►
From YouTube: Minikube - Thomas Strömberg & Medya Ghazizadeh, Google
Description
Join us for Kubernetes Forums Seoul, Sydney, Bengaluru and Delhi - learn more at kubecon.io
Don't miss KubeCon + CloudNativeCon 2020 events in Amsterdam March 30 - April 2, Shanghai July 28-30 and Boston November 17-20! Learn more at kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects
Minikube - Thomas Strömberg & Medya Ghazizadeh, Google
If you want to develop applications for kubernetes and try them out locally first, minikube is going to be your best bet. Come and explore the functionality of minikube via an introductory session from minikube maintainers from Google. We will cover the basics of getting started with minikube, how the local environment compares to a full kubernetes cluster and what customizations you can do to bring the maximum out of your local experience.
https://sched.co/UahA
A
So
I'm
told
they
say
the
best
talks
for
the
end
of
the
day.
So
thank
you
for
being
patient
everybody.
We're
very
excited
to
give
give
this
talk.
I
know
you
all
have
parties
to
get
to
and
everything
stick
with
us
at
least
til
til
the
end.
We
have
we're
gonna,
be
demoing,
a
bunch
of
prototype
work
that
we've
been
doing.
So
if
you
enjoy
watching
failure,
we're
gonna
have
it
and
yeah
guaranteed
failure
today,
okay!
Well,
it's
guaranteed
failure
in
the
in
the
event
of
unexpected
success.
A
All
right,
it
looks
like
we
are
at
time,
so
I'm
gonna
get
started.
My
name
is
Thomas
Stromberg
I
am
one
of
the
maintainer
x'
for
mini
Q
and
I
am
here
to
convince
you
that
mini
cube
is
wonderful
and
it
should
be
the
first
repo
within
the
kubernetes
community
that
you
contribute
to
because
we
love
new
contributors
and
we're
gonna
show
off
a
bunch
of
new
features
that
we've
been
working
on.
So
so
the
three
people
presenting
today
we've
got
media
Sharif
and
myself.
A
A
So
there's
there's
this
famous
book,
the
Kalevala,
which
is
like
the
epic
story
of
the
people,
and
it
has
this
idea
that,
in
order
to
to
really
cast
the
best
spells,
you
have
to
know
why
something,
how
something
came
into
being
and
if
you've
been
working
a
lot
with
software.
This
actually
might
sound
kind
of
intuitive
to
you,
because
you
know
when
you're
debugging
systems,
knowing
why
they
were
designed
in
such
a
way,
actually
really
helps
you
understand
what
failure
case
is
to
look
for.
A
Some
of
you
might
actually
recognize
a
a
particular
theory
that
is
espoused
within
sre
teams,
about
Chesterton's
fence,
which
has
a
similar
concept,
which
is
basically
it's
basically
the
the
story
of
when
you
go
and
see
a
fence
blocking
a
road.
You
know
the
newbie
software
engineer
will
just
say:
I'm,
just
gonna
refactor
this
out
of
the
way.
There's
it
doesn't
make
any
sense.
Why
would
anybody
put
a
fence
here,
but
the
more
experienced
software
engineer
will
say.
A
A
So
why
am
I
telling
you
all
this?
It's
basically
because
I
want
to
make
you
guys
contributors.
Everyone
in
this
room
could
easily
be
a
contributor
to
mini
cube.
Having
a
beginner
mindset
is
hugely
valuable,
especially
for
a
tool
such
as
ours.
That
is
supposed
to
be
easy
to
use
for
people
to
learn.
Kubernetes
we'd
love
to
make
you
one
of
our
four
hundred
plus
contributors
we'd
be
happy
to
guide
you
on
your
journey
to
your
first
commit
and
quite
honestly,
mini
cube
is
not
rocket
science.
A
A
So,
who
is
here
running
kubernetes
at
v?
1.2
didn't
think
so,
so
that
gives
you
kind
of
an
idea
of
how
far
back
mini
cube
goes.
This
was
the
proposal
for
the
implementation
of
mini
cube.
Originally
in
April
2016,
the
world
was
a
different
place
back
then
there
was
no
docker
for
desktop.
There
were
no
local
kubernetes
options,
I
would
say
yeah
well,
you
would
have
to
be
slightly
crazy
to
be
running
kubernetes
at
this
point
in
time
things
were
pretty
rough
around
the
edges.
A
So
a
a
brief
history
and
time
for
for
mini
cube
and
I
realized.
Now
that
these
numbers
are
so
completely
short
that
you're
not
going
to
make
heads
or
tails
of
it,
but
so
the
first
version
came
out
in
summer
of
2016
VirtualBox
only
it
was
running
something
called
local
cube,
which
was
slightly
madness.
It
was
basically
bundling
all
the
kubernetes
services
into
a
single
binary
and
running
instead
of
pods.
Everything
was
go
routines,
but
it
was
an
easy
to
setup
very,
very
low
resource
way
of
running
kubernetes.
It
still
required
a
Linux
VM.
A
Time
went
on
and
by
I
guess,
mid
2017
hue
ATM
support
was
added
and
finally,
in
mid
2018
we
actually
removed
the
local
cube
support.
Thankfully
it
was
really
impressive
at
the
time.
Actually,
the
person
who
wrote
cube
ATM
wrote
local
cube
first,
so
I
can't
really
knock
it
too
much.
But
so,
if
you
see
on
the
chart
things
kind
of
started
to
accelerate
once
we
stopped
using
crazy
and
actually
had
like
real
kubernetes
running
in
mini
cube,
we
were
able
to
run.
A
We
were
able
to
keep
up
much
more
easily
with
the
mainstream
kubernetes
branches,
and
nowadays
the
kubernetes
that
you
run
in
mini
cube
are
the
same
exact
production
binaries
that
you'll
be
using
in
production.
So
the
same
ones
you
download
are
the
same
ones.
We
download
we
ended
up,
adding
emoji
just
for
fun,
people
love
it
or
hate
it.
There's
an
environment
variable.
If
you
really
hate
it
that
you
can
turn
it
off.
A
V1,
oh
came
out
earlier
this
year
was
I,
guess
March
April,
so
we
added
offline
multi
cluster
support
and
we've
basically
been
doing
like
a
monthly
release.
Cadence
since
then
adding
a
variety
of
things
to
make
mini
cube
easier
for
users,
so
our
principles,
so
simple,
local
kubernetes
for
local
development
and
learning.
A
We
find
that
actually
about
40
to
60
percent,
depending
on
like
when
we're
actually
doing
the
polling.
So
we
have
these.
These
survey
results
where
we
ask,
you
know
how.
Why
are
you
trying
mini
cube
and
roughly
half
of
the
people
are
actually
trying
mini
cube
to
learn
many
kubernetes,
which
is
very,
it
definitely
drives
the
direction
of
mini
cube.
So
our
our
first
and
most
important
principle
is
to
be
inclusive
and
community
driven
mini
cube
for
many
people
are,
is
their
first
foray
into
kubernetes,
not
just
as
an
end
user.
A
Who's
curious,
like
the
first
time
I
read
about
this
kubernetes
thing,
I
was
like
oh
I
should
try
it
out.
How
is
this
thing?
How
does
it
work?
I
saw
mini
cube,
I
tried
it
and
I
wouldn't
say
that
my
computer
went
up
in
flames,
but
I
probably
kicked
it
a
couple
of
times.
I
was
really
really
frustrated
by
the
experience
of
my
first
time
running,
mini
cube
and
then
an
opportunity
actually
came
up
for
me
to
get
paid
to
contribute
to
mini
cube.
So
I
took
it
recommended.
B
A
A
A
So
we
end
up
serving
two
masters,
really
where
we've
got
these
new
users
and
we
want
to
give
them
a
great
experience,
but
we
also
have
users
who
are
trying
to
simulate
production,
and
you
might
actually
say
that
it's
competing
priorities,
but
I
would
argue
that
as
long
as
you
respect
the
principle
of
least
surprise
for
both
sets
of
people,
you'll
actually
make
both
of
them
very
happy.
You
don't
want
to
be
different
than
production.
It
should
be
an
accurate
simulation.
A
You
don't
want
to
surprise
the
people
who
are
who
are
trying
to
replicate
production
where
they've
got
cryo
and
kubernetes
113,
and
this
strange
set
of
plugins
and
we
don't
want
to
scare
them
users.
So
if
this
sounds
familiar
the
term,
the
combination
can
actually
work
pretty
well.
If
any
of
you
have
ever
ever
played
a
Kerbal
space
program,
it's
kind
of
the
same
principles
where
you
can.
You
know
you've
got
all
these
options,
but
you're
kind
of
user,
friendly,
you're
kind
of
trying
to
steer
a
user
to
the
golden
path
into
space.
A
A
We
always
guarantee
six
six
minor
releases,
which
is
much
more
ambitious
than
the
normal
current
ideas
deprecation,
but
we
find
that
a
lot
of
people
are
running
older
versions
of
kubernetes
in
production,
and
we
have
to
respect
that.
We
do
support
a
variety
of
runtimes.
I,
see
Kat
is
marked
as
plan,
but
it's
actually,
your
mileage
may
vary
at
this
point.
A
A
A
So
when
I
came,
I
I,
presented
on
mini
cube
about
a
year
ago
in
Seattle
was
anyone
there.
Ok,
excellent
well,
I
was
on
stage
and
I
just
joined.
The
team
I
had
basically
contributed
to
mini
queue.
Maybe
five
weeks
before
presenting
on
it,
thankfully
had
a
co-presenter.
You
knew
what
they
were
talking
about,
but
I
didn't
so
I
learned
a
lot
in
the
process.
We
did
push
for
same-day
cadence
with
kubernetes
major
releases.
We
we
have
simplified
the
installation
a
lot
since
then
we
went
multilingual.
A
It
turns
out
if
you're
trying
to
teach
people
around
the
world
about
kubernetes,
you
really
have
to
be
multilingual,
because
only
what
27
percent
of
earth
can
actually
read
English.
So
if
we
want
everybody
to
be
invited
to
the
party
we
have
to
cater
to
the
world
Multi
cluster
Whitstable
we've
got
all
sorts
of
new
options,
a
lot
more
environmental
awareness
and
some
reduction
in
start
latency,
and
we
got
a
new
website
for
documentation
when
we
host
tutorials,
among
other
things,
so
architecture.
A
So,
yes,
it
is
pointing
down
on
purpose.
So
right
now
at
this
is
timings
on
head.
A
fresh
start
on
hyper
kit
with
no
VM
is
about
85
seconds.
So,
what's
going
on,
why
is
it
taking
so
long?
So
a
lot
of
the
time
is
spent
honestly
firing
up
a
new
hyper
kit.
Vm
when
you
run
mini
Cube,
start
that's
kind
of
what
we
have
to
start
with.
A
Unlike
other
environments,
that
might
have
a
VM
running
in
the
background.
We
currently
don't
really
recommend
it,
but
we
will
soon
so
we
we
spend
you
know
25
seconds
30
seconds,
waiting
for
the
VM
to
boot
up
and
hyper
kit
to
get
an
IP
address.
This
is
using
something
called
Lib
machine
which
is
unfortunately
deprecated.
A
So
that's
we'll
get
into
that
a
little
later.
Once
the
machine
is
up,
you
start
actually
setting
it
up
to
be
a
worthwhile
node.
We
ask
the
state
Chen,
we
sync
time
we
set
up
the
CRI
and
the
CRI
in
this
case
you
know,
can
be
cryo
container
D
docker.
We
actually
we
default
to
docker
for
some
interesting
development
reasons,
but
then
the
next
step
is
we've
got
this
cache,
so
many
cubes
designed
to
work
offline.
We
cache
a
bunch
of
the
images.
You
can
actually
inject
your
own
images
into
the
cache
that
takes
another.
A
You
know
19
seconds,
to
load
all
the
images
into
docker
we'd
like
to
make
that
a
little
faster.
But
then
we
have
this
thing
called
a
bootstrapper
which
is
basically
just
to
install
cube,
ATM
and
run
QB
ADM
the
cube
ATM.
We
tell
it
to
pull
the
images
in
case
any
of
our
pushes
failed.
Then
we
tell
cube
ATM
to
init
and
cube
ad.
A
A
minute
is
the
same
way
that
you
start
a
curry,
92
cluster
in
production,
then
for
the
last
88
seconds
or
if
you're
in
an
older
building,
you
might
be
waiting
for
the
API
server
and
all
the
pots
to
come
up.
This
is
an
intentional
design
difference
between
mini
cube
and
a
lot
of
other
local
kubernetes
solutions.
In
that
we
we
wait
for
health
rather
than
just
assuming
health,
because
we're
trying
to
be
reliable.
A
A
A
Many
people,
don't
realize
that
there's
this
mini
cube
config,
where
you
can
set
persistent
options
like
if
you
want
to
lock
all
of
your
VMs
in
the
future.
Until
you
change
the
setting
to
have
you
know,
4
gigs
of
RAM
beep
into
a
particular
kubernetes
release.
That's
how
you
do
it
so
I'm
gonna
go
ahead
and
do
that
and
then,
oh,
let's
see
so,
I
could
just
type
mini
keep
start
here
and
it
would
be
really
boring.
A
I
heard
that
117
with
beta
2,
like
in
the
last
couple
of
days,
while
we're
sitting
here
at
Q
Khan,
so
one
of
the
things
you
can
do
with
mini
cube
is
you
can
actually
just
tell
it
any
arbitrary
kubernetes
version
and
it'll
download
the
binaries.
That
said,
if
you're
trying
to
run
like
a
mini
cube
version
from
2016
against
kubernetes
117,
that's
probably
not
going
to
work
very
well
because
ApS
have
changed.
A
We
have
this
new
flag
where
we
can
just
enable
add-ons.
So
we're
gonna
enable
a
helmet
iller
who
here
is
excited
that
tiller
went
away
I
sure
as
heck
am,
but
we
got
this
add-on
that
someone
can
tribute
it
anyways
and
I
wanted
to
show
it
off
so
P
I'm
gonna
set
name.
My
cluster
cube
Khan
and
I
mentioned
that
we
wait
for
the
API
server
and
all
the
pods
to
be
healthy.
Well,
it's
a
demo,
so
I'm
just
gonna
say
weight
equals
false.
A
So
what's
this
doing
under
the
hood,
so
we
have
this
command
start
go.
If
I
go
scroll
down
to
the
more
interesting
stuff
here
you
can
see.
We
just
have
a
function,
displaying
some
versions.
Checking
some
registry
mirror
stuff,
selecting
a
driver.
So
here's
actually
one
of
the
things
you
might
notice
that
we
didn't
actually
specify
a
driver.
One
of
the
more
recent
changes
in
mini
cube
is
that
we
auto
select
the
best
appropriate
driver
on
your
machine
depending
on
what's
installed.
If
you
have
hyper
kit
installed,
it'll
use
hyper
kit.
A
Now
we
used
to
hard
code
for
VirtualBox.
Here
you
can
see,
I've
got
hyper
kit,
VirtualBox
and
VMware
fusion
will
basically
walk
down
the
list
of
which
drivers
do.
We
have
the
best
CI
signal
for
so
it
created
the
hyper
kit
VM
and
you
can
kind
of
see
what
goes
on
when
we're
creating
a
hyper
kit
VM.
We
basically
take
your
config
options
and
give
it
hints
like
here's.
It's
passing
the
memory.
A
You
can
pass
a
number
of
CPUs
there's
some
deprecated
stuff
here.
I,
don't
want
to
talk
about
because
it
doesn't
work,
but
you
can
see
actually
there's.
So
we
have
this
health
checking
now
on
the
bottom,
where
we
look
if
hyper
kit
is
installed
and
if
not,
if
the
user
really
wants
to
run
it,
we
tell
them:
hey,
run
brew,
install
hyper
kit,
I
realize
that
means
you
should
have
brew
installed,
but
so
okay,
so
what's
up
and
it
is
a
normal,
normal
kubernetes
cluster.
A
You
can
do
all
the
things
you
care
about
and
you
can
see
tiller
is
still
pending,
but
that's
okay,
it'll
get
there
eventually
one
thing:
what
we
added
more
recently
is
we
had
a
lot
of
problems
with
users
complaining
that
their
cluster
came
up
but
cube
cuddle
was
having
problems
applying
stuff
to
it
and
it
ended
up
because
they're
cute
cuddlin
was
so
old.
It
couldn't
talk
two
or
two
knew
that
I
couldn't
talk
to
the
kubernetes
cluster
version
that
they
deployed.
So
we
added
a
little
warning
check
here.
A
A
A
I
had
named
my
cluster
cute
cotton
and,
if
you're
not
familiar,
we
actually
have
support
for
plumbing
load,
balancer
services
and
it
does
require
route
to
set
up
routes
and
there
is
no
possible
way
that
I
will
be
able
to
type
my
password
mobs
on
stage
looking
at
all
of
you,
so
I'm
just
gonna
go
to
the
next
slide,
so
mini
cube
versus
prod
like.
Why
is
it
different?
I
mentioned
this
because
someone
else
wrote
the
description
of
this
talk
and
they
mentioned
that.
A
I
should
talk
about
this,
so
so
in
many
production
environments
the
master
does
not
allow
jobs
to
be
rescheduled
or
scheduled
on
to
it,
because
mini
cube
is
single
node.
We
have
to
allow
jobs
to
be
scheduled
on
it,
and
that
turns
out.
That's
really
easy
to
do.
The
node
doesn't
resolve
kubernetes
dns
records,
but
pods
do
for
99%
of
people.
This
doesn't
matter,
but
if
you're
trying
to
do
things
with
like
NFS
or
something
you
might
actually
notice
this.
A
If
you're
trying
to
like
issue
mount
commands
on
the
node
to
a
kubernetes
spot,
you'll
notice,
this
that's
something
that
we
might
change
in
the
future.
Mini
cube
is
much
much
more
likely
to
run
out
of
memory
than
production
by
default.
Mini
cube
only
allocates
two
gigs
of
ram.
This
is
actually
one
of
the
most
popular
failed
failure.
Cases
for
people
when
they're
deploying
their
giant
Java
stack
and
everything
ins
and
one
of
the
things
that's
really
interesting-
is
when
you're
in
such
a
constrained
environment
in
a
normal
kubernetes
cluster,
the
kubernetes
stur.
A
Will
you
know
cute
cute,
cut
a
little
notice
and
say:
no.
You
can't
schedule
this
or
no
you're,
taking
up
too
much
RAM
I'm
gonna
kill.
You
turns
out
when
you're
so
constrained
as
you
are
in
mini
cube.
The
Linux
kernel,
notices
first
and
just
drops
the
hammer
on
whatever
the
hell
it
wants,
so
it
can
kill
docker
D.
We
actually
tell
API
server
now
to
kind
of
be
somewhat
exempt,
but
you
can
actually
kill
API
server.
A
As
well,
so
if
you
run
out
of
memory,
things
get
really
ugly
and
I'll
talk
about
how
to
notice
that
later.
But
you
can
set
a
dash
memory
flag
and
we
do
default
to
the
docker
run
time
which
a
lot
of
production
environments
don't
do,
because
it
allows
for
direct
image.
Pushes
and
I
will
actually
show
what
I
mean
by
that.
A
So
if
you
want
to
use
the
docker
command
to
actually
build
against
a
running
mini
cube
or
push
images
to
it,
you
can
use
this
docker
and
the
command
and
it'll
just
tell
you
exactly
how
to
access
the
docker
that's
running
inside
mini
cube.
So
it's
a
really
quick
way
to
sideload
things.
So,
alright,
so
many
key
versus
others.
So
I
I
stole
this
from
tilts
Guide
to
cluster
selection.
I,
don't
actually
use
tilt,
but
I
really
loved
what
they
had
to
say
here.
A
Many
cubes
what
we
recommend,
if
you're
willing
to
pay
some
overhead
for
a
more
high
fidelity
cluster,
and
these
are
some
of
the
pros
that
they
cited
I
kind
of
reworded
their
things
to
fit
on
the
slide.
But
basically
it
is
the
most
full
featured
local
kubernetes.
We've
got
a
lot
of
options.
We
can
build
images
in
cluster
with
docker
end.
We've
got
a
bunch
of
developer,
focused
things
like
the
load.
Balancer
support
add-ons
dashboard
integration,
support
for
many
releases,
able
to
use
technologies
that
require
kernel,
module
support
like
gee,
visor
and
cata
cons.
A
Higher
overhead
you'll
want
to
stop
it
when
not
in
use
running
within
a
VM
makes
cougar
neighs
performance
slower
hypervisors
can
be
difficult
to
configure,
especially
the
first
time
and
incompatible
with
VPNs
that
block
when
ip's
does
not
yet
support
multi-node
or
ipv6.
So
all
the
asterisks
here
are
things
that
we'll
talk
further
about
Indus
well,
I,
wouldn't
say
dispel,
but
at
least
discuss.
A
The
EM
overhead
is
actually
less
than
you
might
think.
I
was
really
blown
away
by
this
slide
and
I
didn't
actually
believe
it.
So
I
had
someone
else,
rather
the
numbers
for
me
and
we
both
came
to
the
same
conclusion
that
if
you
run
the
mini
qvm,
this
is
hyper
kit.
If
you
run
the
mini
qvm
but
don't
start
kubernetes,
it
is
almost
imperceptible
the
load
on
a
machine
so
we're
talking
on
a
six
core
laptop
like
this,
we're
talking
less
than
10%
of
one
core
of
overhead.
So
you
know
where
it's.
A
Actually,
you
can
see
the
yellow
line
and
the
red
line
are
almost
identical
to
the
baseline
load
of
the
system.
However,
when
you
run
kubernetes
on
a
machine
which
is
the
blue
line,
kubernetes
does
require
a
lot
of
resources.
Cubelet
will
consume,
it
does
a
lot
of
polling,
API
server
Opole.
We
have
a
lot
of
work
to
do
with
performance.
We
actually
have
a
contributor
who
is
worried
about
performance
more
than
anything
else,
and
that's
what
she
wants
to
do.
A
A
A
How
would
anyone
use
this,
and
now
it's
only
like
you
know
twelve
percent,
so
still
a
big
issue
that
we
want
to
solve,
but
if
we,
the
the
dark
pink
in
there
is
start
latency
which
also
much
lower
than
he
used
to
so
on
the
documentation
front,
what
do
users
want
to
see
not
too
surprising
here,
access
networking
access
tutorials
would
be
a
huge
win
for
us
if
you're
interested
in
any
of
these,
we
would
love
to
see.
Prs.
A
Tutorials
is
also
another
big
area.
So
what
features
do
people
want?
Anyone
here
care
about
multi-node
yeah?
If
you
it
turns
out
to
be
the
most
requested
feature
for
mini
cube,
which
kind
of
blew
us
away
and
one
of
the
other
ones
was
like.
Can
you
get
rid
of
the
VM
part
I?
Don't
like
it,
so
we're
gonna
actually
do
some
demos
here
where
we
demonstrate
multi,
node
and
VM
free,
so.
A
A
A
A
It's
getting
a
better,
it's
it's
a
better
idea.
It's
a
much
better
idea
now
than
than
none
windows.
Support
for
us
used
to
be
not
so
wonderful.
There
was
a
long
time
where
we
didn't
have
anybody
using
Windows
who
is
contributing.
Now
we
actually
have
some
contributions
from
from
Microsoft
from
random
college
kids.
In
India
we
got
a
bunch
of
like
great
contributions
out
there,
so
things
are
getting
much
better.
The
hyper-v
support
today
compared
to
a
year
ago,
is
miles
better.
However,
we
still
have
confounding
windows
errors.
A
We
still
tell
users
to
keep
deleting
your
VM
then
complain
about
the
config
file
missing
and
there's
a
lot
of
things
to
do.
On
Windows,
Great
Firewall
of
China
is
a
big
issue
for
us.
Kait,
CEO
or
GC
arcade
cio
is
not
accessible
in
China
and
that's
where
we
pull
our
images
from
that's
a
problem.
We
have
some
works.
There
are
workarounds
for
it,
but
we
don't
like
auto,
detect
and
suggest
nested
virtualization.
A
Don't
do
it
just
don't.
So
we
have
a
lot
of
people
complaining,
I'm
running
the
VirtualBox
driver
under
VMware
on
a
Windows
machine,
and
it's
complaining
that,
via
a
virtual
box,
won't
work
I'm
like
yeah,
there's
there's
a
any
incompatibility
between
the
two
nested
KVM,
on
the
other
hand,
works
wonderfully
and
it's
what
we
use
for
CI
non
driver.
We
don't
actually
clean
up.
Even
we
say
we
do
and
is
we
made
a
mistake
and
relying
on
cube
ATM
reset
to
do
all
the
cleanup
and
that's
not
enough
services,
confusion.
A
Some
people
is
mostly
for
kubernetes
newbies.
They
don't
understand
the
services
model
and
they're
like
hey,
I,
tried,
mini
coop,
I,
don't
understand
services
help
me
out.
Kvm
and
liberabit
sometimes
have
baffling
networking
issues.
I
have
run
into
ones
where
I
did
the
Windows
user
approach
and
rebooted,
because
I
spent,
like
30
minutes,
trying
to
figure
out
why
the
KVM
bridge
was
no
longer
routing
packets
and
I'm,
hoping
that
somebody
who
is
much
more
knowledgeable
than
I
am
will
solve
it.
Vpn
firewall
issues,
that's
a
biggie
and
people
running
obsolete,
kubernetes
Flags.
A
It
turns
out
that
when
you
do,
when
you
try
to
pass
a
flag
from
like
kubernetes
1/6
to
1/16
it,
they
changed
the
flags
and
we
don't
give
the
best
error
messages
and
we
take
too
long
to
deliver
them
so
mini
cubile
and
invariably
fail
on
you
someday,
and
you
might
want
to
know
how
to
look
into
why
it's
doing
that.
So
one
of
the
most
important
things
is.
We
have
this
also
log
to
standard
air
flag
to
come
and
fly
with
go.
A
It
will
tell
you
the
actual
it'll
show
you
the
hypervisor
logs,
which
are
crucial
for
debugging.
Why
hypervisors
are
failing
under
you,
you
might
get
a
cubelet
failed
message
is
kind
of
unlikely,
but
if
the
cubelet
does
fail,
there's
the
mini
a
few
blogs
command
and
there's
a
bunch
of
sections
and
the
cubelet
section.
A
If
jobs
are
not
scheduling,
it's
it's
probably
because
you're
out
of
RAM
cube,
CTL
described
node
actually
does
a
great
job
of
telling
you
here
are
the
resources
that
you
are
constrained
on
we'd
like
to
actually
build
some
of
that
that
knowledge
into
mini
cube
status,
API
server,
crashed
you
probably
whom
mini
cube
blogs.
Api
sir,
this
should
say:
API
server
did
not
appear
a
common
issue
for
none
users.
A
One
of
the
problems
with
a
non-driver
is:
we
depend
on
your
environment
to
mimic
exactly
our
VM
environment
and
people
have
different
cgroups
drivers,
different
docker
configs
things
can
get
pretty
wild,
but
those
sections
connectivity
failures,
almost
always
a
VPN
or
firewall.
We
also
have
this
mini
mini
cube
logs
problems
that
has
some
regular
expression.
Filtering
we'd
like
to
do
better
so.
A
So,
what's
on
tap
for
20/20,
so
we
have
multi
node,
we
decided
to
go
there.
Finally,
we
by
popular
appeal-
we
should
have
probably
went
there
a
year
ago,
but
better
late
than
ever.
Improving
VM
free
deployments
we'll
be
demoing
that
in
just
a
moment
where,
instead
of
using
the
non
driver,
you
can
just
deploy
straight
to
docker
or
pod
man
very
similar
to
the
kind
model
we
want
to
make
mini
cube
background
friendly.
A
For
many
users
it
would
be
useful
to
have
a
you
of
some
sort,
so
we're
talking
about
a
very
limited
UI
to
give
you
resource
when
you
run
out
of
quota
or
when
your
pods
aren't
scheduling,
automating
hypervisor
installation,
we'd
like
to
revamp
add-ons
totally
and,
of
course,
decreasing
overhead
decreasing
latency.
Looking
at
a
post
live
machine.
World
Lib
machine
was
something
that
docker
had
started
and
then
dropped
when
they
released
docker
for
desktop,
which
was
sad
but
we'd
like
to
look
at
other
technologies
that
allow
us
to
use
the
same
drivers
or
same
support.
A
The
same
hypervisors
anyways
terraform
would
be
one
possible
place
that
we
would
go
kernel,
assisted
mounts.
Anyone
use
the
mini,
cube
mount
command,
it's
slow
as
hell.
Isn't
it
yeah?
It
is
so
it's
all
user
space
using
nine
PFS.
We
would
like
to
replace
it
with
CIFS
and
NFS
support,
among
others.
Some
hypervisors
actually
have
cool
mounting
technology
support
so
yeah.
A
Actually
we
just
had
a
contributor
who
sent
us
a
PR
for
CIFS
still
in
review
and
improve
CNI
integration
with
multi
node
comes
kind
of
a
requirement
for
improved
CNI
integration,
we'd
like
to
make
sure
that
you
can
basically
just
tell
us,
you
know,
see
and
I
equals
we've
or
Flamel
or
whatever,
and
just
just
do
ever
do
the
needful.
So
one
issue,
that's
not
on
the
roadmap
yet,
but
is
a
personal
pet
peeve,
so
I
don't
bring
it
up.
So
I
could
convince
other
people
that
we
should
put
it
on.
A
The
road
now
is
configuration
persistence.
Our
model
is
very
confusing
to
me.
Right
now,
start
flags
are
not
persistence
across
reboots.
There's,
not
a
separation
of
cluster
creation
and
restart,
but
there
is
this
mini
cube,
config
command.
One
thing
that
has
a
lot
of
people
have
complained
about
is
when
they
upgrade
mini
cubes,
that
it
automatically
upgrades
their
cluster
to
the
new
kubernetes
release,
we're
changing
that
we
got
some
PRS
from
another
contributor
so
that
it
will
be
sticky
across
restarts
and
upgrades
and
add-on.
A
C
Everybody
my
media
I,
want
to
start
with
expressing
my
gratitude
to
mini
cubes,
a
community
post,
the
users
and
the
contributors
since
I've
been
working
on.
This
truly
has
been
the
one
of
the
best
part
of
my
my
work.
So
earlier,
Thomas
talked
about
that.
One
of
the
complaints
on
mini
cube
is
the
start.
Bm
takes
long
longer
than
we
like
to
see.
So
we
have
a
non-driver.
C
That's
only
works
on
Linux,
but
if
we
has
something
better
than
non
driver,
it
would
work
everywhere
and
we'd
have
mini
cube
inside
a
container
instead
of
a
VM.
So
I'm
gonna
show
you
a
demo
of
mini
cube
for
the
first
time
working
inside
a
container.
So
if
I
gonna
use
my
bash
history,
so
so
I
started
implementing
this
feature
with
kind
but
I
had
a
lot
of
issues
of
you
know:
I
couldn't
get
set
the
CPU
size
or,
as
around
around
amount
of
RAM
or
different
container
runtimes.
C
So
I
wanted
to
have
kubernetes
in
any
type
of
containers,
not
just
docker,
so
I
I,
codenamed
cake
kubernetes
in
container
inside
the
code,
basically
see
it.
As
that
and
I
told
a
few
other
contributors
and
they
got
so
excited
about
this-
and
one
of
them
even
messaged
me
Josie
done-
is
that
he
said
I'm
gonna
take
vacation
from
work
to
work
on
this
and
I
got
excited
from
his
excitement.
So
I
I
got
him
excited.
He
got
me
excited
even
more.
There
was
an
explosion
of
excitement.
C
So
now
you
can
see
that
the
result
is
less.
Let's
see
it
keeps
etal
get
past.
I
can
type.
So
that's
fine,
because
there's
no
pods
here
right
now.
If
I
do
all
the
parts
you
can
see,
the
pods
are
coming
up.
So
let's
deploy
something
I'm
gonna
use
my
bash
for
you
guys
I,
don't
trust
myself
typing
right!
Now
there
you
go
so
I
created
deployment.
I'm
gonna
create
a
service
for
it.
From
my
batch
history,
there.
C
C
C
I
didn't
specify
that
the
profile
name
so
I
guess
came
on
there.
You
go
so
we
see
the
services
coming
up,
and
so
it's
acting
just
like
a
normal
company
cluster
and
this
will
be
working
on
all
type
of
operating
system,
not
just
Linux.
This
will
save
you
Sippi
overhead
in
Linux,
but
it'll.
Also
save
you
a
starting
time
in
in
other
than
Linux.
A
So
one
thing
I
want
to
mention
just
quickly
yeah,
so
I
think
you
mentioned
pretty
much
all
the
things
on
that
yeah
we
so
we
prototyped
with
kind,
but
we
ended
up
rewriting
it
to
match
our
free
feature
set
I'm
personally
hoping
for
a
kind
convergent
story
in
our
future
right
now
kind
is
hyper
focused
on
testing
and
I.
Think
we
can.
We
can
find
a
way
to
bridge
gaps
here,
container
deployment
shortcomings.
This
applies
to
all
the
types
swap
issues
can't
use,
gee,
visor
or
cata.
A
A
B
Little
over
time,
so
this
will
only
take
like
20
30
minutes.
It's
fun
we're
doing
some
some
one-handed
demoing
here.
So
this
is
so
multi
node
is
the
is
the
single
most
requested
uploaded
of
issue
on
the
in
the
history
of
the
the
mini
cube
repo?
So
if
you
could
actually
go
to
the
issue
asking
for
multi
node,
it
has
like
4000
of
those
or
something
so
so
I'm
gonna
use
this
really
really
simple.
Mini
coop
start
command.
B
The
thing
that
changed
is
basically
so
I,
basically
plumbed
out
all
the
important
stuff
that
mini
keep
start
was
doing
for
one
single
node
and
then
took
the
important
stuff,
and
now
we
can
add
any
number
of
worker
nodes.
This
does
not
work
with
several
control
plane
nodes,
so
the
master
is
still
going
to
be
receiving
requests
for
now.
That'll
change,
obviously,
and
how
this
works
is
right.
B
Now
we're
still
opening
up
the
we're
still
starting
up
the
first
VM,
but
each
worker
node
is
gonna,
have
its
own
separate
VM,
that's
kind
of
how
that's
going
to
work,
and
so,
as
the
second
VM
starts,
we
will
be.
Basically,
we
mentioned
that's
starting
up
a
VM
starting
up
a
mini
cube.
Cluster
is
just
running
cube,
ADM
in
it.
So
adding
a
worker
node
to
the
cluster
is
just
cube.
Idiom
join.
You
do
that!
Well,
I
apologize
for
all
the
debugging,
but
I
got
this
working.
What
time
is
it?
B
So
so
this
joint
command
that
we
we
pumped
out
that
we
were
printing
out.
We
basically
get
a
we
get
a
discovery,
token
from
the
master,
and
then
we
pass
that,
along
to
the
worker
node
to
pass
back
to
the
master
to
for
authentication
purposes.
This
is
stored
nowhere.
So
it's
actually
really
secure.
Not
the
medicube
needs
security,
but
security
is
good.
B
B
In
the
speaker,
lounge
next
door,
I'm
just
gonna,
try
it
again
and
see
what
happens
all
right.
Well,
what
was
gonna
happen?
What
was
gonna
happen
was
we
were
going
to
see
both
nodes
show
up
in
cube,
CTL
get
nodes,
and
then
we
were
going
to
apply
flannel,
so
they
would
both
actually
come
back
as
ready,
since
the
CNI
would
be
valid
and
then
had
a
deployment
in
a
service
to
basically
ping
the
IP.
That
I
was
getting
back
and
you
see
a
ping
pong
between
the
two
and
it
was
gonna,
be
really
magical.
B
C
B
Maybe
yeah,
maybe
a
delete
and
restart
would
work
it.
But
that's
that's
my
fault.
It's
alright!
It's
alright
it'll-it'll!
So
though,
they're
Christian
testing,
you
should
see
the
actual
code,
it's
real
good,
but
yes,
so
this
this
will
be.
This
will
be
ready
in
proper
form
early
next
year.
That's
my
that's
my
top
priority
along
with
Windows,
so.
C
A
A
A
That
we
want
to
make
sure
that
mini
cube
can
be
run
in
the
background,
so
we're
adding
this
mini
cube,
pause
command
that
basically
pauses
all
the
containers-
pauses
kubernetes
inside
the
VM
to
return
you
to
that
like
1
to
10
percent
overhead
on
the
system,
and
it
is
literally
400
milliseconds
to
pause
and
unpause
your
cluster.
So
if
you
are
tired
of
burning
your
lap,
this
is
gonna,
be
a
pretty
awesome
thing
for
you.
It
turns
out
it's
really
easy
to
implement.
This
is
basically
like
this.
Let
started
the
implementation
here.