►
From YouTube: Kubernetes Office Hours 20190918
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.
Discussion thread for this episode: https://discuss.kubernetes.io/t/office-hours-for-september-2019/7944
For more info: https://github.com/kubernetes/community/blob/master/events/office-hours.md
A
A
D
A
A
The
mute
buttons
at
the
bottom
of
your
UI
in
the
zoom
window,
alright.
Well,
you
get
your
zoom
I'll
go
through
the
intro
and
then
we'll
come
back
to
you
here.
Alright,
so
welcome
everybody.
Let's
start,
let's
start
with
some
ground
rules.
Before
we
start.
First
of
all,
you
can
find
us
on
hash.
Offish
hours
on
slacks,
say
hello.
When
you're
from
there
we
see
people
from
Michigan,
Tel,
Aviv,
Netherlands,
Kazakhstan
awesome.
So
what
we
do
is
during
a
live
stream.
A
We
put
the
chat
on
the
side
there,
so
you
could
participate
when
kind
of
have
a
lot
of
fun
here
together,
but
before
we
start
there's
a
few
ground
rules.
So
this
is
a
judgment-free
zone.
Everybody
had
to
start
from
somewhere.
There
are
no
done
questions,
so
please
help
someone
out
when
they
are
and
be
supportive
of
them
when
they
are
asking
a
question,
and
while
we
will
do
our
best
to
answer
your
questions,
the
panel
doesn't
realize
access
to
your
cluster,
so
live
debugging
is
off-topic.
A
That
goes
for
the
audience
as
well
love
to
have
all
the
experts
in
the
audience,
sharing
their
experiences
and
helping
people
out
that
are
getting
started.
We
always
encourage
people
to
do
that.
Audience.
You're
also
can
help
us
by
pacing
URLs
to
documentation
or
when
we
mention
things
finding
the
github
URL
whacking
it
in
their
blog
posts
about
related
topics.
A
Anything
that
you
think
can
help
on
the
audience
feel
free
to
just
stick
it
in
the
slack
channel,
while
we're
in
the
middle
of
talking
and
what
I
do
is
at
the
end
of
the
show,
we'll
grab
all
the
URLs
and
put
it
together
in
the
show
notes,
so
that
we
can
publish
those
to
the
community
and
then
people
can
check
out
all
the
stuff
we're
talking
about.
There
is
a
discuss
thread
for
this,
which
bob
will
be
fishing
out
and
putting
it
to
the
select
Channel.
A
You
can
also
help
us
out
by
tweeting
spreading
the
word
paying
it
forward,
letting
someone
know
that,
like
he
came
here
and
got
some
help,
anything
to
help
grow,
the
audience
will
be
really
appreciated
and
if
you're
an
expert-
and
you
want
to
give
back
to
the
community,
you
can
always
just
sit
on
the
panel
for
an
hour
once
a
month.
It's
fun:
that's
where
a
puddle
s
came
from
he's
just
hanging
out.
He
didn't
like
my
answer,
so
he
was
like
I'm
gonna.
A
Do
myself
no
just
kidding,
we
love
to
have
a
lot
of
people
around
and
a
lot
of
the
companies
that
these
people
work
for.
Do,
allow
them
to
come
on
here
and
grant
them
an
hour
a
month,
so
a
big
shot
at
the
Giants
warm
stock,
X
packet,
pusher,
calm,
we've
works,
VMware,
lyft,
University
of
Michigan
Red,
Hat
utility
warehouse
and
the
City
of
Ottawa
Ontario
Canada
for
letting
their
engineers
come
and
help
out
the
community,
and
always
a
special
thanks
to
the
CNC
F
for
sponsoring
a
t-shirt
giveaway.
A
So
the
way
it
works
is,
if
you
ask
a
question,
we
will
put
you
in
automatically
for
a
raffle
and
if
you
read
your
question,
live
on
air.
What
we'll
do
is
at
the
end
of
the
show
we
will
raffle
away
at
each
two
t-shirts,
two
to
you,
so
you
have
to
stay
till
the
end
for
the
raffle,
but
keep
on
asking
your
questions
and
we
will
get.
We
will
get
to
it
all
right
with
that
panel.
We
ready
any
last
minute.
A
Comments,
looks
like
the
chat
is
hopping
today,
so
it's
gonna
be
a
good
one.
I
can
already
tell
all
right
the
first
question
so
I'm
those
of
you
listening
just
keep
posting
your
questions
and
we'll
put
them
in
a
little
notes
document.
If,
after
a
while,
it's
obvious
that
we
either
missed
your
question
or
or
something
like
that,
feel
free
to
repeat
it.
If
we
answer
your
question,
feel
free
to
post
up,
follow
up
and
then
we'll
go
to
the
next
question,
then
we'll
come
back
to
you.
A
If
you
need
a
more
detail
or
anything
like
that,
okay,
so
it's
six
minutes
after
let's
get
started,
Marko's
says
hello.
I'm
from
Spain
I
see
that
customized
with
a
k'
and
helm
address
similar
promise
to
template.
Kubernetes
definitions,
any
insights
on
the
best
tool
to
describe
complex
kubernetes
applications.
That's
a
good
one
to
start
off
with.
D
Actually
was
yeah,
I
was
kind
of
trying
to
think
of
what
to
say
here.
This
is
really
a
hard
one.
Home
can
be
tedious.
I
think
customized
makes
things
easier,
but
it
doesn't
solve
all
the
problems.
You're
not
gonna,
find
a
hundred
percent
perfect
solution.
I
would
say,
try
them
both
with
with
an
application
that
you're
using,
and
we
kind
of
did
that
a
little
bit
and
we
decided
that
home
was
just
kind
of
made
the
most
sense,
especially
from
a
Seavey
perspective,
continuous
delivery
perspective.
D
So
yeah
I
mean
you
say,
complex
Bernays
applications
I
mean
yeah,
as
you
start
getting
more
complex
databases
and
and
CR
DS
and
other
resources
using
doing
cron
jobs,
and
things
like
that
I
think
for
me,
hell
makes
it
pretty
easy,
because
I
can
use
kind
of
a
default
Tundra
for
what
a
cron
job
might
look
like
and
and
as
a
template.
D
It's
easy
to
just
kind
of
drop
that
into
a
chart
for
a
particular
service
and
then
just
kind
of
provide
the
variables
that
are
needed
through
the
values,
file
and
so
kind
of
the
initial
zero-to-sixty
of
basic
service
dockerfile.
Getting
at
to
helm
is
difficult,
but
once
you're
there
it
makes
you
know
adding
other
objects
that
you
need
for
the
application
a
little
bit
easier.
So
an
example.
This
is
an
application.
We
just
did
that.
D
We
actually
need
necess
h
tunnel
for
whatever
reason
to
something
else,
and
that
was
really
easy
to
just
deploy
a
very
simple
deployment,
and
then
you
know
so
the
domain
for
how
the
other
services
contacted,
etc.
So
yeah,
there's
customizers
like
three
other
ones
that
are
picking
up
steam
and
they
all
have.
They
all
solve
the
problem
differently.
So
yeah,
definitely
just
try
things
out
in
a
very
minimal
capacity
would
be
my
suggestion
there.
The.
B
Other
thing
is,
like
people
actually
use
them
together
too,
it's
there
isn't
a
single
best
solution
out
there,
and
a
lot
of
people
are
sort
of
revisiting,
which
one
is
best
right.
Now,
with
all
the
changes
come
three
I
can
say
like
if
you
have
more
complex
logic
and
things
like
that,
helm
is
going
to
be
better
than
customized.
A
Moving
on
Christian
asks
hi,
guys
I'm,
currently
migrating
an
application
from
docker
compose
to
kubernetes.
My
problem
is
that
I
have
different
algorithms
to
run
that
each
run
on
a
pod
and
they
should
analyze
the
same
image
in
dr.
kapositas,
same
storage
to
all
of
them.
How
should
I
accomplish
the
same
thing
in
kubernetes
I
read
that
it
is
not
a
good
thing
to
have
a
global
shared
space
in
kubernetes,
so
we
wanted
this
one.
Yes,
I.
C
Did
so
if
you're
running
on
something
like
AWS,
you
could
have
like
a
shared
EBS
volume
across
everything,
but
performance
wise.
That's
not
good
and
actually
that's
kind
of
an
anti-pattern
when
it
comes
to
doing
analysis
like
this,
typically
in
which
you'll
see
in
the
HPC
space,
which
does
a
lot
of
analysis
like
this
is
you'll
actually
have
a
which
we
call
it
in
an
it
container
or
like
a
job
before
the
job
that
will
download
the
image
to
wherever
you
are
executing
your
algorithm
and
then
run
it
locally.
So
it's
just
running
in
scratch.
C
Space
if
you
are
running
large,
wide
jobs
that
makes
the
most
spent
most
sense,
especially
since
you
don't
want
say
a
hundred
pods
hitting
the
same
EBS
volume.
But
if
a
hundred
pods
download
that
image
and
then
do
the
analysis
across
a
hundred
nodes,
you
know
it's
much
more
performant.
So
in
that
case,
I
would
just
have
sort
of
an
init
container,
an
ephemeral
container
that
will
download
the
image
and
do
any
pre-processing
before
the
algorithm
runs.
So
so.
D
D
There's
a
lot
of
factors
there,
but
if
it's
all
the
same
data
that
you
just
need
to
analyze
in
different
ways,
then
keeping
the
data
the
same
I'm,
providing
an
entry
point
that
maybe
script
that
can
you
know
do
something
different
with
that
data
could
also
make
it
easier.
You
don't
exactly
need
to
have
some
sort
of
global
storage
back
lane.
B
D
You,
but
might
add,
it
really
depends
on
the
nature
of
what
you're
trying
to
do
so
I've
seen
people
trying
to
fight
like.
Oh,
let
me
get
the
data
in
some
other
place
and
all
that
when
it's
you
know
a
gig
of
data,
maybe
or
something
or
it's.
You
know
it's
not
a
static
data
set.
It's
it's
just
a
little
bits
and
pieces
they
can
pull
in
and
out.
So
it
depends
on
the
situation.
A
A
Thanks
for
joining
I've
been
able
to
make
a
very
simple
proof
of
concept,
image
a
neck
console
that
that
continuously
outputs
texts
running
locally
and
push
it
to
an
artifact
terrine
repository
I've
already
got
my
kubernetes
doc
cube
configured
with
my
secrets,
and
such
I
was
told
if
I
can
successfully
get
no
resources
found
as
a
response
to
cute
cuddle,
get
pods
and
then
I
was
set
up
properly.
So
at
this
point,
I'm
hoping
to
get
some
help
on
how
to
create
a
kubernetes
cluster,
some
pods
and
deploy
my
proof
of
concept
that
to
it.
D
He
doesn't
really,
it
sounds
like
he
hasn't
really
actually
spawned
a
cluster
as
any
starting
point
about
he's,
just
kind
of
starting
from
from
zero.
Okay
here
but
yeah.
This
is
kind
of
a
wide
question.
Oh,
it
really
depends
on
what
he's
trying
to
do.
He
could
be
cloud
might
make
the
most
sense,
but
yeah
definitely
like
make
sure
the
it
sounds
like
locally
about
your
choreography
ironed
out
and
yeah.
It
looks
like
he
has
a
you.
Have
a
cluster
book
yeah.
B
You
written
deployment
or
anything
like
that
yet
or
just
you
just
have
the
image.
E
F
E
Could
the
nether
subject,
which
will
create
basically
launcher
application?
You
should
also
look
into
pods,
so
yeah,
basically
at
the
beginning,
I
would
just
like
try
to
run
a
simple
pod
with
your
image.
Just
see
it
in
kubernetes
take
a
look
at
logs,
whether
you
actually
get
the
same
exact
thing:
mm-hmm
I
guess
we
can
link
to
a
couple
of
top
aerials
yeah.
A
Bobby's
got
the
link.
I
was
searching
for
deployments
as
well.
I'm
gonna
expand
on
this
question
a
little
bit
here.
When
you
all.
Let's
say
you
have
your
container
and
you're
gonna.
Do
your
service
do
y'all
start
with
a
deployment
or
do
you
sit
down
and
say
you
know,
deployment
replica
set
pod
like
do
you
do
all
of
them
at
once,
or
do
you
kind
of
just
start
with
a
deployment
and
the.
B
A
B
F
B
A
E
Yeah
more
like,
if
you
have
like
a
thing
which
actually
needs
to
run
until
like
completion,
you
should
start
with
a
job
instead
of
a
deployment,
because
jobs
are
actually
the
things
which,
like
run
once
and
runs
until
completion,
and
it's
done
and
you
can
see
by
logs
in
in
kabedon,
Epis
mmm-hmm,
so
yeah,
okay,.
A
So
the
ployment
looks
like
the
place
to
start
that
should
get
you
started.
He
says,
eventual
gonna
be
a
web.
App
to
spruce
is
just
to
get
everything
as
far
as
running
in
the
kubernetes
cluster,
so
yeah.
So
then,
so
then
what
how's
the
work
flow
I
just
want
to
talk
generally
here
to
be
useful
for
more
people.
So
you
do
your
deployment
at
some
point.
He
doesn't
service
those.
Are
you
using
Helmand
stuff?
Do
you
still
go
in
that
direction,
or
do
you
start
off
with
like?
Let
me
find
a
hound
chart.
D
So
I
always
find
a
helm
shart,
usually
something
to
base
off
of
or
I
might
have
my
own
or
like
another
project
most
of
the
times.
I
do
something
else.
I've
done
previously
that
I'm
very
intimate
with
that
I'd
configured
almost
from
scratch
for
myself
or
some
need
I
had
and
then
I
like
to
kind
of
take
that
and
run
with
it.
Because
again
it
has
those
templates
already
kind
of
in
place
for
the
base
model.
I
need
to
change
a
couple
things.
It
just
makes
it
easier
to
get
going
and
get
something
out
quicker.
A
D
A
For
your
simple
laughs,
I
think
it
makes
sense
to
start
with
the
deployment,
because
you
need
to
know
that
stuff
right,
like
that's
kind
of
the
thing,
looks
like
a
lot
of
people
are
sharing
their
documentation,
kershaw
how
you
doin,
he
shared
his
looks
like
a
telegram
kept
Shabbat
little
service.
That
should
be
useful
and
then,
of
course,
Bob
and
Jeff's.
Kubernetes
intro
tutorials
always
have
good
information
like
this.
Okay,
so
start
with
deployments
and
services
that
you
get
you,
but
that's
plenty
of
homework
there
for
you
uh-huh
to
get
started
all
right
anything
else.
A
Any
other
tips
in
check
form
before
we
move
on
all
right.
Awesome
thanks
for
joining
back
sad-ass
hi.
My
question
in
the
office
hours
is:
why
is
it
bet
a
bad
or
good
idea
to
run
calico
on
cluster
nodes
as
a
systemd
service,
and
when
we
will
use
kubernetes
cluster
auto-scaling?
Will
it
be
painful
by
running
calico
as
systemd
on
masters
and
as
pods
on
the
worker
nodes?
A
B
D
F
So
I
I
have
seen
few
cases
that
people
run
it
as
demonstrate
having
a
system
deep
process.
In
view
of
pods,
for
example,
I've
seen
you
see
advisor
case
that
the
AWS
runs
as
a
system
D,
you
don't
have
to
install
a
support,
but
I
would
go
as
pots,
because
the
lifecycle
management
is
quite
easy
system.
D
becomes
very
you
know,
to
manage
to
update
the
config.
You
know
to
change
the
image
to
change.
The
image
version
becomes
a
very
difficult
thing,
so
I
would
advise
to
go
with
the
system
apart
mechanism.
D
F
F
A
Great
always
awesome
to
have
a
networking
person
on
and
for
those
of
you
listening.
You
can
basically
just
think
me
before
the
show,
if
you
want
to
get
on
the
panel-
and
we'll
just
add
you
on
here,
because
we
like
to
make
it
fun
so
welcome
curious
here
for
colic
I
was
just
wondering
if
there
was
an
operator
for
calico.
Is
that
a
thing.
B
F
B
E
A
A
A
B
It
I
see
I
hate
to
say
it
depends,
but
it
sort
of
depends,
like
we've
run
our
own
custom
thing
using
container
Linux
and
cube
EDM
on
top
of
that
with
some
bash
glue,
that's
mostly
like
the
the
container
looks
like
update
agent,
so
that
makes
it
easy
I
know
so
much
yeah,
and
then
you
know
there's
a
lot
of
other.
If
you're,
you
know
familiar
with
ansible
and
you
do
a
lot
of
ansible
stuff
already
keep
spray
might
be
a
good
option.
The
other
thing
is
like
I
would
definitely
advocate.
B
D
I
mean
I'm
gonna,
actually
like
okay,
so
for
a
bare-metal,
I,
I
really
don't
know.
I
really
am
not
as
well-versed
back
when
I
was
doing
eyes
on
bare
metal
like
the
one
four
one
five
around
there
like
it
was
tectonic
or
less
tectonic
made
it
really
easy,
had
a
lot
of
great
scripts
and
and
whatnot
for
fixing
booting
nodes
and
things.
D
However,
I'm
gonna
like
venture
to
ask
him,
like
maybe
think
cloud
depending
on
your
thing
like
managing
nodes,
is
a
lot
of
work
from
experience
and
you
know
gke
or
other
EES
aks,
like
they've,
got
a
lot
of
features
now
a
lot
of
things.
You
know
that
they
can
do
just
kind
of
automatic,
for
you
and
it'll
make
life
a
lot
easier.
So
I'm
definitely
thinking
about
about
possibly
doing
cloud
depending
on
your
use
case,
but
yeah.
E
Policy,
you
have
an
opinion
yeah,
so
I
think
it
mostly
like
depends
on
scale.
If
you
have
like
you
know
like
three
nodes
like
five
nodes
I'd
like
if
it's
a
super
small
cluster,
then
I
would
just
like
go
simple
way
and
use
cube.
A
DM
cube.
A
DM
has
all
the
instructions
for
upgrading
your
cluster
for
like
doing
all
the
operations
it
does
like
backups
for
you,
it
gives
you
like
certificates,
it's
it
sort
of
I
think
like
a
most
simple
way
to
start
off.
E
F
So
I
have
seen
customers
are
deployed
on
battery
meter,
one
of
the
biggest
Sunday
or
have
is
how
do
I
maintain
my
HDD?
How
do
I
maintain
my
API
service
when
I
go
to
Cloud
I?
Already
all
this
the
cube
system,
pods,
is
already
maintained
by
the
cloud.
Maintainer
and
I.
Don't
have
to
do
anything
so
is
there
a
better
way
to
do
it?
I
didn't
really
have
enhancer,
you
know
they
were
looking
for.
Somebody
like
project
Pacific
is
probably
the
right
answer
is
what
I
thought
you
know.
So
it's
a
very
open
area
there.
F
You
know
something
something
you
know:
I
don't
really
have
an
answer,
but
this
is
the
discussions
we
typically
end
up
having
with
customers.
You
know
this
is
what
their
concern
is.
How
do
I
maintain
my
CD?
How
do
I
scale
it
if
I
have
thousand
nodes
over
I
scale
it?
You
know,
you
know
those
questions
is
an
open,
an
open-ended
right
now
so
I'm
concerned.
Do.
B
Alright,
this
is
gonna
make
like
the
cluster
API.
There's
a
lot
of
efforts
going
on
to
sort
of
like
more
streamline
that
entire
experience
for
you,
I
think
one
bare
metal
thing.
It's
called
like
metal,
3
or
metal
cubed
that
we
use
like
OpenStack
ironic
to
actually
handle
the
lifecycle
of
all
the
machines
on
bare
metal.
So,
like
you,
don't
manage
to
a
saying
like
that,
it
will
take
care
of
it
for
you.
A
All
right
and
then
my
my
answer
is
basically
an
echo
ABAB
cube
admin
today,
if
I'm
on
bare
metal
and
a
cluster
API
gets
developed,
I
feel,
like
that's
gonna,
be
the
future.
Just
reminder,
though,
when
you're
doing
bare
metal
it's
when
I
did
my
first
one,
a
dick
you,
man
and
I,
have
my
cluster
up
and
running
and
everything
it
still
doesn't
work.
You
have
to
do
like
the
load,
balancer,
there's
all
the
other
stuff.
You
still
have
to
do.
So
there
really
is
no.
A
Simple
I
was
able
to
do
it,
no
problem,
but
then
there's
all
the
other
stuff
that
I
did
not
plan
for
that
kind
of.
Do
me,
and
it
sounds
like
a
lot
of
these
installers
like
are
moving
towards
using
q,
either
cube
admin
directly
or
consuming
cluster
api
in
some
ways.
So
there's
definitely
some
consolidation
there.
The.
B
Other
thing
I
just
wanted
to
touch
on
quick,
is
even
in
some
of
the
cloud
providers.
Instead
of
using
the
MER
service,
people
are
starting
to
like
actually
deploy
their
own
thing
on
because
mm-hm,
you
know
either
Google
Microsoft
or
Amazon
like
let
you
use
a
different
OTC
provider.
You
have
to
tie
it
in
and
use
like
there.
I
am
roles
or
I.
Did
there
like
sort
of
authentication
service
for
a
lot
of
this?
If
you
require
like
more
customization,
you
essentially
have
to
hand
roll
your
thing.
Mm-Hmm.
A
Yeah,
it's
like
bonus,
I,
don't
have
to
do
with
the
Masters
bummer
I.
Don't
get
control
of
my
own
masters
and
long
would
like
to
add.
Rangers
is
good.
He
likes
that
for
bare
metal
and
then
there's
some
link.
Okay
moving
on
because
the
questions
are
piling
up
now,
Thomas
Papa
sec.
Can
you
guys
see
this
link
here
to
the
CMO
file?
Because
it's
quite
long
since
I
would
like
to
react.
A
question
from
kubernetes
users
were
running
cube
1.14
on
digitalocean.
A
We
have
configured
pods
to
both
stick
to
some
nodes,
node
affinity
and
then
to
be
repelled
from
other
specific
pods
pod
anti
affinity.
However,
kubernetes
seems
to
be
ignoring
the
anti
affinity
and
then
there's
a
configuration.
I
hope
you
guys
can
see
the
configuration,
because
I
don't
want
to
read
the
mo
on
the
air.
D
So
I
actually
just
did
some
of
this
more
specifically
to
balance
workloads
better
across
availability
zones
in
ABS.
So
you
know
I
can't
exactly
say
what's
going
on
here,
just
that,
obviously
the
scheduler
is
building
a
score
for
each
node
when
it
goes
through
and
decides
where
to
place
a
workload,
and
you
do
have
both
both
of
these
as
required.
D
So
you
know
definitely
it's
trying
to
to
appease
those
I
know
for
sure.
There's
plenty
of
resources,
so
it
shouldn't
that
should
be
a
problem.
Sure
yeah,
I,
don't
exactly
know
a
you
know,
instance
type
for
node
affinity,
I.
Believe
that's
going
to
probably
take
over
again.
It
depends
on
what
he's
doing.
Yeah
I
can't
really
like
debug.
This
live
per
se.
It
looks
pretty
sane
I'd,
maybe
check
your
match.
Expressions
for
the
DNC
affinity,
rule
I've.
B
B
A
Yeah,
that
is
a
good
one.
Good
point
all
right,
so
believe
the
question
there
audience
if
you
could
help
him
or
her
out
I'm
sure
they
would
appreciate
it
all
right.
Moving
on
shy
cat
says
I'm
really
interested
about
kubernetes
dashboard
recommendations.
I
feel
that
the
official
UI
is
not
maintained,
they're
not
going
anywhere
Jeff
I
added
that
Jeff
part
any
other.
A
C
Wanna,
you
know
say
a
couple
things
as
one
of
the
chairs
of
cig
UI.
First
off
in
the
last
week,
we've
merged
over
20
PRS,
so
I
I
would
not
say
that
it's
not
maintained
it
probably
just
doesn't
have
the
velocity
of
a
smaller
project.
That
being
said,
I
do
know
of
several
other,
pretty
good
dashboard
alternatives,
one
of
them
that's
kind
of
the
new
hotness.
That's
come
out
as
VMware
is
octant.
That's
mainly
that's
not
meant
to
be
a
dashboard
that
multiple
people
consume.
C
C
That
was
completely
written
from
the
ground
up
in
what
was
it
react
instead
of
angular,
because
dashboard
is
written
in
angular
both
of
those
are
pretty
solid
and
do
show
multiple
integrations.
The
current
kubernetes
dashboard
does
have,
or
it's
about
to
have,
support
for
CR
DS,
which
might
help
when
it
comes
to
showing
AWS
specific
things.
However,
a
lot
of
the
dashboards
that
you'll
see
don't
actually
give
insights
into
the
clouds
themselves.
A
That
all
right
and
I
have
a
thread
that
there's
actually
a
few
that
I
will
paste
in
the
channel
here
once
once,
I
find
it
that
I
know
honey
did
a
long
blogpost,
comparing
all
the
different
kinds
of
dashboards
and
stuff,
and
we
talked
about
this
in
detail
last
week,
so
I
will
find
the
notes
and
publish
them.
While
you
all
answer
the
next
question,
which
is
IO
and
r
says
hey
folks,
do
you
happen
to
be
aware?
Is
ole
and
O's
cop
F,
it's
a
Python
framework
for
writing.
Kubernetes
operators.
A
20
once
going
twice:
okay,
so
I
guess
we
don't
know
a
lot
about
that.
One
I
see
an
open
shift.
Restclient
Python
is
that
the
Red
Hat
one,
let's
wait
for
looks
like
Chris
short,
is
afk.
Let's,
let's
circle
back,
when
we
seen
him
our
salmon
says:
I
looked
at
coffin,
we
adopted
another
soul
and
O'toole's
seemed
sound,
but
we
haven't
leveraged
it
for
anything.
It's
good
to
know.
A
All
right,
let's
move
on
from
this
one,
then
I'm,
sorry
about
that
wish
we
had
more
experience
than
that.
Dimitri
I
hope
I
got
that
right,
question
for
observability.
It
seems
a
lot
of
Agri
fauna
community
charge
for
kubernetes,
either
don't
work
or
only
parts
of
the
chart
work.
The
most
companies
create
their
own
church
or
am
I.
Looking
in
the
wrong
place,
great
question.
E
So
my
recommendation
would
be
to
look
into
kubernetes
monitoring,
I,
think
it's
maintained
by
Sigma
instrumentation
and
basically
it's
a
it's.
There
is
a
project
called
monitoring,
mixin
or
something
like
that,
and
it
basically
gives
you
grana
dashboards
alerts,
all
that
stuff
yeah
and
it
basically
works.
It
may
have
like
basically
release
versioning
and
they
have
a
couple
of
versions
for
like
kubernetes
related
when
1.14
and
we
have
a
version
which
works
like
prior
to
that
so
yeah.
Basically,
they
are
well
maintained
and
it'll
be
good.
A
A
Awesome
all
right.
It
looks
like
that's
plenty
of
links
to
get.
You
started
feel
free
to
ask
a
follow-up,
and
thank
you
for
listening.
Those
of
you
joining
us
we're
about
halfway
through
this
is
the
kubernetes
office
hours
hop
into
hash
office
hours
on
slack
urban
Edison
I/o,
and
ask
your
questions
something
that,
if
you're
listening
in
something
that
we
like
to
do
is
we
do
keep
the
channel
open
throughout
the
whole
month
so
feel
free
to
hang
out
in
there.
A
B
C
D
So
we
have
lots
of
labels
and
other
things
like
that
to
give
us
data.
So
when
we
troubleshoot
it
makes
it
easier
in
terms
of
what
that
thing
is,
it
runs
those
commands.
That's
again,
it
really
doesn't
matter.
I
think
folks,
get
lab.
I'll
have
a
little
bit
more
strapping
for
kubernetes.
That
might
make
it
easier,
especially
when
starting
out
so.
A
All
right,
any
other
opinions
on
this
one.
All
right,
Edward
T,
welcome
ass
I
want
to
use
the
get-ups
model
to
deploy
apps
to
a
cluster
I.
Would
one
update
the
repo
that
maintains
the
cluster
state?
Would
you
just
use
said
to
replace
the
image
tag
and
then
commit
it?
Are
there
tools
that
do
that
any
good
practices,
Thanks
yeah.
D
This
is
I'm
literally
at
the
flux,
cdio
site,
the
get-ups
operator
for
kubernetes,
and
you
know,
tools
the
auto
tool
that
automatically
ensures
the
state
of
your
kubernetes
cluster
measures,
configuration
supplied
and
get
so
that's
that's
gonna,
be
probably
something
you'll
definitely
want
to
look
at.
I
haven't
used
it
so,
but
we've
puts
out
some
great
tools
you
can
CTL
is
one
of
them.
So
now
we
trust
it
does
what
it
says
it
does.
Yeah.
B
A
A
B
A
A
A
So
yeah,
sorry
about
that,
not
as
smart
on
get
up
says:
I
should
be
alright.
Moving
on,
as
so
check
out,
hash,
get
ops
and
then
he'll
send
a
bunch
of
we
perks,
people
that
way,
and
that
will
start
from
there
mary,
I
says
hello.
How
do
you
recommend
a
deal
with
cron
jobs
during
deployments?
Currently
I
set
them
as
suspended
deployed
the
applications
new
version
and
after
that
I
apply.
A
C
D
And
I
don't
think
you
need
to
disable
them
per
se.
You
know
in
your
check,
maybe
like
we're
just
saying.
Definitely
do
that
is
important,
but
the
cron
job
itself
is
not
the
job.
It's
important
to
remember
that
a
cron
jobs
definition
for
when
to
run
a
specific
job
and
a
job
is
an
object
and
of
itself.
So
a
cron
job
is
just
kind
of
a
declaration
of
what
you
want
to
run.
So
you
can
definitely
update
the
cron
job
with
the
latest
image
version
after
it
can
built.
It
says
it
like
it.
D
That's
a
nasty
ICD
process
but
yeah
that
the
current
version
running
of
that
cron
job
is
obviously
running
the
old
version.
If
you
know
you
do
it
at
the
same
time,
so,
but
yeah
you're,
definitely
safe
too,
to
update
the
cron
job
object.
Well,
a
job
is
running
so
remember
that
that
job
that
is
currently
running
is
detached
from
the
cron
job
kind
of
thing
like
it's,
it's
it
was
spawned
by
the
cron
job,
but
it's
now
its
own
job.
So.
A
We
are
almost
caught
up
on
questions
here
and
we
have
about
15
minutes
before
the
t-shirt,
giveaway,
so
I'll
keep
them
coming.
Mauricio
ass.
When
I
see
devices
being
passed
to
kubernetes,
it
seems
that
they
must
be
in
slash
dead
first,
like
in
docker,
so
way
to
expose
slash,
pass
a
device
that
only
exists,
a
slash,
slash
bus
instead,
I.
B
A
A
And
while
we're
waiting
for
that
just
a
reminder
that
kubernetes
1.16
is
probably
coming
out
today,
I'm
barring
any
problems,
so
it's
gonna
be
a
busy
day.
Those
of
you
asking
yes,
we
do
do
these.
The
third
Wednesday
of
every
month,
I
always
announce
them
in
the
channel
a
week
before
or
you
can
just
subscribe
to
the
kubernetes
channel
on
YouTube.
Every
time
we
go
live,
you
will
get
a
notification
but
yeah.
We
we've
done
this
for
what
almost
two
years
I
don't
know,
quick.
A
A
If
I
haven't
answered
your
question
yet,
please
repeat:
it
I
think
next
we
have
Ramesh's
question
and
then
ash
looks
like
they
have
a
question
so
I'll
go
ahead
and
add
that
Rama
asks
I'm,
seeing
nodes
going
unhealthy
in
the
logs
chose,
plague
and
pods
go
into
a
zombie,
State
Alceste,
full
pod,
stuck
in
termination
state.
What
could
be
the
reason?
A
D
So
I've
earned
in
this
multiple
times
a
while
back,
and
it
was
almost
always
the
run
time
in
in
his
case,
docker
and
I
would
be
on
the
hosts
actually
kind
of
systems
system
D
getting
logs
from
from
docker
and
trying
to
figure
out
what's
going
on
there.
A
lot
of
it
was
like
maybe
odd
issues
like
something
space-related
or
lock
somewhere
or
you
know
other
things
like
that
or
resource
to
tender,
or
something
like
that.
I
never
really
found
like
a
singular
like
this
always
causes
this.
D
But
if
you
do,
some
searching
on
Google
you'll
definitely
have
like
a
ton
of
results
and
you
can
kind
of
learn
more
about
there
and
what
it
means,
but
yeah,
most
of
the
time,
it's
very
specific
to
the
node
and
it's
basically
something
related
to
cubelet
working
with
the
wrong
time
to
get
the
job
done
and
a
lot
of
it
I
think
one
of
the
times
this
docker
wasn't
even
responsive
to
cubelet
and
so
that
it
was.
It
was
a
big
problem.
D
So
definitely
look
at
your
notes
and
what's
going
on
there
and
check
out
the
logs,
so.
A
F
F
A
D
I
was
gonna,
suggest,
actually
there's
some
pretty
good
home
charts
for
this
that
we
use
that
work
really
well.
So
if,
if
this,
if
ash
ever
looks
into
it
may
be
doing
something
with
home,
you
know
I
know
this
was
kind
of
the
initial
implementation
of
Prometheus
on
on
urbanites
with
the
Eliquis,
as
we
know
yet,
JSON
in
case
I,
don't
really
support
anymore,
so
yeah
libraries.
E
D
B
E
Yeah,
but
looking
at
that
repository,
it
actually
has
like
customization
file.
So
it's
probably
it's
its
own
customized.
So
I
would
look
at
that
file
and
actually
try
to
figure
out
which
manifests
you
actually
need,
but
this
seems
to
be
like
this
wrap
is
actually
still
maintained
and
it's
like
the
last
time
it
was
five
days
ago,
so
it
should
be
doable.
Mm-Hm.
B
A
A
A
A
D
F
Yeah
recently,
probably
a
month
back
I
did
an
AWS
Prometheus.
The
Helen
shards
really
worked
out.
Well,
that's
very
simple:
they
have
the
node
exporter,
cube
state
map,
matrix
alert
manager,
everything
as
one
package
it
just
works,
beautifully.
A
A
Max
guy
wants
to
ask:
what's
your
take
on
running
traditional
sequel
databases
like
Postgres
and
my
sequel,
on
kubernetes
on
pram,
so
not
RDS
like
I've
been
using
helm
chart,
but
it
could
be
painful.
I'm
experimenting
with
zalando
Postgres
operator
and
cockroach
DB's
operator.
That's
part
of
look
anything
you
can
recommend
other
than
don't
do
it
there's
a
great
blog
series
that
I'm
gonna
find
the
yarra
here
from
Josh
burkas,
who
is
in
kubernetes
and
has
been
in
the
Postgres
community
for
a
while
about
what
you
need
to
do.
Here's
a
video.
B
B
D
I
was
gonna
say,
but
that's
for
my
school
I'm
guessing
they're
using
Postgres,
but
it
definitely
like
it
takes
out
the
client
burden
of
doing
the
charting
and
it
kind
of
takes
that
on
itself
and
helps
load
balance
a
lot
better.
So
definitely
look
at
that,
but
yeah
Josh
burkas,
it's
kind
of
one
of
I
consider
them
one
of
the
leaders
of
yeah.
A
A
So
hopefully
that'll
give
you
the
information
that
you
need.
Thanks
for
asking
the
question,
and
probably
our
last
question
for
the
day:
Ramesh
one
interesting
thing:
I
said
of
kubernetes
cluster
with
metrics
enabled,
but
whenever
I
run
cute
cuddle
top
nodes,
it
always
says
metrics
not
available.
The
metrics
pod
is
running,
though
I
gave
a
lot
of
time
for
it,
but
it's
still
the
same
result.
What
could
be
wrong?
D
Yeah,
so
this
is
for
us
actually
in
our
deployment
process,
bootstrapping
new
clusters,
we
actually
install
metrics
server
and
cube
state
metrics.
We
actually
run
a
new
relic
agent
as
well,
that
you
consumes
cube
state,
metrics
and
then
I
believe
criminais
days.
Natively
is
using
metric
server
for
those
pod
and
node
stats,
you're
looking
for
so
our
metrics.
If.
B
D
A
Ok,
and
with
that,
that's
our
final
question
for
the
day
and
we're
about
to
wrap
it
up.
I'd
like
to
thank
our
panels,
great
job.
Everyone
welcome,
BB,
your
first,
your
first
office
hours,
I,
hope
you
join
us
in
the
future.
Those
of
you
listening.
If
you
want
to
find
a
great
way
to
get
back
to
the
community,
you
can
just
hop
on
here
and
try
to
answer
as
many
questions
as
you
can't
third
Wednesday
of
every
month.
A
We
if
you
join
us
and
with
that
it's
time
to
give
away
the
t-shirts
so
I
went
ahead
and
and
ran
the
raffle
here,
and
our
two
winners
are
max
at
and
Thomas
post
pic
SEC.
Sorry,
if
I,
if
I
butchered
fat
I,
will
DM
you
after
the
show,
with
a
t-shirt
ship
code
to
CNC
fo,
you
just
go
to
store
dot,
C
and
C
fi
o
any
of
you
and
get
a
shirt
for
your
favorite
cloud
native
projects.
I
give
you
a
little
code
there
as
we
appreciate
you
participating.
A
We
give
away
two
to
t-shirts
every
month,
so
I'm
not
wearing
one,
but
it's
similar
to
this,
except
with
the
official
kubernetes
logo
and
in
a
nice
dark
gray,
and
with
that
audience,
I
hope
you
have
a
good
time
panelist
any
last
minute
things.
Congratulations!
Everybody!
On
kubernetes
dawa!
When
that's
sixteen
coming
out
today,
it's
gonna
be
a
busy
day.
Last
word.
Anyone.
D
Just
eks
supports
1.14
any
KS
CTL
also
supports
1.14,
just
the
exciting
news
of
my
last
couple
weeks.
So
just
move
throw
my
clusters
to
it.
So
I'm
super
happy
with
that
and
stock
X
is
hiring
and
you
can
work
with
me.
I'm
kubernetes
and
auto-scaling
and
challenges
of
being
a
start-up.
So
I'll
stop
talking
now.
I
promise.
That's
awesome.
A
If
you're,
looking
for
a
pair
of
Air
Jordans
like
you,
are
I,
definitely
snagged
mine
through
your
kubernetes
cluster,
anything
else,
alright,
thanks
audience,
we
hope
you've
enjoyed
it.
We
hope
you
join
us
again.
Please
help
us
by
tweeting
and
letting
someone
know
if
we
helped
out
and
with
that
everyone
have
a
great
day
and
let's
keep
those
clusters
up
and
running.
Everyone
cheers
thanks.
Everyone,
oh
I'll,
publish
the
show
notes
with
the
URLs
and
stuff
in
the
discussed
thread
and
I'll
put
that
in
the
channel
as
well
thanks
everyone.