►
From YouTube: TGI Kubernetes 082: Weaveworks Ignite
Description
Full notes and links at https://github.com/heptio/tgik/blob/master/episodes/082/README.md
00:00:00 - Welcome to TGIK!
00:04:52 - Week in Review
00:13:52 - Context/positioning Ignite
00:19:22 - Ignite blog post
00:22:31 - Installing Ignite
00:30:25 - Running our first VM!
00:33:11 - Using OCI images for VM images
00:42:46 - Ignite architecture review
00:43:06 - Storage with devicemapper
00:50:05 - Using Docker and exploring networking
01:05:52 - Launching k8s into Ignite VMs
01:32:50 - GitOps and Ignite
01:43:10 - Wrapping up!
Come hang out with Joe Beda as he does a bit of hands on hacking of Kubernetes and related topics. Some of this will be Joe talking about the things he knows. Some of this will be Joe exploring something new with the audience. Come join the fun, ask questions, comment, and participate in the live chat!
This week we will look at Weaveworks Ignite (https://www.weave.works/blog/fire-up-your-vms-with-weave-ignite) a project to provide a docker like experience around lightweight VMs built on AWS Firecracker. This was a project building on work that Lucas Käldström has been doing for a while.
A
Hello,
hello,
hello,
hello,
welcome
to
another
TGI
kubernetes.
My
name
is
Joe
Beda
and
I
am
your
host.
Today
this
is
a
blurry.
This
is
a
weekly
YouTube
livestream
that
I
or
some
other
folks
on
our
team
will
do
where
we
work
to
explore
the
kubernetes
space.
Try
out
new
things
learn
some
stuff
together.
You'll
see
everything
happen
in
excruciating
detail,
which
is
part
of
the
fun.
A
So,
first
of
all,
if
I
need
a
sound
check,
if
y'all
can
hear
me,
let
me
know
and
I
also
like
to
say
hello
to
all
the
folks
that
are
saying
hi
in
the
chat.
So
we
have
sent
deep
from
Poland
good
to
see
you
Martin
from
the
Netherlands
Rory
from
Scotland.
All
these
people
who
who
are
spending
their
Friday
evening
with
me,
I,
appreciate
it
Sean
Emmerich
on
from
Turkey.
Let's
see,
we
got
suresh
from
hamburg
ruff
from
bangladesh,
marcin
from
Cracow
M&E
from
Strasbourg.
A
Let's
see
oh
and
Sean
says
that
we
sound
good.
So
that's
good
Rory.
How
is
gone?
Oh
eh,
like
you're,
saying
it
sounds
good
Omar
from
Israel
first
time
with
the
livestream
yeah
11
p.m.
but
thanks
for
coming
and
staying
with
us
for
a
little
bit.
If
you
miss
stuff,
you
can
obviously
catch
up
on
things
later.
A
Let's
see
a
Elko
from
the
Netherlands
Ivan
from
Belgrade
Chris
from
India.
Oh
thank
you
all
for
joining
me.
I
really
appreciate
it.
Today's
topic
is
going
to
be
a
new,
offering
open
source
project
that
was
just
released
a
couple
of
days
ago
from
we've
works
called
ignite.
It's
essentially
a
framework
around
this
other
project
that
we
had
done
some
work
on
called
that
we've
done
an
episode
on
called
AWS
firecracker
for
managing
VMs
and
we'll
get
into
the
details
there.
A
So,
let's
see
we
have
a
Thorsten
how's
it
going,
and
then
Lucas
from
Finland
here
is
is
actually
the
author
of
the
project.
Longtime
contributor
to
the
kubernetes
project
help
help
build
out
cube
a
DM
plus
a
whole
bunch
of
other
things,
and
it's
just
back
from
a
stint
in
the
the
Finnish
military
and-
and
this
is
the
first
big
piece
of
work
that
he's
done
since
he
had
since
he
finished
up
there.
A
So
thanks
for
joining
us
Lucas
and
helping
out
as
we
explore
your
tool,
you
can,
let
me
know
where
I'm
screwing
things
up
and
and
hopefully
we'll
have
a
have
a
good
time.
Looking
at
it
together,
we
have
a
show
car
from
Atlanta
and
Mark
from
Melbourne
good
to
see
ya
so
Wow
Melbourne,
it
is
like
you
know,
6
a.m.
in
Melbourne,
6
a.m.
ish
right.
A
Thank
you
for
for
getting
up
early
here,
alright,
so
a
lot
of
times
what
we,
what
we
try
and
do
when
we
start
out
at
TGI
K,
is
I,
go
through
and
talk
about
what's
new
in
the
sort
of
kubernetes
world
over
the
last
week
or
two,
we
didn't
have
an
episode
last
week
because
of
the
u.s.
fourth
of
July
I
did
not
light
anything
on
fire,
which
is
good.
A
We
actually
went
to
a
weighing
out
in
this
this
town
outside
of
Seattle
called
Leavenworth,
which
is
like
a
Bavarian
village
up
in
the
mountains
outside
Seattle.
It's
it's
kind
of
crazy,
but
it
was
a
lot
of
fun.
It
was
a
beautiful
wedding.
So
that's
what
I
was
doing.
I
wasn't
actually
doing
TGK
then
Roberto
from
Pennsylvania
Nick
from
Sheffield
England
Shaw
from
Stockholm
spiral
dynamics
from
France,
all
right,
awesome,
yeah,
so,
okay!
A
So,
let's,
let's
jump
in
so
oh
and
one
of
the
things
that
we
have
here
and
I'll
post
it
again
is
we
have.
Let
me
switch
to
the
screen
is
that
we
have
we
use
this
tool
called
hack
MD,
where
we
can
sort
of
do
collaborative
editing
of
a
doc
in
this
is
where
I
put
all
the
links
that
I'm
going
to
be
working
with
and
if
folks
want
to
like
crowdsource
actually
did
I
go
through
9.
A
Ok,
so
there's
a
bunch
of
people
looking
at
it
I
can
you
all
edit
access
right,
signed
in
users?
Ok,
cool
so
yeah,
so
you
all
can
can
go
through
and
if
there's
other
notes,
if
there's
stuff
that
you
want
to
share,
if
there's
like
time
codes
that
are
interesting,
you
know
if
you
want
to
help
out
and
put
it
in
the
in
the
hack
MD.
That
would
be
wonderful
and
then
I.
Take
this
and
I
put
this
in
our
github
repo
after
the
episode
all
right.
A
So
let's
just
going
through
sort
of
the
the
news
of
the
week.
So
the
big
news
here,
obviously
or
this
is-
is
big
news-
that
sort
of
in
the
business
side
of
the
house.
Let's
see,
oh
geez
focus
on
the
wrong
thing
there.
Okay,
there
we
go
is
IBM
completed
the
acquisition
of
Red
Hat,
obviously,
Red
Hat,
you
know
through
openshift,
is
a
big
contributor
to
kubernetes.
So
I
think
you
know
the
the
name
of
the
game
here
and
there's
a
bunch
of
analysis
out
there.
For
those
who
are
interested.
A
Is
that
it's
going
to
be
you
know,
red
hat's
gonna
continue
to
be
the
independent.
Yet
there's
going
to
be
a
lot
of
good
IBM
stuff
coming
through
I.
Think
everybody's
still
waiting
to
figure
out
what
exactly
that
means,
but
congratulations
to
all
the
folks,
all
our
friends
at
Red,
Hat
and
IBM,
for
for
seeing
this
happen,
these
things
before
an
acquisition
like
this
happens
because
I
have
you
know
when
we
sold
hep
tio
2
VMware.
A
These
things
were,
you
know
it's
sort
of
a
limbo,
so
I'm
sure
they're
happy
to
finally
get
this
thing
done
and
and
and
be
able
to
start
tackling
the
hard
problems
of
integrating
where
it
makes
sense
all
right.
So,
let's
see
so
we
have
other
I'm,
not
sure
I
may
be
doing
these
in
a
weird
order.
Let's
see
what
do
I
got
here,
yeah,
okay,
so
this
is
a
blog
post
Lucas
and
is.
A
And
apparently
the
screencast
is
running
on
top
of
the
Ignite
vm.
So,
like
you
know
we're
bringing
things
full
circle
there,
Federico
how's
it
going
serve
good
to
see
ya
all
right,
let's
see
and
so
Sean
says
currently
watching
the
f5
engine
X
acquisition
from
the
inside,
and
it's
super
interesting
exactly
right.
It's
one
of
those
like
before,
while
you're
doing
all
the
due
diligence
and
all
the
regulatory
waiting
times,
it's
a
really
weird
place
to
be
so
so
there's
there's,
certainly
a
certain
amount
of
like
alright.
Now
we
can
get
started
once
the
once.
A
The
acquisition
happens
all
right,
so
this
is
the
toy
I
wanted
to
share
with
folks
download
kubernetes
com.
So
one
of
the
the
VMware
engineers
xft
Oh
folks
cobbled
this
together
really
cool.
This
essentially
is
an
easy
way
to
get
access
to
all
the
various
binaries
that
you
might
want
for
different
releases.
A
Now
all
of
these
things
are
stored
out
on
a
Google
Cloud
Storage
bucket,
using
a
common
scheme,
but
like
remembering
how
to
actually
construct
the
paths
to
download
is,
is
kind
of
a
pain
and
so
Chuck
Chuck
haw
put
this
together,
and
so
you
can
go
through.
It's
like
well,
I
want
to
download
the
you
know
the
Darwin
version
of
cube
control
for
the
1.15
boom.
There
I
go
through
and
you
can
see
you
know,
there's
a
link
that
comes
down
there.
A
That
goes
directly
into
the
release
buckets
out
of
that,
but
you
can
also
use
this,
for
you
know
the
the
server
binaries
there
also
so
like
Linux
amd64,
the
1.15
super
super
useful
to
be
able
to
actually
get
access
to
that.
So
this
is
really
cool.
This
is
something
that
I
sort
of
in
the
back
of
my
head,
I
kind
of
wanted
to
build
for
a
long
time
and
it's
like
independently
chuck
just
went
off
and
did
it
because
that's
that's
how
Chuck
roll
so
very
cool,
shot
good
job.
A
Ok!
So
today,
I,
don't
know
what
time
today,
but
you
probably
still
have
an
hour
to
this
time,
is,
is
the
last
day
for
you
submit
a
CFP
if
you
want
to
speak
at
cube
con
in
San,
Diego
and
so
I
just
figured.
This
is
the
last
call
right
now.
If
you
work
well
under
pressure,
you
can
maybe
all
together
something
and
actually
get
it
out
there
and
get
something
submitted
cube.
Con
has
been
a
fascinating
event
to
watch,
grow
and
evolve.
A
You
know
every
time
I
go
I'm,
always
blown
away
with
the
energy,
with
the
creativity,
with
the
innovation
that
everybody
brings
to
it.
So
if
you
have
something
that
you
want
to
want
to
talk
about
revolving
around
kubernetes
or
cloud
native
in
general,
this
is
a
great
place
to
do
it
and
yeah.
So
please
submit
some
some
talk
proposals
it.
A
And
so,
if
you
know
folks
who
are
sort
of
looking,
there's
Craig
his
picture
up
there,
it's
not
a
bad.
It's
really
hard
to
take
a
good
picture.
Craig
I
got
to
be
honest,
that
one's
not
bad
but
but
there's
a
this
is
sort
of
a
summary
of
where
we're
doing
kubernetes
stuff
there.
If
you
know
folks
who
are
in
the
VMware
world
or
a
few
yourself
are
part
of
the
VMware
world,
this
is
a
great
resource.
A
If
you're
going
to
be
at
the
VM
world,
let's
see
other
stuff
I
want
to
go
through
quickly,
so
steep
slope
put
together
a
guide
on
running
contour,
ingress
controller
envoy
based
ingress
controller
on
top
of
kind,
so
that
you
can
actually
go
through
and
start
playing
with
some
of
the
the
features
of
contour
there.
So
a
good
guide
coming
there,
and
then
this
is
really
cool.
So
this
is
upstream
is
Cal.
Here,
helped
drive
the
process
of
getting
dual
stack
support
into
kubernetes.
A
Robottom,
what's
going
on,
can
you
can
y'all
hear
me
fine,
all
right
if
there's
some
problems
at
the
stream?
Let
me
know
so
so
this
is
called
who
did
a
lot
of
the
work
and
then
here's
the
the
link
to
the
okay.
So
it's
okay
for
everybody,
Sean
I,
don't
know!
What's
going
on
I'm
sorry,
man,
okay
cool,
so
so
this
is
a
I
P
v6
dual
stack
support.
So
when
folks
talk
about
dual
stack,
what
they
mean
is
that
your
computer
can
talk
both
ipv6
and
ipv4.
A
A
Always
a
little
bit
tricky
these.
You
never
quite
know
how
this
stuff
is
gonna
work
and
there's
always
always
a
little
bit
of
trickiness
doing
this
stuff.
So
hopefully
this
this
will
improve
things.
A
little
bit.
I
made
a
change
there
to
the
bitrate,
okay,
so
alright,
so
yeah,
so
dual
stack
supports.
So
it's
really
cool
to
see
this
getting
in
there.
I
think
this
is
one
of
those
things
where
early
on
in
kubernetes.
A
We
didn't
think
about
v6
as
much
as
we
should
have,
and
so
it's
really
great
to
see
folks
start
to
start
to
actually
sort
of
fill
in
the
gaps
there,
which
is,
which
is
really
really
cool.
Okay,
so
that
is,
are
there
any
other,
any
other
sort
of
an
you
know,
news
items
or
releases
or
stuff
that
people
think
that
I
mix
missed?
A
All
right,
so
all
right,
so
here's
I'm
going
to
go
ahead
and
start
jumping
into
into
ignite
talking
about
sort
of
what
it
is
and
how
it's
working
as
I
demo
stuff.
What
I
have
here
is
I
launched
on
Amazon
a
c5
metal
instance
type
okay,
solo,
Mattie's,
saying
hey
Joe
before
you
start.
Can
you
comment
on
how
stable
firecracker
is
gone
during
episode?
58
you
mentioned
was
still
a
bit
rough
around
the
edges.
I
haven't
played
with
it
since
then
I
think
I'm.
If
I
remember
correctly
episode
58,
it
was
awhile
ago.
A
Definitely
using
it
in
you
know
like
Lukas
is
saying
in
production
type
settings
and
it's
been
working
well
so
I
think
fire-control
is.
That
is
the
thing
that
I
was
missing
in
that
first
episode
that
really
helps
fill
in
some
of
the
usability
gaps.
That
I
you
know
saw
is
rough
around
the
edges
there
and
some
of
that
I
think
is
just
you
know.
As
people
come
to
this,
that's,
like
you
know,
I
wouldn't
be
surprised.
I
haven't
looked
that
you
know.
A
So
I
think
you
know
definitely
listen
to
folks
like
like
Lucas,
who
have
been
using
it
more
directly,
alright,
so
so
yeah
so
I
created
a
an
Amazon
c5
metal
instance,
which
runs
me
somewhere
on
the
order
of
four
bucks
an
hour
and
it's
been
running
since
about
10:00
a.m.
and
so
I
think
this
is
probably
going
to
be.
You
know
the
most
expensive,
TGI
K
episode
ever
so
yeah,
hopefully
I
won't
forget
to
shut
it
down
when
all
said
and
done.
A
But
the
thing
to
keep
in
mind
is
that
if
you
do
want
to
play
with
things
like
fire
cracker
or
if
you
want
to
play
with
ignite
you're
gonna
need
a
place
where
you
get
access
to
hardware
virtualization
features,
because
that's
that's
what's
going
on
here.
So,
okay,
so
folks
are
still
hearing
the
robot
stuff.
Let
me.
A
A
How's
this
is
this
any
better,
maybe
that'll
actually
work.
This
is
now
going
through.
My
like
fancy
microphone,
yeah
well,
is
one
of
the
things
is
that
like
whenever
you
do
OBS
and
you
plug
in
on
plug
stuff,
sometimes
it
actually
gets
confused.
Hopefully
that's
actually
taught
sounding
better.
Even
for
those
who
who
didn't
see
the
the
robot
stuff,
all
right
cool,
that's
good
to
hear
all
right,
so
we're
better
off
there
and
the
levels
look
like
they're,
pretty
good,
too
alright.
A
So
the
thing
to
keep
mind
if
you're
gonna
play
with
any
of
this
stuff,
you
need
a
place
where
you
can
actually
do
hardware
virtualization,
and
so
so,
if
you're
running
on
a
sure
they
support.
What's
called
nested
virtualization
out
of
the
box.
If
you're
running
on
something
like
if
you're
running
on
GC,
p
GC
p
also
supports
nested
virtualization,
where
it's
like,
you
can
launch
a
hardware
accelerated
vm
from
within
a
vm
to
do
it
with
GC
p.
A
So
it's
a
little
bit
tricky
to
actually
get
access
to
this
stuff,
and
it's
you
have
to
make
sure
that
you
yeah
so
Nick
saying
that
packets
body
instances
probably
could
be
a
good
way
to
play
with
it.
Also,
oh,
really
faint.
Now:
okay,
really!
Okay,
let
me
see
if
I
can
turn
up.
Sorry
for
all
this
folks
turn
up
the
gain:
chak
chak,
chak,
chak
chak
part.
Hopefully
that's
gonna
be
a
little
bit
louder
for
y'all.
A
Actually,
I
can
do
this
and
yes
there
we
go
okay,
hopefully
oh
now,
I'm
in
the
red,
okay,
they're,
real,
alright!
So
that's
the
first
thing
to
keep
in
mind
here:
alright,
so
now
here's
the
blog
post
that
announces
this
and
Alexis
post
of
this,
but
I'm
assuming
Luke,
has
had
a
lot
going
on
with
this.
It
talks
about
what
ignite
is,
and
so
what
ignite
is
so
now
we're
too
loud?
Okay,
I've
never
been
accused
of
having
to
quiet
a
voice.
Let's
see
how's
that
Oh
still
too
loud
still
too
loud.
A
Okay,
that
looks
like
it's
probably
pretty
good,
okay,
so
so
the
so
the
question
is
ignite
is
a
rapper.
It
creates
an
experience
around
firecracker,
VMs
and-
and
this
is
not
I-
think
what
a
lot
of
people
assumed
it
was
going
to
be.
So
there
are
efforts,
and
actually,
if
you
go-
and
if
you
look
here
in
the
firecracker
organization,
there's
fire
craft
container
D,
which
is
essentially
a
way
to
launch
firecracker,
VMs
and
host
containers
inside
of
a
VM,
so
you're.
Essentially
then
using
firecracker
as
a
VM
based
isolation
mechanism
for
a
container.
A
That's
I
think
what
a
lot
of
folks
think
about.
As
you
start,
mixing,
containers
and
firecracker
is
that
you're
going
to
use
the
VMS
as
a
as
a
wrapper
now.
Other
projects
that
are
sort
of
similar
in
that
vein
would
be
things
like
like
cata
containers,
which
is
a
project
for
using
KVM
in
other
ways
and
I
believe
there's
some
integration
to
use
firecracker
as
the
VM
technology
sitting
underneath
cata
containers.
A
Another
sort
of
similar
thing
would
be
something
like
the
cloud-based
vm
container
as
a
service
that,
under
the
covers,
are
fundamentally
based
on
DM.
So
these
would
be
things
like
as
your
container
instances
or
in
the
in
the
in
the
Amazon
world.
This
the
Fargate
is
the
name
of
their
product
and
and
my
understanding
is
that
Fargate
actually
does
use
firecracker
under
the
covers.
Also
so
so.
At
the
end
of
the
day,
like
a
lot
of
people
are
looking
at
firecracker
as
a
way
to
actually
create
a
container
around
VMs.
A
What
ignite
is
is
actually
the
other
way
around.
How
can
we
actually
create
something
that
is
more
of
a
traditional
vm,
but
we
launch
and
we
manage
those
things
using
some
container
technologies
and
using
some
of
the
experiences
that
folks
have
gotten
used
to
with
containers.
So,
instead
of
it's
like
how
do
we
make
containers
use
VMs
for
isolation?
A
Let
me
login
there
okay,
so
this
is
a
big
old
machine.
If
we
do
a
cat,
proc
CPU
info,
you
can
see
like
there
are
96
processors
on
this
machine.
That's
that's
that's
a
lot
48
physical
and
but
with
hyper-threading.
It
shows
up
in
the
OS
as
96.
Ok,
this
is
way
more
machine
than
I
need
here.
So
it
says,
first
of
all
make
sure
that
we
have
the
right.
Okay,
we
have
VTX
support,
make
sure
that
we
have
k
vm.
A
A
And
we
asked
you
to
assess
a
night
version:
okay,
so
we're
up
and
running
firecracker
version
0.17.
This
is
a
great
install
experience
here,
because
essentially
we
went
from
zero
download
one
binary
fire
crackers
included
with
it.
There's
not
a
lot
of
other
random
dependencies
that
I
have
to
install
okay,
and
so
Lucas
is
saying
that
okay,
so
other
dependencies
that
we
might
need
host
requirements.
A
Well,
that
was
okay,
so
we
got
all
that
I
think
we're
probably
good
there.
Mountain
you
mount
is
probably
their
that's
their
that's
their.
We
need
docker.
These
are
probably
their
yeah
DM
setup
for
device.
Mapper
tar,
as
this
agent
get.
We
should
be
good,
okay,
okay,
so
tar
will
set
up.
The
doctor
will
set
up
all
the
the
IP
forwarding
and
stuff
okay
I
haven't
installed.
Docker
like
this
in
a
while,
install
doctor
a
boon
to
18
204.
Let's
see
what
I
got
to
do
here.
A
A
So
we're
Verto
says:
oh
it's
1809,
now,
okay,
that
it'd
be
awesome.
If
I
can
run
this
on
old
or
older
hardware
got
some
2010
boxes,
I
think
2010
should
work.
It
might
be
a
little
bit
old.
The
not
all
processors
support
the
virtualization
instructions
and
I
think
over
time,
Intel's
been
adding
those
to
more
and
more
processors.
But
if
you
have
really
old
hardware
like
that
old,
it
may
or
may
not
work
for
you.
Okay,
so
Lucas
a
saying:
apt-get
install,
docker,
dot,
IO
Oh
like
that
it
works
okay.
A
What
is
it
fan?
Does
that
actually
control
the
fan?
Can
I
make
the
fan
run
faster
than
some
Amazon
data
center
somewhere?
If
I
spin
it
fast
enough,
does
it
light
it
on
fire?
Can
I
burn
down
a
rack,
so
there's
some
slight
difference
between
a
mood
to
die?
Oh
and
Dockers
docker,
but
they
both
seem
to
work
okay.
So
hopefully
we
won't
hit
any
issues
there.
I
don't
know.
Arun
can
I
burn
down
my
Amazon
datacenter
by
by
going
through
and
hacking
the
BIOS
from
within
the
virtualization
should
I
not
give
people
ideas.
A
A
So
this
is
okay,
so
so
just
a
little
bit
of
background
on
firecracker,
while
we're
waiting
for
this
for
firecracker
is
a
minimal
virtual
machine
manager
vmm
for
KVM
and
so
with
KVM
based
VMs.
There
is
essentially
a
bunch
of
kernel
capabilities
built
into
the
Linux
kernel
that
largely
give
you
access
to
the
hardware,
accelerated
features
that
are
built
into
your
process.
A
So
it
takes
these
things
built
into
the
processors
and
it
exposes
them
up
through
through
a
device
interface
called
/dev
KVM
and
gives
you
access
to
those
so
that
you
can
actually
go
ahead
and
get
access
to
the
hardware.
Acceleration
features
most
people
when
they
start
running
VMs
using
KVM.
They
pair
it
with
something
like
QEMU,
ki,
mu
q
mu.
A
Is
this
ancient
project
for
doing
processor,
emulation
it
does
all
sorts
of
emulation,
not
just
x86
and
and
but
usually
what
you
do
is
use
QAM
you,
you
don't
use
all
of
QAM
you
for
a
lot
of
the
sort
of
hardware
accelerated
stuff.
It
then
pairs
with
KVM
to
be
able
to
run
a
generalized
vm
now
qmu
is
because
it's
trying
to
do
so.
Many
things.
A
It's
a
little
bit
scary
from
a
security
point
of
view,
and
it's
has
a
lot
more
than
you
actually
need
to
do
sort
of
a
minimal
vm,
and
so
that
impacts
things
like
startup
time
and
and
that
type
of
thing,
and
so
what
what
firecracker
is
is.
It
is
the
bare
minimum
vm
m
necessary
to
run
something
with
with
qmu
I
mean
with
k,
vm
pair
it
down
as
much
as
possible.
So
the
only
devices
that
are
attached
to
the
vm
are
a
variety
block
driver
Verdi
own
net.
A
So
this
means,
instead
of
emulating
hardware,
there's
actually
para
virtualized
drivers
that
it
works
with.
So
it
doesn't
have
a
lot
of
that
sort
of
legacy
that
you
get
from
hardware
interactions,
serial
console
and
then
a
keyboard.
That's
only
used
to
do
one
thing,
reboot
the
thing
and
so
really
really
minimalistic,
vmm
or
so
Lucas
saying
yeah
the
history
around
this
ok,
so
Lucas
is
history,
so
he
spent
like
yeah
like
I
was
saying
military
service
in
Finland.
A
A
Building
some
automation
on
top
of
fire
cracker
learned
a
lot
and
then
my
understanding
is
correct
me
if
I'm,
wrong,
Lucas
and
then
after
building
a
bunch
of
that
experience,
is
left
the
military
and
then
started
from
scratch
and
rebuilt
a
lot
of
that
stuff.
For
what
we
see
here
all
right,
so
we
should
be
good
to
go,
and
so
that's
the
installation
guide
getting
started
all
right.
Let's
just
try
and
make
this
work,
so
I
can
do
ignite
run.
A
A
Okay,
fascinating,
ok,
so
we're
seeing
some
so
we
downloaded
an
image
somewhere
imported
the
OCI
image
and
all
I
have
a
good
clue
in
terms
of
what's
happening
behind
the
scenes
here,
but
will
actually
go
ahead
and
okay
pulling
another
image.
Now,
let's
go
ahead
and
it
launched
a
VM
and
the
VM
has
an
ID
and
it's
running
in
a
container.
So
there's
there's
layers
going
on
here.
Let's
see
and
we'll
go
into
the
detail.
Search
so
link
has
just
dropped
this
okay.
So
this
is
the
docker
file
for
that
image
that
we
just
launched.
A
All
right,
so
this
is
okay.
This
is
this
is
bacon
my
noodle.
This
is
some
weird
stuff
here
and
then
here
is
images
kernels.
This
is
the
kernel
builder,
so
this
actually
builds
okay
got
it
got
it.
Okay,
I
see,
what's
going
on
here,
I'll
explain
this
Lucas
I
see
what's
what's
happening,
okay,
so
so
ignite
SSH
my
VM
boom
I'm
actually
running
in
a
VM
now,
which
is
actually
pretty
darn
cool.
That's
how
fast
and
easy
it
is,
and
so
I
did
the
SSH.
A
A
Yes,
I
will
draw
some
diagrams
on
the
stack,
actually
there's
a
design
doc
that
that
has
some
interactions
in
terms
of
what's
happening.
So
the
user
experience,
here's
okay,
so
the
first
thing
I
want
to
actually
recognize
here
is
yeah.
The
doctor
requirement
is
a
head-scratcher,
but
I'll
explain
it
in
a
second,
because
it's
actually
fascinating
how
Lucas
essentially
hacked
on
you
know,
use
docker
as
part
of
the
bootstrapping
here
and
there's
other
good
things
that
you
get
around
docker.
That's
that's
super
cool
yeah,
so
it's
it's!
A
It's
firecracker
running
inside
of
a
docker
container,
we'll
get
to
it
because
that's
I
think
part
of
the
fun
okay.
So
so
the
idea
here
is
that
these
are
real
vm's.
These
are
meant
to
be
VMs.
These
are
not
meant
to
be
containers,
and
so
the
first
thing
that
the
Lucas
dropped
in
here
was
this
thing
here
is.
This
is
a
this.
Is
the
root
filesystem
for
VM?
A
That
idea
of
like
I,
want
to
have
a
tarball
I
want
to
name
it.
I
want
to
push
it.
I
want
to
pull
it.
I
want
to
push
and
pull
Delta's
to
it.
That's
a
really
useful
set
of
features
regardless
of
containers.
I
think
you
know
in
to
some
degree,
these
things
got
bound
into
containers,
but
distribution
of
named
signed,
large
tar
balls
is
actually
a
really
really
cool
thing,
and
so,
if
you
think
about,
if
you
use
something
like
vagrant,
the
first
thing
you've
got
to
do
with
vagrant.
A
A
This
is
the
open
container
initiative
and
there's
a
spec
on
this,
which
is
the
OCI
image
format,
and
one
of
the
things
that
we're
starting
to
see
is
that
the
OCI
image
format
and
the
way
that
you
upload
and
download
these
things
is
useful
for
more
than
just
containers.
You
can
use
it
for
things
like
VM
images
like
what
Lucas
is
doing
here,
but
you
can
also
use
this
thing
for
packaging
up
other
things.
A
When
you
look
at
something
like
like
scene
out
scene,
AB
is
actually
looking
now
to
use
the
OCI
image
format
for
packaging
up
a
group
of
container
images
versus
actually
dealing
with
with
just
a
single
image,
and
so
so.
The
first
interesting
thing
here
is
that
we're
reusing
all
of
the
plumbing
all
of
the
research
for
distributing
container
images
for
distributing
VM
images
and
and
I
don't
want
to
be
a
little
bit
clear
here
when
I
say
VM
images.
A
A
Ok,
Roberto's
saying
it
is
working
on
one
of
the
older
Zeon's
that
he
has
sitting
around.
So
all
right,
so
that's
that's.
The
first
thing
that
that's
happening
here
is
is
is
what
we
did
here
when
we
were
downloading
this
docker
docker
image.
We
were
actually
downloading
the
we're
using
that
docker
image
as
the
root
filesystem
of
the
VM
that
we're
going
to
launch
all
right
and
then
there's
an
image,
import
and
so
importing
the
OCI
image.
Is
this
thing?
So
this
thing
ends
up
to
being
220
megabytes
and
there's
an
image
ID
there
now.
A
The
next
thing
that
works
is
the
way
that
fire
cracker
works
is
that
it
doesn't
emulate
a
traditional
bios
that
you'd
find
in
there
in
a
typical
machine,
and
this
is
one
of
the
shortcuts
or
optimizations
or
improvements
that
fire
cracker
made
versus
using
something
like
QEMU
qm.
You
can
take
the
exact
same
image
that
you
could
like
boot,
a
machine
off
of
and
then
boot
that
using
a
traditional
bios
boot
process,
fire
cracker,
on
the
other
hand,
actually
skips
the
boot
process,
skips
things
like
grub.
A
This
way,
also,
we
did
it
to
actually
reduce
the
surface
area
with
GCE
when
we
were
first
launching
it
and
and
if
you've
been
using
ec2
long
enough.
You
know
you
used
to
both
have
it
both
a
and
AMI,
but
there
was
also
this
thing
called
an
8k
I
and
is
you're
an
Amazon
kernel
image,
which
is
essentially
was
the
same
thing
here.
So
the
next
thing
that's
happening
here
is
that
Lucas
is
taking
the
kernel
that
we're
taking
and
he's
using
OCI
and
docker
images
to
distribute
that
thing.
A
So
that's
the
first
thing:
that's
going
on
so
Rory
says
sounds
a
bit
like
windows,
subsystem
for
Linux
v2,
where
MS
is
actually
providing
a
kernel,
and
then
you
can
load
different
route
file
systems,
I
believe
with
and
I
haven't
taken
a
close
look
at
WS
l2,
but
I
believe
they're
not
actually
running
that
in
a
VM.
It's
actually
running,
you
know,
side
by
side,
I,
don't
know.
Maybe
it
is
running
on
the
VM,
maybe
maybe
that's
how
they're
doing
it
I
need
to
actually
dig
into
wsl
alright.
A
So
that's
the
first
thing
that
happens
here
and
then
he
creates
a
VM
and
then
and
then
actually
it
turns
out
that
there's
another
like
the
the
fire
cracker
process
actually
runs
in
a
docker
container
also,
and
so
he
has
to
go
ahead
and
pull
that
down.
And
finally
the
whole
thing
goes
ahead
and
gets
launched:
okay,
so,
okay!
So
that's
where
we're
at
with
that.
Let's
start
digging
a
little
bit
into
the
architecture
here
and
I
wouldn't
we're
gonna
do
the
get
up.
Stuff
is
interesting,
but.
A
A
Ignite
kernel,
we
can
see
here's
the
kernel
that
we're
using
when
kill
VMs,
logs,
PS
right
and
so
a
lot
of
so
what
you'll
see
here
is
that
the
experience
of
this
mirrors
the
docker
command
line,
and
so
at
the
end
of
day,
what
Lucas
is
doing
is
he's
bringing
sort
of
a
docker
type
experience
with
run
and
images
and
pushing
and
pulling
he's
bringing
that
stuff
to
the
vm
world.
So
it
mirrors
the
the
sort
of
the
docker
user
experience
here.
So
there
we
have
run
start
and
stop
and
version.
Ok,
so
managed
VM.
A
So
one
thing
that
I'm
not
seeing
here
is
disks,
so
I
don't
know
if
there's
probably
there's
plans
to
actually
think
about
separate
disks.
How
do
you
manage
those
things?
Oh
ignite
logs,
ignite
blogs?
My
BM
I
can't
spell
ignite,
that's
the
first
problem,
alright,
so
yeah,
and
so
when
I
look
at
the
logs.
This
is
actually
output
from
the
serial
port
of
the
VM
and
that
actually
shows
the
boot
logs.
A
And
so,
if
your
kernel
crashes,
for
whatever
reason
you'll
actually
see
it
show
up
here
so
Roberto
says,
would
ignite
run
on
a
Raspberry
Pi
for
I,
don't
know
what
the
Raspberry
Pi
4
actually
has
virtualization
instructions.
I
don't
know
if
Lucas
does
love
himself
some
arm,
but
he
says
he
has
plans
all
right
so
we're
making
progress
here.
A
A
All
right,
so
a
lot
of
the
intuition
that
you
bring
from
from
docker
would
actually
come
here.
Well,
okay!
So,
let's
just
let's
we'll
take
a
look
at
this
architecture
diagram
here
of
what's
going
yeah,
this
must
have
been
the
one
that
I
was
looking
at
okay,
so
the
things
that
are
coming
in
is
that
you
have
the
image
for
the
VM.
A
You
have
the
image
of
the
VM.
You
have
the
image
for
the
kernel
and
then
we
have
the
container
image
because
that's
where
we're
actually
running
firecracker
inside
of
the
container
image,
and
so
so
the
first
step
here
is
that
you
import
an
image.
We
download
that
using
OCI
and
then,
as
part
of
where
you
create
the
the
VM,
it's
actually
creating
a
it's
essentially
cracking
open
that
docker
image
taking
out
the
out
all
those
files
and
then
writing
those
things
to
disk
and
behind
the
scenes.
A
What
it's
using
is
actually
this
system
inside
of
Linux
called
device
mapper
and
device.
Mapper
is
a
way
that
you
can
essentially
create
block
devices,
do
copy-on-write
not
at
the
file
level,
but
at
the
bottle
and
the
device
mapper
method
of
actually
doing
this
is
is
a
was
what
Red
Hat
used
for
the
longest
time
to
actually
back
docker
and
so
device
mapper
over
time
has
actually
gotten
a
lot
more
stable.
As
the
docker
usage
of
device,
mapper
has
really
pushed
it.
Super
super
super
hard.
A
It
only
knows
how
to
consume
a
block
device
from
the
sort
of
host
side
and,
and
so
that
means
that
you
either
need
to
create
a
big,
sparse
file
or
use
something
like
qko
and
I.
Don't
think
even
supports
qk2
to
be
able
to
interface
or
you
can
actually
sort
of
do
pass-through
to
a
direct
block
device
on
the
host
kernel.
A
And
so
that's
essentially
what
firecracker
is
doing
is
it's
doing
pass-through
to
the
device
in
the
kernel
with
with
device
mapper,
and
so,
if
we
actually
go
through
and
we're
on
the
host
again
here,
cat
gov
or
LS
def,
and
we
look
at
let's
see,
I
think
if
we
block-
or
here
we
have
DM
0
I,
don't
know
device
mapper.
Oh
no,
that's
fraction
controlling
device
mapper,
but
if
we
do
a
block,
you
know
I
think
some
of
these
probably
dab
mapper.
Okay,.
A
All
right
yeah,
so
there
we
actually
see
here's
the
base
and
here's
the
let's
see
so.
Okay,
so
Roberto's
gonna,
gonna
hook
up
with
Lukas
and
try
and
contribute
over
slack
and
YouTube.
Apparently
doesn't
like
you
to
talk
about
slack
I.
Don't
know,
I
had
to
approve
that
message.
Okay,
but
if
we
look
at
mapper,
what
we
actually
see
here
is
that
we
have
ignite
base.
A
So
this
is
actually
taking
the
that
that
image
decanting
it
creating
a
filesystem
to
canteen,
mounting
that
rocket,
using
device
mapper
to
create
a
block,
a
new
block
device
taking
that
block
device,
initializing
a
filesystem
on
it,
mounting
that
file
system
and
then
cracking
open
the
OCI
image
and
actually
writing
it
in
that
file
system.
And
then
the
next
thing
that
that
ignite
is
doing
is
actually
creating
a
copy
on
write
clone
of
that.
That's
what
we
have
here
and
then
it
does
some
certain
specialization
to
essentially
specialize
that
particular
image
for
a
VM.
A
So
this
would
be
things
in
the
SSH
case,
like
writing
a
unique
SSH
key
and
they're-
probably
doing
things
like
configuring,
some
networking
stuff
configuring,
the
hostname
stuff
like
that,
and
so
this
is
the
base
in
the
overlay.
If
I
were
to
launch
another
VM.
So
if
I
go
through
and
I
do
run,
we're
gonna
call
this
one,
my
VM
to
number
one
I'm
not
gonna,
have
to
download
anything.
This
should
be
blazingly
fast
and
number.
A
Two
will
actually
see
a
bass
plus
two
two
derived
things
from
it:
boom
I
just
launched
a
VM
there
now
I
have
two
VMs
actually
running.
Okay,
so
Thorsten
is
saying:
can
you
show
the
output
of
Olo
setup?
We
will
in
a
second,
but
now,
if
I
do
LS
device
mapper
now
we
actually
see
we
have
two
bases,
and
it's
interesting.
Why
is
the
base
I
would
expect?
I
would
expect
Lucas
that
there
would
be
a
graph
and
we
could
actually
be
able
to
share
the
base,
but
apparently
we
actually
create
two
bases
there.
A
A
Taking
I
don't
know
exactly
what's
going
on
here,
but
I
believe
we
can
see
the
layers
here.
Okay,
yeah
I
am
NOT
a
device
like
the
last
time,
I
mucked
with
this
stuff.
I
can't
remember
how
long
ago
that
was
okay.
So
there
we
see
that
they're
sharing
the
same
image
text
for
file
under
the
covers:
okay,
I,
get
to
what
hey
what's
happening
here
is
that
okay,
it's
even
more
complicated,
so
okay.
So
this
is
essentially
a
sparse
file
for
the
file
system
sitting
on
disk
under
Lib,
firecracker
images,
etc.
A
And
then
what
you
can
do
is
you
can
loop
back
mount
that
which
essentially
take
this
file
and
then
use
that
file
and
expose
it
into
the
kernel
as
a
block
device,
and
then
there's
overlays
going
on
on
top
of
this,
so
the
fundamentals
sort
of
like
so
we're
not
having
amok
with
partition
tables
on
our
route,
OS
we're
actually
going
through
and
working
within
a
sparse
file,
that's
actually
sitting
in
the
file
system.
So
that's
why
the
so!
A
This
is
loopback,
not
loopback
as
a
network
but
loopback
in
terms
of
disk
mounting
stuff,
and
then
Rory
says.
If
you're
using
desktop
Linux
and
do
it
ignite
run
the
file
manager
be
like
hey
new
file
system,
let
me
go
ahead
and
mount
that
so
there
there
you
go
okay,
so
that
is
okay
and
then
the
way
you
can
initialize
this
X
for
using
essentially
user
mode
toolkits
from
for
mostly
they
were
built
for
like
riparian
file
systems,
but
you
can
also
use
it
to
initialize
those
those
file
systems
Wow.
A
This
is
taking
me
back
to
when
we
did
early
Google
compute
engine
stuff
all
right.
So
let's
look
back
at
our
diagram.
So
now
we
actually
have
a
block
device
that
we
can
point
firecracker
out
now.
The
next
thing
we
need
to
do
is
there's
a
couple
of
things.
We
need
to
set
up
networking
because,
oh
and
I,
we
didn't
actually
do
this
like
if
I
do
SSH.
A
A
All
right
so
so
Lucas
is
saying:
yes,
oh,
you
can
do
a
DD
zero
into
an
image,
and
then
you
can
do
make
FS
against
that
image
and
all
that
so
yeah
okay,
so
idea
will
get
networking
set
up
at
some
point
and
figure
out
what's
going
on
there.
But
one
of
the
reasons
that
what
we
got
to
do
here
is
that
we're
running
firecracker
inside
of
a
container
I
believe
for
for
two
different
reasons.
So
what's
the
race
condition
around
networking
did
I
not.
A
Yeah
resolved
comp
is
actually
empty.
All
right
so
will
it
could
be
because
I'm
running
on
this,
like
ginormous
machine,
and
so
it's
actually
too
fast
and
there
might
be
a
race
condition
there.
So
that
might
be
something
that's
going
on.
If
you
can
help
me
figure
out
how
to
fix
that
in
a
second
Lucas,
we
can
go
ahead,
but
I'm
going
to
talk
about
sort
of
why
we're
running
in
that
container.
So
there's
two
reasons
why
running
in
a
container
the
first
one
is
that
firecracker
provides
isolation
at
the
VM
level.
A
It
doesn't
provide
isolation
at
the
sort
of
CPU
and
RAM
level.
So
if
firecracker,
let's
say
that
you
launched
a
VM
inside
a
firecracker
that
have
12
CPUs
in
it,
but
you
only
but
like
you
could
then
theoretically
say,
but
I
want
to
actually
sort
of
restrain
that
to
like
two
CPUs
on
my
host
machine.
It
would
be
the
container
that's
running
firecracker,
that
wood
down
constrain
that
actually
more
likely.
A
A
So
that's
the
other
thing
and
then
the
last
thing-
and
this
was
I
was
going
to
get
to-
is
that
we're
actually
using
the
the
container
runtime
to
configure
networking
and
so
docker
will
then
go
through
configure
networking,
configure
CNI
and
then,
when
ignite
launches
docker,
it
gives
it
a
little
bit
extra
permissions
so
that
it
can
do
set
up.
What's
called
a
I
believe
it's
a
tunnel
I
always
get
confused
between
tabs
and
tunnels,
and
so
the
idea
being.
Is
that
what,
when
you
launch
firecracker
firecracker
knows
how
to
speak
blocks
and
packets?
A
Okay,
but
generally
your
OS
knows
how
to
speak
files
and
streams
and
and
sockets
right.
So
you
have
to
like,
like
machines,
speak
blocks
and
packets
processes
speak,
speak
files
and
and
sockets
streams,
and
so
with
respect
to
the
block
device
for
storage,
we're
doing
things
with
all
the
device
mapper
remounting
this
stuff.
A
They,
you
know
ignite,
creates
a
tap
and
then
it's
actually
there's
forwarding
of
packets,
both
ways
that
happens
between
ignite
and
the
kernel
at
that
level.
So
we're
not
using
the
TCP
stack
of
the
underlying
kernel,
but
we
are
actually
using
the
networking
stack
to
both
inject
and
get
packets
out
of
it.
A
A
A
Inside
the
VM
I
do
an
apt
get
update,
we're
using
the
TCP
stack
running
inside
the
VM.
It's
producing
a
bunch
of
packets.
Those
packets
are
sent
across
the
divert
I/o
net
interface
to
a
firecracker.
Firecracker
then
is
configured
to
take
those
packets
and
send
those
directly
to
the
tap
device
that
ignites
set
up.
That
tap
device
then
goes
through
and
essentially
drops
those
things
on
the
Network.
A
Let's
see
so
I'm
reading
comments
from
Lucas
here,
so
we're
removing
the
IP
from
the
container
and
then
transferring
the
IP
into
the
VM,
and
so
it
turns
out
that
there's
probably
no
easy
way
to
run
some
sort
of
networking
inside
the
container
that
actually
exposes
so
ok.
So
this
is
a
question
for
you.
Lucas
firecracker
actually
exposes
metrics
over
an
HTTP
service.
How
do
you
all
get
access
to
that
HTTP
service?
Is
it
something
like
you
know,
there's
a
you
open
up
a
socket
and
then
reexpose
it
and
then
release
the
IP
address.
A
A
A
It's
not
working,
oh
that
worked.
Oh
the
I
thought
the
oh!
No,
yes,
you
did.
The
is
the
arguments,
the
thing
okay,
so
now
I
can
do
IP
a
derp.
So
this
thing
now,
if
we
look
at
the
the
ve
that
we
have
here,
you'll
see
that
this
actually
doesn't
have
an
IP
address
assigned
to
it,
and
so
we
essentially
the
container,
that's
running
firecracker,
doesn't
have
an
IP
address
assigned
to
it.
A
Okay,
Michael
Payne
says
funny
that
for
years,
team
VMs
said
that
containers
don't
contain
and
you
should
run
them
in
VMs
with
a
night
we
have
VMs
running
in
containers
or
any
not
in
yeah
exactly
so
we're
finding
new
way.
Three
mix
these
things.
It
turns
out
that
that
there's
this
other
project
that
I
didn't
mention,
which
is,
is
cube,
vert,
which,
when
I
say
it
fast
Qbert.
It
sounds
like
Qbert
like
that
game
from
the
80s.
You
know
we
have
the
guy
across
the
grid.
A
Cube
FERC
is
essentially
a
way
of
using
KVM
to
actually
run
containers
inside
of
kubernetes,
and
so
again
it's
it's.
It's
a
it's
a
more
similar
thing
where
we're
using
containers
to
run
VMs
versus
using
VMs
to
run
containers,
but
this
is
much
more
of
sort
of
a
you
know:
local,
like
sort
of
docker
experience
versus
cube
FERC,
which
is
very
kubernetes,
ish
and
I,
know
that
there's
plans
to
start
using
ignite
inside
of
kubernetes
and
doing
interesting
stuff.
A
Okay,
what's
the
PSA
you
a
PSA,
you
X
W
inside
the
VM.
Here
we
can
see
what's
happening
here
all
right
and
then
the
other
interesting
thing
here
is
that
when
we
run
firecracker,
firecracker
actually
uses
a
UNIX
domain.
Socket
that
can
be
exposed
outside
of
the
container
mapped
in
to
be
able
to
get
access
to
things
like
Prometheus
metrics
out
of
the
VM
out
of
that
so
yeah.
A
A
Okay,
I
guess
we
don't
having
so
one
of
the
things
that
you
probably
want
to
do.
Lucas
is
actually
do
some
sort
of
spoofing
protection
where
you
make
sure
that
the
VM
doesn't
try
and
actually
use
any
IP
addresses
beyond
the
one
that
it
was
actually
assigned
to.
That
might
be
something
that's
worthwhile.
A
So
if
the
VM
tries
to
send
anything
that
is
yeah,
you
know,
and
then
you
probably
have
a
flag
to
disable
that
so
you
can
start
using
these
things
as
routers
and
stuff.
Oh
yeah,
you
got
to
get
paranoid
with
these
things,
alright,
so,
okay,
so
that
is
cool.
So
then
we're
actually
running
the
VM
and
then
the
VM
is,
is
you
know,
packaged
speaking
packets
and
blocks
so
across
here
we're
talking
packets
and
blocks
and
then
using
tap
and
the
and
the
device
mapper
we're
actually
converting.
A
You
know
the
the
block
into
file,
reads
and
writes
through
loop
backs
and
device
mapper,
and
then
we're
converting
and
then
we're
converting
injecting
packets
directly
into
the
into
the
kernel.
So
that's
the
basic
architecture
of
what's
actually
happening
here
and
and
I
mean
the
in
the
cool
thing
about
it.
I
think
like
like
stuff
that
I
think
is
really
really
neat
number
one.
Is
that
doing
the
the
tap
stuff
inside
of
a
container
and
reusing
as
much
as
docker
as
possible?
A
You
know,
I,
don't
know
if
MAC
address
spoofing
actually
matters,
because
it's
joining
a
bridge
and
I,
don't
know
if
there's
any
real
art
going
on
there.
That
bridge
is
not
bridged
to
everything
else.
Maybe
maybe
it
matters
or
not.
I
don't
know,
I
might
be,
might
be
out
of
control
all
right
and
then
Roberto's
saying
do
you
know
what
the
earliest
version
of
cube
admin
is
that
can
be
used
with
ignite.
A
The
downside,
then,
is
that
you
don't
get
things
like
broadcast,
because
your
broadcast
domain
is
like
you
and
your
router,
and-
and
so
that's,
why,
like
broadcast
multicast
is
generally
not
supported
in
cloud
environments
because
they
try
and
hop
to
l-3
as
fast
as
possible.
Okay,
so
now
we
got
some
VMs.
We
got
these
things
up
and
running.
Okay,
some
questions
for
you
Lucas,
and
these
are
things
that
I
don't
know.
Is
it
possible
for
me
to
go
through
shutdown,
a
vm,
take
that
image
and
actually
recreate
an
image
and
upload?
It
again.
A
For
building
container
images-
and
there
might
be
something
in
here-
that
you
can
steal
for
essentially
being
able
to
do
a
dip
of
two
file
systems
and
create
a
container
image
from
that.
So
that
might
be
a
way
to
essentially
say:
hey.
I
want
to
launch
a
VM
I
want
to
do
a
bunch
of
stuff.
Now
I
want
to
save
that
as
a
new
image,
and
actually
you
uploaded.
That
might
be
something
that's
worthwhile.
A
And
because
device
mapper
actually
does
do
copy-on-write,
but
it
does
copy
and
write
at
the
block
level,
not
at
the
file
level,
and
so
that
means
that
then,
to
go
ahead
and
create
container
images
from
it.
You
have
to
go
through
and
read,
arrive
the
file
changes
from
sort
of
like
doing
file,
comparisons
versus
being
able
to
actually
have
the
underlying
sort
of
like
overlay
system,
be
able
to
keep
track
of
it
for
you
all
right.
A
So
the
next
thing
we're
gonna
do
is
we're
gonna
launch
a
kubernetes
cluster
with
a
bunch
of
ignite
nodes.
That's
going
to
be
a
lot
of
fun,
so
the
first
thing
I
need
to
do
is
actually,
let's
sync
ignite
clone,
so
we're
going
to
go
through
and
hopefully
we
won't
have
to
edit
files
because
I
don't
have
like
I'm
such
a
dummy
these
days.
Okay,
so,
let's
see
do
I
have
to
do
this
as
a
root.
A
A
Okay,
so
we
got
that
up
and
we
got
that
now
we're
going
into
CD,
ignite
images,
cubed
and
then
okay,
so
now
we
go
through
okay,
so
set
up
some
files
of
certificate
using
prepare
alright.
So
what
are
we
doing
here?
So
we
got
hah
alright.
So
this
is
actually
some
fun
stuff
here.
Oh
so
one
of
the
fascinating
things
and
I
just
wanted
to
call
this
out
most
vm
systems
that
we
talked
about,
whether
it
dec
2
g,
c
e,
I,
believe
Azure.
Does
this
I
believe
it's
available
and
things
like
vSphere?
A
You
take
a
VM
image.
You
long,
you
boot
the
VM,
but
you
want
to
do
some
sort
of
specialization
in
the
VM.
You
want
to
inject
some
information
to
tell
the
VM
how
to
actually
specialize
itself,
and
this
is
generally,
you
know,
there's
there's
a
ton
of
different
ways
to
do
it,
but
you
might
use
something
like
cloud
in
it,
and
so
this
means
that
inside
the
VM
you
get
access
to
something
like
a
metadata
server.
Where
you
can
say
hey
what
SSH
keys
should
I
actually
use.
Or
you
know
what
are
there?
A
Some
programs
that
I
should
run
on
startup
ignite
doesn't
have
a
metadata
server
and,
and
it
turns
out
that
I'll
removed,
ok,
I'll
remove
the
previous
VMs,
so
it
turns
out
that
implementing
a
metadata
server
and
this
thing
is
actually
really
complicated.
Because
again,
the
VM
on
the
host
side
actually
speeds
packets,
but
you
want
to
speak
HTTP
to
implement
a
metadata
server
and
to
do
that.
A
A
So
that's
actually
really
really
interesting.
Ok,
fire
cracker
does
have.
Oh,
it
does
have
a
metadata
server.
I'm,
sorry
come
in
and
vb0
it
out.
Five!
Okay!
Well,
that's
fascinating!
Look
at
that!
Oh
you
know
what
I
remember:
we've
looked
at
this
and
then
it's
like
I'm
like
what
are
we
using
as
a
TCP
stack
because
and
it's
the
mmds
Network
stack.
Yes,
okay,
yeah,
because
this
stuff
starts
getting
really
complicated.
A
Okay,
dump,
okay,
Oh
Dumbo
is
that
did
the
network
stack,
that's
used
yeah.
This
didn't
exist
back
in
the
day
when
I
was
doing
this
stuff,
okay,
so
okay,
so
yeah,
so
that's
gonna,
be
an
interesting
thing
to
actually
look
into
in
the
future.
Is
what
kind
of
metadata
do
you
want
to
expose
through
the
net
through
the
metadata
service
into
the
your
egnyte
VM
already
says,
the
TC
PHP
is
something
that
Alex
wrote
in
one
weekend.
We
keep
testing
and
it
works,
but
I'm
very
suspicious
of
it
yeah
we
ended
up
using,
oh
god.
A
A
They
were
great
up
until
you
start
really
pounding
them
hard
and
trying
to
scale
up
it's
one
of
those
things
where
you
don't
need
this
stuff
to
be
super
fast,
because
you're
never
reading
a
ton
of
data
over
the
metadata
server,
but
you
want
it
to
actually
be
solid
and
reliable.
Okay,
ok,
very
cool!
Ok!
So
now
let's
go
through
and
ok.
So
the
first
thing
we're
doing
here
is
we're
doing
ignite
PS
ignite
RM
f
of
this
one.
A
Did
I
get
him
okay,
we
got
them
both.
Okay
Russ
just
has
awesome
web
server
stacks
that
were
too
much
for
us.
Okay.
Well,
because
you
did
a
new
HDTV
stack
as
part
of
it
yeah
it's
a
whole
nother
threat
analysis
when
you're,
actually
talking
about
exposing
a
raw,
TCP,
HDB
and
TCP
stack
to
the
to
the
guest
like
that.
Okay,
so
we
go
through
it.
Let's
do
so
the
prepare
here.
What
are
we
doing
is
that
we're
doing
a
docker
docker
run
with
cube
admin
to
create
some
certificates,
create
an
admin
cube
config.
A
A
You
know
we
could
download,
cube
admin
and
run
it,
but
if
you
have
a
docker
container,
you
might
as
well
do
that
then
you'll
be
reliable
and
we're
using
the
same
mechanisms
for
distributing
all
the
binaries
okay.
So
now
we
actually
go
through
and
if
we
look
into
the
run
directory,
we
have
our
cat
run.
So
this
is
a
cube
config
for
the
cluster
that
we're
about
to
launch
there's
config
dot,
yeah
mole,
which
essentially
is
the
certificates
and
everything
that
we're
going
to
be
doing.
A
We
have
the
tokens
and
certificate
keys,
and
so
we
feed
that
stuff
into
the
VM
to
avoid
having
to
copy
stuff
later.
If
you
can
create
this
stuff,
then
you
don't
have
to
sort
of
launch
stuff
and
sort
of
like
read
stuff
from
here
and
push
it
over
there.
You
can
actually
create
it
once
and
then
push
it
out
to
everybody,
it's
much
easier
to
coordinate
around
its
just
added.
If
you
want
to
access
it
on
the
hosts.
Ip.
Oh
I
see
okay.
A
A
We
have
the
certain
key,
the
CA
hash
and
the
token,
and
then
because
those
are
things
that
we're
gonna
actually
be
dealing
with
run
PKI,
and
then
we
have
our
certificate
and
key
for
that
for
the
serta
30,
okay,
very
good,
very
cool
all
right.
So
now
now
we're
gonna
go
through
and
we're
going
to
do,
an
ignite
run
to
actually
launch
the
seed
master
and
so
we're
starting.
So
this
is
going
to
be
our
master
node.
A
A
All
right,
all
right
so
so
to
Bob's
I
mean
this
stuff
is
still
early
I
just
launched
number
one.
Is
that
because
I'm
running
on
such
a
freaking
beefy
machine,
we
probably
tickle
the
race
condition
where
the
name
server
the
result
comp
doesn't
get
set
up
and
and
then
the
second
thing
is
that
for
this
particular
demo,
Lucas
switched
from
using
docker
to
container
D
docker
sets
of
IP
forwarding
container
d-doesn't,
and
so
we
had
to
manually
go
through
and
do
that.
A
Alexander
is
saying
how
far
is
ignite
from
being
a
CRE
compatible
solution
say?
Would
it
replace
doctrine
containers?
So
this
is
the
thing
is
that
like,
if
you
want
to
see
RI
compatible
solution,
then
use
something
like
a
firecracker
container,
D
shim.
This
is
really
not
about
running
containers
and
VMs.
It's
actually
using
firecracker
to
actually
run
lightweight
VMS,
and
so
the
whole
idea
is
that
we
want
to
create
a
super
sort
of
docker
ish
experience
around
creating
VMs.
A
Is
there
any
way
easy
way
of
running
VMs
from
kubernetes
pots
yeah?
So
this
would
be
the
way
of
running
VMs
from
kubernetes
pods,
not
the
way
of
actually
running
pods
in
VMs.
It's
very
confusing.
It
really
is
so
the
competition
to
this
would
be
something
like
cube
vert,
whereas
the
other
thing
that
you
were
talking
about
the
the
solution,
there
would
be
something
like
kata
containers
or
in
a
hosted
environment,
something
like
a
CI
or
or
Fargate.
A
A
A
So
those
things
are
up
and
running
okay
and
then
ssh
into
each
one
and
we're
gonna
do
ignite
as
a
sage
master
one
one
of
the
interesting
things
that
I
think
ignite
does.
Is
it
actually
creates
a
SSH
key
pair
for
each
VM
that
you
launch
so
you're,
not
actually
reusing
the
same
one?
So
by
default
you
can
actually
say
use
this
other
different
thing,
but
there
we
go.
Okay,
we
want
to
do
echo
name
server.
One
got
one
got
one
to
resolve
and
we
want
to
do
echo
1.
A
A
A
So,
instead
of
actually
setting
up
any
sort
of
DNS
you're
like
let's
just
do
it
with
Public
DNS,
all
right
little
hacker
ama
there
all
right,
we
could
have
actually
mapped
that
in
to
resolve
comp.
I.
Think
it's
also
and
that's
why
you
want
me
to
delete
the
other
VMs
to
be
able
to
do
that
yeah.
So
I
think
this
is
the
type
of
thing
where
it's
like,
okay,
now
you're,
actually
starting
to
like
what
is
the
difference
between
launching
some
VMs
and
creating
a
cloud.
A
A
Is
that
when
you
launch
a
VM
and
GCE,
you
give
it
a
name
that
automatically
gets
added
to
DNS
by
default
for
GCE,
whereas
if
you
want
to
actually
use
names
for
for
DNS
with
something
like
ec2,
you
have
to
do
route
53
and
then
populate
that
and
so,
and
so
doing,
some
sort
of,
like
auto
name
to
DNS
type
of
thing
with
with
ignite,
might
actually
be
pretty
cool
all
right.
So
we
got
one
of
those
things
now.
A
A
A
A
A
A
A
A
A
Ready
boom
we're
up
and
running
okay,
so
now
we
got
that
up
and
running
and
I
assume
that
we
can
go
through
and
actually
add,
more
nodes
and
stuff,
like
that,
also
which
would
be
super
super
cool,
but
we're
running
out
of
time.
Okay,
the
last
thing
on
the
guide
here
is:
we
can
go
through
and
actually
say,
go
ahead
and
delete
Master
zero,
and
now
we
can
do
get
nodes.
A
A
Yeah,
it
takes
time
okay,
but
at
some
point
it
takes
some
minutes.
Okay,
yeah.
You
can
tune
this
stuff
with
kubernetes,
but
at
some
point
we'll
realize
that
that
nodes
unhealthy
and
automatically
remove
stuff.
Most
of
the
time
you
know
you
don't
just
sort
of
yank
the
plug
on
nodes
like
that
very
cool
all
right.
This
is
super
super
cool,
I
love.
You
know
just
to
wrap
things
up.
It's
still,
you
know,
there's
there's
a
few
hiccups
we
hit
here,
but
I.
A
Think
in
general,
the
couple
of
things
that
I
think
are
really
really
interesting.
Number
one
reusing
the
docker
image
format
for
distributing
things
that
go
above
and
beyond
just
docker
images
is
actually
really
cool.
This
is
something
I've
been
harping
on
for
a
while
is
that
taking
a
set
of
files,
packaging,
those
things
together,
uploading
them
signing
them,
naming
them
keeping
track
of
them?
That's
a
useful
service
in
and
of
itself,
even
ignoring
all
the
docker
stuff
and
so
I'm
glad
to
see
people
start
recognizing
that
that
facility
is
useful
outside
of
containers
and
I.
A
Think,
as
we
see
that
develop
over
time,
it'll
actually
be
more
and
more
interesting.
Like
imagine
that,
like,
if
you're
running
in
kubernetes,
we
could
have
a
volume
which
is
of
type,
you
know,
OCI
bundle
or
something
like
that
and
where
we
create
a
file
system,
and
then
we
initialize
the
file
system
with
the
OCI
bundle
using
naming
and
stuff
like
that,
where
you
get
all
the
benefits
of
container
naming
and
signing,
but
you
also
get
the
fact
that
you're
just
using
it
for
plain
old
data
and
not
for
you
know
the
full
container
image.
A
Resource
isolation
when
running
the
firecracker
binary,
but
then
also
using
docker
to
set
up
networking
so
that
you
can
actually
set
the
tun
in
the
tap
I
mean
the
tap
through.
That
is
actually
really
cool
too.
So
that's,
actually
pretty
cool,
I.
Think
yeah,
there's!
You
know
missing
features.
Obviously
these
things
are
still
very
ephemeral
like
containers.
A
If
you
know
you
know,
like
saving
data
over
time,
shutting
one
down
cloning,
it
creating
data
volumes.
Those
things
all
look
to
be
missing
right
now,
and
so
maybe
maybe
that's
something.
That's
missing
that
can
be
headed
over
time.
Yeah
Lucas
is
making
excuses
here.
This
thing
started
in
mid-may,
so
it's
like
it's
great
I
mean
I.
I
love
the
ideas
behind
it.
You
know,
there's
gonna
be
things
that
are
gonna,
be
be
added
over
time.
A
That's
a
lot
harder
to
do
with
a
VM
and
so
I
know.
At
one
point,
there
was
talk
of
supporting
Verdi
op9,
which
is
essentially
kind
of
like
a
fuse
like
protocol
that
works
across
the
hypervisor
interface
I.
Don't
know
if
they're
still
still
talk
of
doing
that.
Oh
the
get-ups
thing.
So
let's
get
the
get-ups
thing
real,
quick
I.
Maybe
we
can
demo.
Maybe
we
can't
so
I
do
says
a
really
interesting
to
take
feature
requests
around
this
stuff.
A
Okay
and
get
you
declaratively
store.
Okay,
so
you
can
do
ignite,
get
ops
where
repo
points
to
a
git
repo,
and
then
you
create
a
file
saying
here's
the
VM
that
I
want
to
run
with
the
CPU,
Ram
disk,
etc.
And
then
you
can
do
a
git
push
and
actually
make
it
work:
okay,
yeah
that
seems
pretty
cool,
and
so
it
sits
there
and
it
launches
VMs
based
on
a
directory
versus
versus
anything
else.
So
so,
if
I
do,
let's
see
make
dur
my
get
CD
my
gigs
get
in
it
and
if
I
can
do.
A
A
A
A
A
A
A
A
A
Did
I
miss
a
quote?
Let's
actually
see
what
happens
today,
mrs.
cool
yeah
I
got
the
quote
on
their
narrow
curtain
when
pressing
the
V
M
one
two
three
cannot
change
into
the
created
state
describes
a
VM
with
UID
one
two,
three
four
name
test:
VM,
oh
okay,
well,
I,
think
we're
running
out
of
time.
Lucas
I
gotta
run
because
my
life
is
all
means
now
apparently
remove
the
file.
A
Because
remove
and
then
we
have
to
wait
for
it
to
cream,
then
you
have
to
recreate
it.
Okay,
well,
okay,
so
piece
of
feedback
there
I
love
the
idea
of
get
apps
I'd
love
to
declare
declarative
nature
of
it.
I
would
actually
layer.
This
in
I
would
actually
say
at
the
end
of
the
day
having
something
like
the
static
manifest
capability
of
cubelet
is
super.
A
A
A
I
honestly,
wouldn't
tie
these
two
closely
together
or
at
least
making
an
option
to
actually
pull
stuff
off
of
a
local
Drive
instead
of
instead
of
always
pulling
from
get,
you
know,
cuz,
who
knows,
we
may
actually
be
in
a
world
where
there's
more
than
get
at
the
at
the
end,
so
yeah
awesome,
stuff,
Lucas,
really
really
exciting.
Good
job
I
think
everybody's
really
really
excited
about
it.
A
I'm
definitely
going
to
be
keeping
an
eye
out
to
see
see
where
you
all
take
it
next,
so
very,
very
cool
and
Alexander
I
will
not
forget
to
shut
down
my
very
pricey
p.m.
that
is
way
overkill
for
the
demo
that
we
did
here
and
okay,
oh
and
then
also
use
kubernetes.
As
the
back
of
my
series.
Exactly
that
be
super
quote.
Thank
you,
everybody
for
joining
me.
We're
like
an
hour
and
40
minutes
in
for
those
of
you
who
stayed
with
me.
You
know
good
job
I
learned
a
lot.
A
This
was
a
lot
of
fun
and,
let's
see
tune
in
I,
think
I
might
be
doing.
I
have
to
look,
the
calendar,
I
might
be
doing.
Gg
I
can
next
week,
but
then
I'm
gonna
be
on
vacation
for
several
weeks
and
so
we're
gonna
have
to
fee
and
Josh,
and
and
maybe
we
can
a
Christian
and
and
I'm
working
on
getting
olive
to
to
get
set
up
so
that
she
can
actually
do
a
TGI
K,
so
we're
trying
to
expand
the
set
of
folks
doing
this
stuff.