►
From YouTube: Kubernetes Office Hours 20171018 (European Time Zone)
Description
Office Hours are where people can ask questions about Kubernetes and have experts standing by on the live stream to answer questions. Join us every third Wednesday of the month here and in #office-hours on the Kubernetes Slack channel.
All the information/instructions on how to join are here: http://bit.ly/kubernetes-office-hours
B
All
right
there
we
go
yes,
so
that
slider
all
the
way
left
should
have
been
all
the
way
right,
but
last
time
I
had
it
all
the
way
right,
and
everyone
said
it
was
too
noisy.
So,
as
you
guys
hear
the
audio
levels,
let
me
know
I'll
adjust
over
time
until
we
absolutely
figure
this
out.
We've
only
had
two
of
these
before
so.
Thank
you.
Everyone
for
joining
this
is
the
kubernetes
community
office
hours.
We
ran
two
experiments
before
and
had
some
decent
participation,
so
we're
gonna
do
is
every
third
Wednesday
of
the
month.
B
We're
gonna
have
two
sessions,
one
at
this
timeslot.
It's
gonna
be
every
third
Wednesday
here
and
then
one
six
hours
from
now
towards
the
end
of
my
day
for
the
West
Coasters,
so
you'll
get
a
different
set
of
developers
each
time,
depending
on
who's
around,
depending
if
dr.
Kahn's
happening
or
not
and
who's
there.
The
way
it
works
is
we
bring
these
experts
here.
Let
me
see
if
I
can
point
to
your
windows
there
who
understand
kubernetes,
and
we
answer
kubernetes
questions
from
the
audience
to
kind
of
help.
B
C
A
A
Given
that
E's
world
I
am
also
a
co-lead
of
sig
apps,
which
is
especially
disputes
for
running
applications
on
top
of
these,
so
then
I
contributed
to
given
empty
town.
So
if
you
have
any
question
so
that
I
can
help
on
to
those
and
also
one
of
the
main
tenants
that
looking
at
these
charts
with
all
the
trees
law,
okay,.
B
D
D
B
Okay,
awesome,
so
here's
how
it's
gonna
work,
we've
had
people
in
hash
office
stash
hours
already
asking
questions
feel
free
to
tweet
people
to
come
here.
I
have
a
bitly
link
that
you
should
see
everywhere
that
I'm
sharing
with
people
that
will
bring
them
to
this
location.
So
those
of
you
listening,
if
you
can
help
us
out
by
tweeting,
letting
people
know
letting
co-workers
know
we
had
during
the
first
one,
we
had
a
whole
company
had
a
lunch
and
they
watched
the
stream
over
lunch
as,
like
a
team
I
thought
that
was
really
awesome.
B
So
the
goal
here
is
to
to
those
of
you
that
are
out
there
using
kubernetes
in
the
real
world
help
this
you
know
have
this
be
a
useful
tool
for
you
or
you
want
to
come
and
watch
and
learn
a
bunch
of
things.
So
with
that
the
questions
are
starting
to
pile
up.
So
we're
gonna
begin
as
always
just
do
question
in
all
caps.
So
I
can
notice
the
question
colon
and
then
ask
your
question.
B
Please
try
to
keep
your
questions
answerable
so,
for
example,
we're
not
going
to
be
able
to
SSH
into
your
cluster
and
fix
things
and,
in
general,
the
developers
we're
not
we're
going
to
try
to
turn
everything
into
a
teaching
moment
right.
So
if
you
have
like
a
syntax
error,
we're
not
just
gonna,
say:
hey.
You
fat-fingered
this
kind
of
try
to
make
it
so
every
single
question
we're
learning
a
little
bit
about
something.
B
So
with
that.
The
first
question
is
for
van
borg,
thanks
for
coming,
because
how
do
you
guys
handle
docker
volume
and
not
running
with
the
root
user?
I
got
the
Postgres
set
up
running
with
a
mounted
volume,
but
I
cannot
seem
to
have
the
volume
that's
mounted
writable
for
Postgres
user
instead
of
route.
I've
also
have
ranch
own
capital
are
postcards
column
postcards
on
the
mount
as
well.
Is
the
CH
own
a
good
idea.
B
Excellent
and
if
you
have
a
follow
up
that
for
van
bore
once
we
catch
up,
feel
please
feel
free
to
ask
as
many
follow-up
questions
as
you
need.
I
should
have
mentioned
earlier.
This
is
no
judgement
zone,
so
no
one's
ever
gonna
make
fun
of
anybody
for
any
question.
There's
no
question:
that's
too
too
basic
or
to
beginner
here,
I've
call
I!
Guess
that's
how
you
said:
question
I
need
more
than
256
pods
per
node
in
a
bare
metal
cluster.
Could
you
give
me
an
example,
valley
of
and
then
a
bunch
of,
cluster
cider
files?
D
A
B
E
D
B
Eight
ApS
in
an
expert
use
of
the
threading
functionality
and
slack.
This
is
the
first
time
I've
seen
it
used.
Reg
says
you
can
ash
max
pods
flag
on
the
cubelet
to
bump
the
limit,
but
I
don't
know
if
you
just
double:
that's
like
over
double
what
the
default
is.
I
feel
like
that
might
surface
some
unforseen.
E
B
So
we'll
wait
for
follow-up
on
that.
That
might
be
one
of
those
hey.
Six
scaleability
might
be
interested
in
your
use
case
to
figure
out
what's
up
the
next
one.
This
is
a
live,
likeness,
probe
question.
My
cluster
is
on
gke
and
I
have
a
pie
with
two
containers,
one
for
apache2
and
a
proxy
for
sequel
on
the
next
one.
My
problem
is
that
I
can't
make
liveness
probe
to
work
and
apache2.
I
have
to
find
a
service
as
TCP
load
balancer.
B
B
A
B
And
it
looks
like
ApS
is
asking
him
as
well:
big
shout-out,
thanks
very
much
we'd
love
to
see
it
when
other
experienced
people
are
helping
people
out
in
the
channel.
So
thanks,
that's
I'm
on
Preet
Singh
at
crowdfire.
Thank
you
very
much
for
your
help.
Okay,
let's
see
question
oh
and
then
he
follows
up
with
a
question.
Excellent
I've
seen
pods
get
stuck
in
terminating
state
with
failed
to
kill
container
RPC
error
code
for
context.
Deadlines
exceeded.
I
understand
this,
probably
due
to
docker
hanging.
What's
the
right
way
to
deal
with
this.
D
Stuck
in
terminating
book
because
we
were
running
cubelet
in
a
container
and
double
mounting
speaking
volumes
in
there,
okay,
but
this
seems
to
be
a
different
problem
here
in
general.
If
docker
is
hanging
I've
seen
that
like
once
recently-
and
sometimes
you
need
to
restart
the
document,
if
you
have
one
to
open
up,
you
should
be
able
to
restart
without
restarting
containers,
so
energy
can
be
attached
to
the
shims
and
sometimes
I've
seen
that
that
wasn't
enough
than
you
was,
and
we
also
restarted
qubit
just
to
get
the
the
current
state
again.
B
B
B
So,
even
if
we
can't
find
an
answer
for
this
week,
hang
out
in
the
channel
and
then
I'll
usually
be
able
to
find
someone
that
could
at
least
help
you
out.
So
that's
what
happens
when
we
get
stuck
all
right
next
question.
This
is
gonna,
be
a
nice
long
one.
What
are
the
best
practices
for
doing?
Ci
CD
with
kubernetes
Tara
for
support
is
limited
to
stable
and
I.
Really
don't
want
to
use
deployments.
B
The
best
option
I
found
is
creating
helm,
charts
with
dependencies,
putting
them
in
the
repo
and
using
helm
files
on
the
Jenkins
plug-in
that
can
install
/
upgrade
a
hound
chart
I've
yet
to
implement
this.
So
my
understanding
might
not
be
correct,
I'm.
Looking
for
a
solution
that
would
limit
config
drift
as
much
as
possible.
A
So
so
home
is
a
good
way
to
do
this,
I
mean
you
can
create,
you
can
have
at
the
helm
chart
in
a
repository
and
then
I
haven't
actually
seen
the
Jenkins
bug
in
that
you're.
Referring
to
for
how
long,
but
it's
also
pretty
said
you
know,
I've
done
it
before
we're
just
recording
out
just
doing
a
home
upgrade
install,
which
is
the
same
thing
as
you
CTL
apply
and
then
I
was
going.
I
got
great,
that
home
shot
or.
B
A
I
think
I
think
the
nice
thing
about
the
way
good
Nettie's
works
is
that
doesn't
necessarily
there
isn't
a
necessarily
one
way
to
do
this.
There's
multiple
good
ways
of
doing
this,
and
you
know
helm,
is
one
of
them
just
using
cube
CTL
with
you.
If
you
have
a
very
simple
deployment,
just
using
keeps
you
Kalin,
bra
yeah.
Well,
is
also
fine.
A
You
know,
as
soon
as
you
start
wanting
to
substitute
things
and
doing
sense,
and
that
starts
getting
more
complicated,
and
maybe
you
want
to
look
at
something
like
a
hub
chart
or
or
I
think
there's
some
there's
some
stuff
in
this
domain
like
case
on
it,
which
is
a
chase
on
it.
Everything
with
with
a
library
of
goop,
knows
these
resources
that
you
can
just
modify
the
things
that
you
want.
So
there's
there's
multiple
ways
of
doing
this.
A
A
So,
in
terms
of
values
files,
what
I'd
like
to
do
is
have
a
different
values
for
all
the
different
environments.
So
if
I
have
a
if
I'm
in
one
that
like
that,
would
just
be
more
the
35-foot
production
one
that
could
be,
that
could
just
be
a
fire
that
exists
in
my
my
CI
server.
My
Jenkins,
for
example,
follow
that
sequence
now
and
then
I'm,
just
in
my
Oracle
thing,
I'll
just
but
I
want
for
the
environment
and
then
for
dependencies.
A
So
up
until
I
think
a
couple
of
releases
ago,
there
wasn't
a
way
to
point
to
dependencies
in
the
file
system,
so
you'd
have
to
set
up
a
hound
repository
and
then
point
to
dependencies
there,
which
actually
kind
of
made
it
all
go
to
work.
But
if
you
have,
if
your
chart
was
already
inside
your
repository
setting
now
you
can
actually
point
to
chance
in
the
file
system
as
the
dependencies
and
then
enter
them
in
in
the
chance
of
that
makes
a
lot
easier
to
work
with.
A
B
You
all
right
great
moving
on
vamp
or
gas.
Another
question:
can
you
show
an
example
of
using
FS
group
for
my
issue?
My
setup
consists
of
a
persistent
volume,
persistent
volume
claims
and
a
definition
of
the
pod,
here's
a
yamo
file
and
then
the
Yamma
file
for
the
sake
of
the
audience.
If
you
can
kind
of
talk
talk
through
this,
it
looks
like
you
answered
this
already,
but
if
you
could
talk
through
it
a
little
bit,
oh
yeah,
that.
A
A
B
Okay,
I
believe
so
I,
don't
think
I've
skipped
a
question,
but
if
I
have
please
repost
it
or
pointed
out
to
me
in
slack
and
we
can
absolutely
take
more
questions
if
we
have
more
and
I
see
people
typing
furiously.
So,
let's
see,
let's
see
what
we
get
here
and
I
will
go
ahead
and
see
if
I
can
get
more
people
from
kubernetes
users.
A
B
D
A
A
B
I
really
I
really
want
to
see
so
in
one
of
the
slides
I
saw
that
they
use
cube
admin,
I,
guess
spawn
and
I
really
want
to
see
exactly
how
they
did
that
I'm,
really
glad
that
they
used
cube
admin
and
didn't
write
their
own
tool.
I
was
very,
very
happy
all
right.
We
got
a
question
from
D
Barrett
coming
in.
Are
there
any
plans
for
increasing
the
discoverability
of
issue?
/
PRS
for
specific,
held
charts,
it's
already
quite
hard
to
dig
through
the
repo
and
I
could
see
it
getting
harder
as
it
grows.
B
A
B
B
A
B
A
Say
that's
it
for
the
ad-hoc
right
now.
You
know
we
had
the
laptops
initially
initiated
by
someone
not
related
to
get
lab
and
then
get
lab
cake
came
and
wanted
to
kind
of
over,
because
it
took
so
long
to
get
changes
through
they
decided
to
create
their
own
repository,
which
is
a
perfect.
You
know
perfectly
fine,
like
the
helm
model
works
in
this
way
we
can
just
create
any
repository,
and
then
you
know
music,
an
app
that
and
it's
all
stuck
with
there,
but
in
terms
of
discoverability
and
ease
of
use.
A
It
makes
a
lot
of
sense
to
have
a
centralized
repo
as
well,
so
that
people
could
just
easily
find
things.
You
know
the
way
apt-get
works,
for
example.
So
so
we
are
so
we
do
want
to
have
people
like
it
get
lab,
and
you
know,
data
dog
is
another
example
of
their
maintaining
their
chart.
Cockroach
DB,
it's
so
there's,
actually
quite
a
few
where
we
have
companies
have
to
be
making.
There
are
charts
and
we
just
want
to
give
them
more
control
over
that.
B
Okay,
awesome
and
it
appears
that
I
do
have
an
echo
going
on
so
I've
got
to
toggle
my
mute
every
time.
I'm
not
talking,
see
I
thought
I
was
set
up
perfectly,
but
I
guess
everyone
needs
headphones
at
one
time,
Santhosh
Kumar
says
hi
team
I
could
see,
there's
a
skip
option
enabled
on
the
kubernetes
dashboard
1.7.
Is
there
any
possibility
to
disable
the
skip
button
and
then
he
has
a
screenshot
of
his
actual
dashboard
where
it
looks
like
he
wants?
Yes,.
D
B
B
D
D
B
B
B
Yes,
so
speaking
about
about
the
helm,
just
going
back
to
helmet
feels
like
we
could
have
a
topic
we
keep
discussing
while
we're
waiting
for
questions
yeah,
so
I
think
it
would
be
really
interesting
like
how
would
like
cube
ops,
calm,
indexed
apps
that
live,
maybe
in
some
other
organizations
version
control
right.
Yes,.
A
I
think
what
we
plan
to
do
is
and
if
you've
seen
the
way
doctor
library
does
this.
This
is
the
official
docker
docker
hub
repository.
They
have
a
library
repo,
which
I
think
links
to
uber
repos
and
basically
just
it's
kind
of
like
a
sub-module
type
thing
and
then
to
see
that
repo
itself
has
CI.
That
goes
then
publishes
a
you
know
real
I
guess
for
the
docker
it
publishes
in
the
registry,
but
for
for
our
use
case
it
would.
It
would
be
going
to
the
same.
A
You
know
GCS
bucket,
that
we
have
and
just
building
out
that
repo
there
so
you'd
still
be
the
same
experience
as
it
is
now.
It
still
be
one
single
repository,
but
it's
just
that
underneath
it's
all
pointing
to
two
different
repositories
get
help.
You
know
proposed
fees
yeah.
So
that
way
we
can
keep
the
same
experience
without
having
to
you
know,
deal
with
multiple
repositories.
B
Yeah
and
it
it
removes
blockers
right
as
opposed
to
having
one
centralized
thing
to
review
like
when
I've
been
on
teams
where
we
contributed
to
brew,
which
is
the
package
manager
and
they
do
a
really
great
job.
But
at
the
end
you
know
you,
there
there's
one
bottleneck
that
goes
through
and
there's
there's
bonuses
there,
but
I
also.
C
A
You
know
the
the
same
maintenance
that
we
have
for
kubernetes
slash,
charts
is
they're,
not
necessarily
subject
matter
experts
for
all
those
applications
that
are
in
that
repository
right.
So
you
know
I'm,
not
a
subject
matter
expert
for
forget
lab,
so
am
I
the
best
person
to
reviewed
changes
that
are
very
related
to
or
very
specific
to
how
good
lab
is
run.
A
Probably
not,
but
someone
you
know
from
get
lab
is
obviously
going
to
be
more
knowledgeable
in
that
area
and
they
should
have
they
have
that
knowledge
to
be
able
to
accept
it
er
and
because
it's
so
because
of
the
barrier
of
getting
someone
to
be
a
contributor
to
the
cube,
Nessie
strands
repo.
You
know
we
have
to
get
them
Ritter
the
cube,
Nettie's
org,
so
that
they
can
spin
off
tests
and
all
this
it
makes
it.
B
E
D
B
A
B
B
D
B
So
one
thing
I
really
like
about
how
you
folks
built
this
is
I
love
having
the
old
versions
here
on
the
side
right,
so
I
can
like
keep
track
of
that
stuff,
and
then
it
just
has
like
the
source
link
right
there.
It's
not
it's
not
like
a
complicated
thing,
and
then
you
just
parse
the
readme
and
make
it
all
pretty
I
like
that.
A
lot
yeah.
B
B
B
B
B
B
B
Like
when
I
saw
it,
I
was
like
cuz,
you
know
those
racks
that
they
have,
that
are
just
nothing
but
GPUs
and
I
was
like
if
I
had
a
spare
one
hundred
and
twenty
thousand
dollars,
and
then
you
I
found
myself
kind
of
doing
the
math
on
the
electricity
in
my
head
and
then
I
was
like
okay,
whatever
how.
B
Alright,
so
thanks
for
pasting,
the
link
to
the
RabbitMQ
chart
chart
there
in
slack
next
question
from
Bobo
s,
I
was
installing
cube
admin
and,
for
a
start,
also
used
master
for
gearing
up
a
pods.
Now
all
I
want
hold
on
now.
I
want
all
possibly
remove
from
master
and
forbade
any
pods
from
ever
start
up
master.
D
E
B
B
B
D
B
So
so
I've
actually
done
this.
So
when
I
do
my
cube,
admin
testing
I
have
two
laptops
that
I
also
set
an
upgrade
to
and
in
order
to
test
this,
because
I
think
this
is
the
coolest
thing.
Even
though
it's
like
a
basic
thing
now
is
I
close
the
lid
on
the
laptop
to
kill
the
node
and
then
I
just
watched
the
pods
come
up
on
other
nodes
and
then,
when
I
open
it
back
up,
you
know,
and
it
comes
back
on
line
and
and
all
that
stuff,
then
things
can
get
scheduled
it
on
again.
C
B
D
B
D
B
D
D
B
B
What
happens
in
the
case
where
I
think
what
he's
saying
how
to
recover
the
node
like
if
the
node
itself
is
broken?
Let's
say
you
did
a
kernel
upgrade
or
something
on
the
node,
and
it's
like
total
heart
failure.
You
would
just
so
on
a
public
cloud.
He
would
just
terminate
the
instance
and
get
you
a
new
one
right
and
then
I'm
bare
metal.
You
would
just
do
a
reinstall
or
whatever
rejoin
it
later,
and
then
everything
should
be
fine,
so
I,
don't
I,
don't
think
in
this
world.
B
D
D
B
B
Next,
a
PSS.
We
have
some
JVM
apps
that
take
a
lot
of
CPU
when
starting
up,
but
not
so
much
when
they're,
just
operating
normally
setting
lower
CPU
limits,
increases
the
startup
time
and
higher
CPU
limits,
waste
CPU
or
affect
other
apps
running
in
the
burstable
class.
Is
there
a
way
sort
of
that
like
preload
stuff
in
the
JVM,
when
building
the
docker
image?
Is
this
how
the
JVM
is
supposed
to
work?
What's
the
recommended
way
to
go
about
this.
D
B
B
B
D
B
This
is
something
I've
seen
in
one
of
the
scheduling
SIG's,
where
you
would
have
Prometheus
and
all
the
metric
goodness
always
constantly
running.
Monitoring
your
apps
and
dynamically
adjusting
I'm
gonna
try
to
find
there's
a
blog
post
on
this
work,
but
it's
not
complete,
but,
let
me
add,
let
me
add,
Mike
says
we
experienced
this
and
just
had
to
increase
the
CPU
memory
size.
We
had
to
deal
with
the
high
usage
of
resources
and
Suraj
kind
of
asked
a
follow-up
question:
are
there
any
tweaks
that
the
JVM
has
enabled
to
run
inside
containers
better?
B
B
Let
me
maybe
we
can
move
to
another
question
while
I
do
some
digging
for
people
and
get
them
some
links
here.
So
he
says:
I
heard
some
interesting
things
about
PKS,
pivotal
container
services,
they're
introducing
Bosh
to
take
care
of
nodes
as
well.
The
Bosh
will
spin
up
another
node
like
a
pod,
but
I
didn't
see,
documentation
anywhere,
so
I
think
there's
one
of
those
things.
B
D
B
Leave
I
found
the
link
on
everything
you've
ever
want
to
know
about
resource
scheduling
almost
and
this
is
that
refers
to
one
of
Tim
Hawking's
talks
and
then
Michael,
Hasan,
Blas
kind
of
goes
through
some
of
the
work
that's
happening
in
this
area.
I
think
that's
a
really
great,
read
and
I
found
it
to
be
great
when
I
found
it
so
sorry,
do
we
talk
about
how
we
change
the
limits?
That's
just
in
the
yellow
files
right,
yeah.
B
Sure,
okay,
so
Tim
Timothy
goes
on
to
say
yeah.
It
doesn't
do
that
on
the
fly
by
default.
Those
are
BPA's.
Even
then
CPUs
compressible
memory
is
not
VP
is
vertical
pod,
auto
scaling.
Okay!
So
that's
that's
a
good
one!
I'm
I'm
gonna
I've
linked
this
question
because
I
think
that's
a
that's
a
common
question.
B
If
you
look
a
lot
of
the
workloads
that
are
using
JVM
that
people
are
using
in
real
life,
I,
don't
think
it
would
hurt
for
us
to
continue
to
get
visibility
on
that
area,
because
it
appears
that
a
lot
of
people
are
looking
for
best
practice
there
all
right
minute,
so
I'm
gonna
power.
Through
these
question
we
run
a
fair
amount
of
crown
jobs,
but
in
kubernetes
it
seems
your
urge
to
use
jobs.
But
in
my
opinion
it
seems
overkill
to
deploy
container,
to
run
a
job
and
kill
it
again.
A
D
D
B
D
B
D
D
D
A
E
D
B
D
B
A
B
Sure
yeah,
so
basically,
if
our
backs
enabled
don't
worry
about
the
button,
although
I
can
see
from
a
UI
perspective,
you
know,
depending
on
how
many,
how
comfortable
your
developers
are
might
be
like
go
see
your
administrator
and
get
your
creds
for
this
cluster
instead
and
it
takes
you
to
a
Help
section
or
something
like
that.
Maybe
you
can
hack
yourself
a
custom
button
or
something
yeah.
B
So
I
will
I'm
gonna
post
a
list
of
SIG's
here
real
quick
as
our
last
thing.
Unfortunately,
we're
not
we're
out
of
time
for
questions
and
we'll
be
back
next
month.
All
of
the
features
that
are
written
in
kubernetes
correspond
to
a
certain
sig.
So,
for
example,
the
dashboard
is
written
by
sig
UI
and
all
the
SIG's
have
public
meetings,
mailing
lists
and
slack
channels.
B
So
that's
the
list
of
SIG's.
If
you
have
any
questions
of
what
goes
where
you
can
always
ask
somebody
hey
what
what
SIG's
should
this
be
a
part
of
and
they'll
usually
point
you
in
the
right
direction,
but
for
us
I'd
like
to
thank
our
guests
for
for
attending,
we
got
through
quite
a
bit
of
questions.
We
got
to
hard
ones
that
we're
gonna
try
to
follow
up
here
over
the
next
week
or
so
and
see.
If
we
get
you
some
more
attention
too.
B
We
are
gonna,
have
another
one
of
these
sessions
and
about
five
hours
or
so
for
the
west
coast
of
the
US,
with
a
totally
different
set
of
developers
and
for
those
of
you
who
found
this
useful.
Please
come
back
next
time.
Please
hang
out
in
the
channel.
You
can
hang
out
of
the
channel.
You
know
as
long
as
you
want
and
just
kind
of
keep
it
going
we're
trying
to
build
a
place
where
people
can
feel
comfortable
and
answering
questions
help
pay.
B
It
forward
help
someone
else
that
kind
of
good
stuff,
so
we
will
see
everybody
in
exactly
a
month.
We
are
the
third
Wednesday
of
every
month
and
the
topic
on
the
channel
in
office
hours
will
send
you
to
the
page.
If
you
have
any
feedback
on
that
page,
please
let
me
know
and
other
than
that
any
any
final
words
from
our
guests.