgment-free
zone.
Everyone
had
to
start
from
somewhere.
There
are
no
dumb
questions,
so
please
help
us
out,
even
if
you
think
the
question
is
a
little
bit
too
easy
or
whatever.
A
There
are
no
easy
questions
either,
and
please
do
remember
that
the
CNC
F
kubernetes
code
of
conduct
is
an
effect,
and
while
we
will
do
our
best
to
answer
your
questions,
the
panel
doesn't
have
access
to
your
cluster.
So
we
can't
really
like
a
live
de
bug
into
your
nodes
and
stuff
like
that,
but
we
will
do
our
best
to
get
you
moving
down
to
whatever
the
next
step
is
wherever
the
log
is
where
you
should
be.
A
Looking
that
sort
of
things
panelists
you're
encouraged
to
expand
on
your
answers
with
your
experiences
and
your
pro
tips
audience
you
can
help
us
out
but
typing
along
as
we're
answering
questions.
Maybe
you
have
links
to
the
official
documentation
or
you
have
experience
in
this
a
lot
of
the
value
that
we
get
isn't
just
from
the
people
in
the
panel.
But
the
people
in
the
audience
kind
of
taken
the
time
to
be
like
okay,
everybody.
A
Let's
focus
on
this
one
topic
and
sharing
your
expertise,
that's
what
it's
all
about
and
I
have
been
known
to
just
randomly
give
out
t-shirt
to
people
who
are
just
doing
an
awesome
job.
Please
feel
free
to
just
post
your
questions
to
them
channel
you
can
post
them
on
discuss
like
Riverdance
da
da
yo,
Stack
Overflow
whatever.
A
So
if
you
ask
a
question,
we
will
put
you
into
a
little
raffle
and
we
give
away
two
t-shirts
at
the
end
of
the
hours,
but
you
must
be
present
to
win.
So
you
gotta,
listen
and
listen
to
the
whole,
show
sorry,
and
with
that
panel
we
ready
audience.
How
are
we
sounding
out
there?
It
sounds
like
the
audio
sounding
great.
A
If
you
don't
have
any
questions,
just
let
us
know
your
out
there.
We
do
stream,
the
slack
Channel
in
the
in
the
in
the
live
stream.
So
tell
us
where
you
from
say
hello:
whatever
keep
it
interesting
all
right,
we
ready.
Everybody
comes
up
all
right.
The
first
question
is
for
minifig
thanks
for
joining
us,
because
the
background
on
the
example
I
have
my
I
have
might
be
a
little
bit
complicated,
but
I'm
gonna.
A
C
Say
it
it
ain't,
just
is
it
but
like
it
depends
every
application
is
gonna,
be
different,
specific
looking
at
the
the
summit
domain,
one
like
you
know,
I
have
run
the
sort
of
stuff
in
container
before
and
I've
had
various
issues
where,
like
you
know
it,
will
it
really
doesn't
like
it
when
the
IP
changes
and
sort
of
a
lot
of
like
other
things,
I
got
you
you
could
you
know
use
like
a
staple
set.
Somebody
at
least
have
will
always
have
a
consistent
name,
and
some
some
of
the
things
worked.
Well
with
that.
B
C
D
Yeah
we
have
a
burr.
We
have
a
lot
of
old,
Java
apps
that
were
designed
basically
just
to
run
on
a
single
server.
So
you
know
sessions
and
things
like
that
would
kind
of
go
go
nuts.
If
we
started
scaling
out
with
different
instances
on
the
user
and
so
kind
of
like
what
Bob
said,
we
try
to
make
sure
that
they
run
in,
like
instance,
of
one,
so
we
don't
run
into
those
hurdles.
So
you
kind
of
have
to
know
what
your
applications
limitations
are.
A
D
C
Host
is
is
definitely
one
thing,
especially
when
they
care
about,
like
you
know
the
host
IP
and
things
of
that
nature.
You
do
lose
sort
of
a
lot
of
the
the
power
of
kubernetes
and
sort
of
bring
this
stuff
in
a
che.
But
you
know
we
are
dealing
with
legacy
applications.
You
start
going
to
take
what
you
can
get
and
you
still
do
gain
a
lot
of
benefits
like
just
the
general
better
adoption
of
like
you
know,
we're
keeping
track
of
everything
and
get
now.
C
Yet
the
the
host
stateful
sets
give
you
a
like
persistent
identity
for
the
container,
so
it
always
be
sort
of
like
the
the
name
of
the
thing.
0-1
depend
on
how
many
instances
you
have,
and
so
things
that
care
about
you
know
its
identity
and
being
consistent.
Stateful
sets
work
really
well
and
host.
Networking
is
really
if
it
assumes
it's
going
to
be
managing
like
ports
on
the
host,
and
you
need
to
have
that
persistent
IP.
A
C
A
Awesome
and
as
many
fig
just
did
here,
if
we
ask
a
question,
you
have
a
follow
up,
feel
free
to
just
keep
on
typing
and
we'll
grab
that
information
as
we
go.
Hopefully,
that'll
get
you
started
minifig.
Please
ask
a
follow-up.
Next
comes
Simone
thanks
for
joining,
says:
I
have
a
pod
which
seems
to
be
unreachable.
It
can
only
ping
the
host
where
it
is
running
any
idea
on
how
to
check.
What's
going
on,
I've
tried
the
usual
rough
patch
n
s,
s
dashed
open,
that's
how
I
say
it.
A
D
C
A
Okay,
so
probably
C&I
lines
is
a
good
place
to
start
Symone
that
should
help
you
get
started.
If
you
have
a
follow
up
and
you've
tried
that
or
something
just
let
us
know,
and
then
we'll
wait
for
him
or
her
to
respond.
In
the
meantime,
matt
says
I'm
getting
a
search,
eric
cube,
cuddle
get
pods,
and
then
you
all
see
the
error
there
in
the
notes.
How
do
I
upgrade
this
and
other
search?
I've
tried,
cube,
AB
and
alpha
face
search.
A
C
B
I'll
give
them
a
minute
here:
I
have
a
feeling,
the
cube
ADM
a
long
time
ago,
used
to
use
that
some
kubernetes
SSL
like
originally
and
I,
think
they've
moved
it.
So
what
I
think
probably
happened
here?
Is
this
controller
node
or
with
whichever
one
it
is?
This,
like
controller,
has
just
been
up
for
such
a
long
time,
probably
a
year
or
so
and
not
been
replaced
that
it
doesn't
regenerate
service.
B
If
I
remember
correctly,
at
least
at
one
point,
q
radio
used
to
generate
the
certs
as
the
controller
node
started,
they
had
like
a
little
application.
That
did
that.
So,
if
you
could
add
a
new
controller
to
the
cluster
and
take
this
one
away,
maybe
that
would
solve
the
problem
without
having
to
dig
in
any
further
yeah.
A
C
A
B
Was
replace
the
controller
without
upgrading
it
right,
like
you,
just
bring
up
a
new
one,
eleven
controller
yeah
and
then
take
that
older
one
away.
I,
don't
know
what
environment
on!
If
you
know,
if
you're
on
a
cloud
provider
or
something
that
should
be
very
easy,
if
you
want
bad
metal,
that's
gonna
be
harder
at
111
is
gone.
Yeah.
A
B
A
I
would
definitely
Matt
I
would
definitely
get
a
second
opinion
as
well.
I
mean
it's
out
of
support,
but
I'm
sure
you're,
not
the
only
person
doing
that.
Obviously
we
don't
know
what
your
work
environment
is.
I,
don't
want
you
going
to
where
it
could
be
like
we
have
to
upgrade
rip
down.
Everything
so
definitely
take
that
into
consideration,
but
yeah
I
mean
one
of
the
one
of
the
problems
of
kubernetes
is
you
have
to
you
have
to
stick.
A
C
A
C
A
So
what
about
that
answer?
Matt,
but
hopefully
that
will
get
you
the
information
that
you
needed.
Everyone
think
good
thoughts
for
Matt
all
right.
Moving
on
Ameno,
lass,
hi,
y'all,
I'm
new
in
the
channel.
Welcome,
there's
two
questions.
They
want
to
ask
so
we'll
go
with
the
first.
Have
you
ever
used
metal
lb
to
externally
expose
a
cube
service?
A
C
C
A
A
So
we
could
move
on.
Does
anyone
have
any
experience
with
Samba
+
Active
Directory
in
a
pod?
Do
you
know
if
it's
as
possible
I
can't
find
any
documentation
on
this?
Is
there
a
way
to
reset
from
scratch
the
IP
tables
rule
on
worker
nodes?
Even
if
I
don't
have
an
IP
table
safe
backup,
it
seems
that
in
one
of
my
clusters,
I
have
some
missing
IP
tables
rules
for
kubernetes
I've
deleted
all
the
key
proxy
pods
recreate
the
correct
IP
tables
rules,
but
it
seems
that
this
is
not
enough.
Let's
see.
F
So
I
don't
have
anything
on
the
is
er
I'm.
Sorry,
these
Samba
ad
in
a
pod,
I'm
gonna
highly
recommend
you
look
into
just
using
Azure
as
they
have
this
all
cloud
set
up
and
easy
to
use
on
IP
tables.
I
I
mean
you
could
definitely
flush
rules,
but
when
you're
starting
to
do
things
like
that
you're
starting
to
have
issues
like
you're
talking
about
something
is
just
not
right.
So
I
urge
you
to
take
a
kind
of
10,000
foot
step
back
and
look
at
how
things
are
being
deployed.
F
Usually,
q
proxy
operates
well
on
its
own
enough
that
you
shouldn't
have
to
go
in
manually
to
night
Patil
as
rules
I
haven't
done
a
bare-metal
deployment
in
a
while.
So
I
could
be
wrong
on
that.
Maybe
there's
some
extra
foo
you
need
to
need
to
do,
but
yeah,
usually
when
you
have
to
manually
care
about
IP
tables,
is
when
there
is
some
some
pretty
core
problem
to
to
the
deployment
so
mm-hm.
C
A
A
Let's
keep
those
open
there.
We
can
come
back
to
them
or
if
anybody
has
an
idea,
but
moving
on
to
Ravi
Shankar
welcome.
They
have
a
question
says
we
have
a
large
covered
IDs
cluster
with
400
plus
computers
on
GCE
self-manage,
with
calico
overlay,
recently
we're
observing
problems
with
calico
mesh
creation
delays.
What's
the
suggested
approach,
here's
calico
route,
reflector
the
right
approach,
I.
F
F
You
have
so
many
hosts
that
they
just
haven't
really
seen
a
use
case
like
that,
and
maybe
you
found
some
bugs
or
something
like
that,
but
yeah
I
can't
defend
we
say
yea
or
nay.
If
there's
like,
yes,
you
should
be
using
it,
no
there's
something
better,
etc.
So
it's
awesome
is
400
considered
a
lot
for
a
network
overlay.
C
The
one
thing
or
you
are
using
was
it
canal
or
what's
the
definable
elico
thing
its
canal
canal
is
the
in
in
general
like
if,
if
you
can
avoid
an
overlay
network
that
comes
to
be
better.
A
A
Okay,
unfortunately,
it
appears
the
Calico
is
not
on
the
kubernetes
slack,
but
Chris
has
dropped
a
link
to
the
project.
Calico
slack
so
hopefully
they'll
be
able
to
help
you
out.
Sorry
about
that.
Back
to
Manuel's
question
says
yeah
regarding
Samba
I
cannot
use
Azure
because
they're
doing
on-prem.
Is
there
any
way
to
do
it
in
an
on-premise
environment?
A
Just
a
quick
question
in
manual
is
Samba
a
requirement,
or
do
you
just
need
Active
Directory
I'm
wondering
we
can
ask
our
friends
over
its
sick
windows.
If
that's
an
option
for
you
and
then
minifigures
wait.
The
office
iris
folks
haven't
used
calico.
What
do
people
do
for
Network
policy,
quick
round
of
cni
opinions?
Go
jo
Bob,
Chris,
Jeff
Mario,
using.
C
E
F
I
also
prefer
psyllium
I
think
what
they're
doing
for
the
community
and
some
of
the
the
problems
they're
solving
are
outright
just
more
thoughtful
than
some
of
the
other
CN
eyes.
They're,
not
just
a
basic,
you
know,
install
the
C
and
I
can
go.
They
really
put
thought
into
other
problems
that
modern
enterprises
are
facing
and
it's
really
a
powerful
see
and
I,
like
Bob,
said,
ebps,
etc.
So,
right
now
we're
using
the
a
SS,
C
and
I
and
I'm
looking
to
move
us
to
psyllium
and
get
all
the
the
fancy
features
nicely.
So
awesome.
A
Awesome
and
audience
if
you
have
strong
opinions
on
CNA
feel
free
to
plop
them
in
there
Emanuel
goes
on
to
say:
let's
see,
looks
like
matt
has
some
experience
with
metal
and
be
an
external
IDs
that
might
be
able
to
help.
My
name
is
to
use
a
Saba
pod
and
join
a
domain
controller
from
it.
Ok,
someone
is
not
an
option
all
right.
Let's
keep
that
one
open.
Then,
while
we
move
over
to
Brendan's
question
I'm,
sorry,
we
couldn't
help
you
out
as
much
there.
A
F
Usually,
how
long
so,
what
I
usually
run
with
the
debug
dry
run
flags
before
I?
Do
anything,
and
usually
that
gives
me
there's
also
a
homeland
command
that
I
that
we
used,
especially
in
our
pipelines.
That
usually
gives
me
like
somewhere
in
the
Yamma,
where,
like
I'm
being
stupid
and
then
I'm
like
oh
yeah,
that
doesn't
look
right
and
I
can
usually
find
what's
going
on
there
other
than
that
I'm,
not
sure
on
anything
more
advanced
than
that
I
know.
F
There's
the
I
forget
what
the
channel
is
in
slack
I
used
to
be
in
it
all
the
time
there
was
a
home
users
Channel
there.
It
is
so
you
could
also
ask
them
there.
If
you
do
there,
just
isn't
making
sense,
but
in
the
init
for
the
most
part,
it's
generally
something
in
the
template,
where
you're
trying
to
use
a
variable
and
the
type
is
wrong
or
there's
a
misplace
character,
etc.
That's
those
tend
to
be
the
errors
that
I
seem
to
find
so
mostly
because
again,
I'm,
just
stupid,
yeah.
D
F
A
F
So
there's
a
home
template
command
that
can
build
the
manifest,
and
then
you
can
type
that
into
cube
eval
and
you
probably
want
to
use
it
with
a
strict
option
as
well.
I
haven't
looked
at
it
in
a
while.
It's
it's
not
tied
to
Q
Bernays.
Officially,
it's
just
our
open
source
project
from
from
someone
in
the
community,
but
it's.
A
F
Yeah,
we
actually
do
that
in
our
pipeline
as
well.
We
couldn't
use
strict
because
it
didn't
recognize
the
latest
API
for
like
HPA
or
something
this
was
months
ago,
but
I
definitely
would
run
it
through
there
because
of
helm
thinks
everything
is
fine,
it's
not
counter,
then
it's
something
kubernetes
side
and-
and
that's
so
thank
you
for
mentioning
I
should
have
yeah
just
said
something,
so
that
is.
A
Good
to
know
part
of
the
reason
we
do
this
is
we
can
learn
about
new
tools
and
those
of
you
listening
in
will
grab
all
the
URLs
that
are
mentioned
during
the
show
and
whack
him
in
the
show
notes.
He
goes
how
manifest
was
also
something
I
was
unaware
of.
These
are
definitely
some
new
options
to
dig
into
awesome.
A
Whoo,
we
needed
a
win.
We
had
some
tough
questions
today,
so
hopefully
the
hopefully
that'll
help
you
out.
Let
us
know
Brendon
how
that
goes
all
right.
Moving
on,
we
are
we're
about
halfway
through
those
of
you
joining
us
on
YouTube
welcome.
This
is
the
kubernetes
office
hours.
Ask
your
questions
at
slack,
best
hash
office
hours
on
the
live
stream.
A
Follow
the
instructions
below
and
you'll,
be
able
to
join
the
channel
feel
free
to
ask
your
questions:
will
queue
them
up
and
then,
if
we
ask
your
question,
we
will
put
you
in
a
raffle
to
give
you
a
t-shirt
at
the
end
of
the
show
thanks
everyone
for
joining
us.
Ok,
moving
on
the
next
questions
from
Demetri.
C
One
sec:
sorry,
we
missed
a
second
print
of
a
question.
We
did
what's.
A
Okay,
they're
just
pointing
out,
we
missed
a
section
question.
Okay,
all
right!
Let
me
reread
this
okay,
so
Ravi
said
we
had
a
large
kubernetes.
Cluster
is
400
million
self-managed
with
calico
and
now
they're
getting
delays
with
mesh
creation,
and
then
they
go
on
to
say.
What's
the
suggested
approach
here
is
calico
route
reflector
the
right
approach.
C
A
I
go
okay,
so
calico
rut,
reflector
was
last
updated
many
months
back,
not
sure
if
this
maintained
anymore
also,
how
do
we
this
in
a
live
cluster?
Secondly,
we
see
delays
and
deem
except
odds
getting
created
on
newly
joined
minions
via
cube
admin,
specifically
cute
proxy,
in
calico,
no
there's
no
sed
latency
is
observed
or
in
the
cube
control
plane.
How
do
we
debug
this.
C
C
Think
what
the
fault
sink
intervals,
is
it
a
minute
or
two
minutes
knows
checking
in
on
what
they
should
be
running
like
what's
sort
of
delay?
Are
you
like
how
how
big
is
it
the
way
that
you're
seeing
because
you
can
also
check
it
like?
Is
it
how
long
does
it
take
for
it
to
pull
the
pod
or
the
image
and
then
no
sync,
all
that
stuff
looks.
A
C
C
A
C
A
F
I
would
want
to
deal
with
her
have
on
my
memory
when
I
on
my
head,
when
I'm
trying
to
do
an
upgrade
or
rule
a
new
demon
set
to
update
something
or
tune
the
cni
or
at
policies
etcetera,
so
I
yeah,
we
actually
here,
we
don't
have
a
ton
of
the
applications,
but
we
we
separate
like
front
end
versus
back
end,
I'm,
actually
thinking
I
think
creating
another
one
as
well,
because
backends
getting
pretty
hefty.
So
it's
not
really
number
of
nodes
per
se.
It's
more
a
number
of
services.
For
me,
okay,.
A
Number
sir
yeah
yeah
good
to
think
about
all
right.
Bob
is
typing
with
Ravi
now
we'll
let
them
we'll
let
them
keep
swapping
info
and
then
we'll
get
back
to
that.
Let's
bust
up
Dimitri's
question
here:
real
quick,
Bob,
we'll
just
keep
typing
the
media
says:
hey
guys.
I
have
a
three
node
cluster
running
on
e
KS.
Each
node
is
on
a
different
zone,
a
b
and
c
from
my
understanding
of
pod
widgets
PVC
and
a
cannot
run
in
Zone
B
due
to
the
PVC
being
in
a
is
this
correct.
F
So
I,
actually
don't
do
any.
This
is
I,
feel
really
stupid.
I,
don't
really
do
any
stateful
workloads
across
zones,
but
co-worker
of
mine
brought
for
eks
at
least
has
done
this,
and-
and
that
is
the
case
where
you
have
to
have
an
oh
group
per
zone
and
then
you
schedule
a
workload
and
that
workload
does
need
to
stay
in
the
node
group,
where
the
zone
that
the
PC
exists
and
there
really
isn't
a
workaround
I'm,
not
sure
on
the
power
of
your
EBS
or
EFS,
whatever
you're,
using
in
terms
of
replication
across
zones.
F
I'm
not
that'd,
be
an
8
OBS
construct,
I,
don't
I,
just
don't
know,
because
I
haven't
any
need
to
use
that,
but
I
presume
that
most
clouds
now
are
probably
freighting.
Some
other
way
of
providing
the
storage
across
zones,
especially
since
zones
are
meant
to
be
incredibly
fast
between
them
themselves.
So
I
would
say:
look
at
what
a
Tobias
offers
maybe
talk
to
an
AWS
rep
if
you're
running
eks,
they
should
be
able
to
support
you
in
that
realm.
But
I
know
from
an
e
KS
like
deployment
standpoint
using
u
KS
CTL.
B
I
think
it
depends
on
what
backs
your
persistent
volumes
as
well,
because
if
you
use
EBS,
then
it
is
attached
to
a
zone
because
all
the
disks
are
directly
attached
to
the
nodes
and
you
end
up
in
the
same
zone.
If
you
use
something
like
EFS.
Instead,
it's
slower
you've
also
got
burst
in
capabilities
like
you
can
provision
that
now
the
EFS
isn't
so
much
so
we
use
EFS
for
our
previous
storage
and
that
works
ish
for
us
and
it
can
run
on
any
note
and
any
zone.
But
we
don't
use
eks.
F
True
freak-ass,
but
it
might
be.
Ok,
I
should
just
shut
my
mouth
and
let
Joel
talk
because
I
don't
know
what
I'm
talking
about
Thank
You
Joel
yeah
ETS
should
be
roughly
the
same.
I
think
so
and
then
I
know.
Ets
CTL
has
some
strapping
in
there
for
defining
volumes
and
whatnot
when
you
deploy
cluster
and
things
so
yeah
I
would
definitely
look
at
the
documentation
there,
but
that
should
be
possible
depending
on
your
needs.
You
could
probably
go
to
DFS.
Cuz
EFS
is
just
effectively
NFS
I.
Think
so.
A
A
C
A
Sure
I
think
it
might
help
with
you
know:
I
hate
to
go
down
this
road
of
like
what
are
you
trying
to
accomplish
here,
but
you
could
give
us
a
use
case.
It
might
be
able
to
to
figure
out
exactly
how
to
get
to
your
end
goal.
There
FC
would
like
to
add
on
the
we've
supporting
multicast,
that
data
will
do.
Membership
registration
and
people
still
talk
about
EFS.
A
That
would
be
fantastic
Jeff
if
you
could
pop
over
into
kubernetes
users
and
let
them
know
that
we
are
going,
you
could
use
more
questions,
that'd
be
great
or
if
anybody
has
a
follow
up,
feel
free
to
keep
on
asking.
While
we
figure
out
manage
NFS,
whether
it's
fun
or
not,
I
do
have
a
question
about
the
the
zones
that
the
mitri
was
asking
about
there,
with
a
PVC
being
in
one
zone.
A
F
B
A
A
A
F
F
Could
be
safe
with
using
EFS
or
at
least
try
it
for
his
workload
and
see
what
happens
so
I
think
that
one
we
can
probably
close
the
case
on
that
I
know.
There's
people
doing
that
already
too.
So
there
should
be
a
plenty
of
Docs
and
support,
probably
a
file
provider
yeah.
B
One
final
tip
I
put
on
that
is
when
you
create
the
EFS
I
want
to
say
bucket
I,
don't
think
they
call
it
bucket,
then
try
and
look
at
provisioning,
your
iOS.
It
used
to
be
kind
of
like
burst
based
and
although
it
was
it
used
to
be
proportional
to
how
much
data
you
had
stored
in
it.
So
we
needed
more
fruit
put
than
you
could
offer.
So
we
just
wrote
a
bunch
of
zeros
to
fill
up
some
space
to
get
more
frequent,
terrible
idea.
F
B
F
Is
a
hard
problem,
so
yeah,
but
I
would
definitely
look
at
EFS
if
I
was
going
to
have
to
use
something
because
I
really
like
cross
zone
separation
and
redundancy.
Therefore
you
know
failover.
It
would
not
sure.
A
A
C
B
I
was
just
gonna
say.
We
are
well
aware
that
we're
not
meant
to
be
doing
this,
but
it
does
seem
to
work
for
us
so
like
we
haven't
really
seen
any
problems
with
it.
Occasionally
we
get
a
failed
compaction,
but
for
the
most
part
it's
been
fine,
but
I
think
it
really
depends
on
how
how
scale
how
much
scale
you've
got
there
our
biggest
clusters.
B
B
C
A
A
A
A
Yes,
this
got
complicated
quickly,
so
all
right,
hopefully
Dimitri-
that
gives
you
a
bunch
of
options,
feel
free
to
hang
out
here.
Thank
that
goes
for
everybody
feel
free
to
just
keep
on
hanging
out
in
the
channel
throughout
the
month
and
how
you
get
on,
because
we
do
love
to
see
when
people
solve
their
problems.
E
A
A
That
is
good
one
and
we
will
definitely
make
sure
we
get
all
these
URLs
in
the
show
notes.
Cuz,
that's
a
good
room,
wow
there's
a
lot
of
places
you
can
remote,
send
Prometheus
stuff
I've.
Never
heard
of
some
of
these.
That's
awesome,
alright,
awesome
to
meet
you
I
hope
that
gives
you
some
options.
They're
checking
in
any
word
on
Ravi's
problem.
Here,
oh,
we
were
talking
about
it
in
a
threat.
Okay,
all
right,
I
will
come
back
to
you.
Then.
Let
me
look
here.
We
have
some
new
questions
here.
A
One
second
minifig
has
a
question,
says
I'd
love
to
hear
opinions
about
ipv6
last
year
before
dual
stack.
Now
that
it
is
out
of
alpha-I,
will
you
just
give
everyone?
A
quick
warning
here:
I
am
going
to
try
to
get
a
cig
and
network
specific
office
hours
on
the
calendar
for
next
year;
I'm
kind
of
shopping
it
around
to
the
cig,
Network
folks,
so
stay
in
tuned
for
that,
but
quick
opinions
on
dual
stack:
anybody
anybody
looking
into
this
well
I
grab
other
questions.
A
F
A
F
F
It's
awesome
I.
We
have
no
use
case
at
all
for
it,
but
we
were
a
very
different
type
of
needs
and
whatnot.
So
the
basic
stuff
gets
us
by
I
think
for
the
more
advanced
deployments
and
as
kubernetes
starts
to
penetrate
enterprises.
I
think
it's
gonna
become
something
that's
pivotal
and
I'm
really
glad
to
see
that
the
work
has
been
done
to
get
us
there.
So.
A
Did
did
any
of
you
see
Hawkins
talk
on
the
dual
stack
at
cube.
Con
I
am
going
to
find
that
talk
and
stick
it
in
the
show
notes.
While
you
all
answer
this
next
question
from
DevOps
day
manager,
I,
guess
that's
what
that
means.
I
have
1,000
Q
of
work
in
my
right,
a
system
to
assign
parallel
work
queue
which
is
working
only
once
after
complete
all
pods
in
completed
state
after
some
time
assigned
queue
to
Redis,
which
is
not
taking
my
work.
C
A
E
A
A
E
Thing
as
usual,
it
depends
based
on
the
behavior
that
you're
describing
so
the
example
is
a
bunch
of
kubernetes
batch
jobs
that
are
not
actually
so
the
documentation
is
wrong.
That's
actually
the
first
problem.
The
documentation
is
showing
an
example
of
a
bunch
of
persistent
pods
that
are
popping
tasks
off
of
a
stack
and
then
running
them,
but
they
are
keeping
the
pods
alive.
A
E
Realistically,
what
you
should
be
doing
is
having
like
your
work,
queue
as
needed,
whether
that's
you
know
in
Redis
or
whether
that's
you
actually
creating
individual
batch
jobs
within
kubernetes
and
having
it
manage
that.
That's
that's
up
to
you,
but
the
example
is
wrong
because
it
is
actually
showing
jobs
as
a
persistent
thing
that
is
popping
things
off
a
rhetta's
queue
and
executing
them,
but
still
staying
alive,
whereas
jobs
should
be
one-off
tasks,
not
tasks
compromises.
A
A
A
All
right
so
bob
is
the
raffle
totally
guaranteed
to
be
random
and
scientific
I
want
to
thank
all
of
you
that
listen
today,
217
of
you
tuned
in
that
is
by
far
a
world
record
for
us.
So
thank
you
very
much
for
participating.
The
winners
are
Dmitry
and
DevOps
day.
M
we're
I,
don't
know
how
to
pronounce
a
DevOps
day
manager.
I,
don't
sure
I
will
PM
you
after
this
and
give
you
a
code
to
the
CNC
F
store.
So
you
can
have
your
kubernetes
t-shirt.
A
Anything
else,
any
last
comments
from
the
audience
or
the
panel.
Before
we
wrap
it
up.
We
will
be
back
at
the
new
year
as
always
the
third
Wednesday
of
the
month.
If
you're
interested
in
contributing
to
kubernetes,
we
have
a
sister
program
called
meet
our
contributors,
which
is
more
about
asking
questions
on
how
you
do
a
pull
request.
How
do
you
get
your
changes
into
kubernetes
that
sort
of
thing,
so
we
highly
recommend
you
browse
the
kubernetes
youtube
channel.
This
is
posted
on
and
check
out
some
of
our
other
programs,
including
mentorship
and
whatnot.
A
As
we
said
before,
thanks
everyone
for
coming.
This
is
an
all-volunteer
panel.
If
you
want
to
sit
on
this
panel
and
help
rotate
out
some
of
these
people
here,
so
we
can
get
more
diverse
opinions
on
it.
We
would
really
appreciate
it
yeah
and,
as
he
said,
it
kind
of
sucks
that
the
last
question
of
2019
goes
unanswered,
bring
it
on
in
2020,
so
I
think
what
I
will
do
is
I
will
actually
tweet.
A
Let's
do
this,
we'll
tweet
for
opinions
on
customize
on
Twitter
and
I'll,
get
a
bunch
of
responses
from
a
bunch
of
people
and
then
we'll
stick
that
in
the
show,
notes
and
we'll
just
leverage
the
power
of
the
community,
because
the
five
of
us
can't
can't
figure
that
out.
This
is
hey.
Will
that
work
for
everybody
or
what.
E
A
Kubernetes
memory
yeah:
what
do
I
have
to
go?
First,
Oh
nuts,
probably
cube
con
Kelsey's
keynote
Kelsey's
keynote
this
year
at
San
Diego.
Those
of
you
that
have
not
seen
that
it
is
in
that
playlist
I
put
absolutely
I'm
gonna
go
with
that
yeah
Troy
says
cube
Khan,
but
yeah.
D
F
D
A
C
E
B
A
That's
really
good,
that's
really
knitting!
Well
thanks
everybody!
This
has
been
the
most
successful
show
of
the
year.
So
far
viewership
wise,
we
will
have
happy
holidays
and
you
celebrate
and
we
will
see
everyone
in
a
good
year.
Good
luck,
everybody
with
your
clusters
that
if
we
need
an
answer,
your
question
please
give
us
a
shot
again,
feel
free
to
hang
out
in
the
channel
feel
free
to.
Let
us
know
how
it
goes.