►
From YouTube: Scalable Multiplayer Game Design with OpenShift (E8)
Description
This monthly series explores what it takes to design a scale-out multiplayer video game architecture using containers and OpenShift. There will be live coding, philosophical design discussion, and everything in between.
In this episode we'll try to get the container build of the game server running on OpenShift. We'll also deploy the web client. And, if time allows, we'll try to apply some GitOps principles to automating all of it.
Host:
Erik Jacobs
Guests:
* Derek Reese
* Michael Clayton
* Roderick Kieley
* Jared Sprague
A
A
Yes,
jerry,
give
me
a
thumbs
up.
Okay,
so
roddy
and
jafar
had
done
a
little
bit
of
pre-work
behind
the
scenes
to
figure
out
how
to
get
the
game
server,
built
and
deployed
using
openshift
pipelines
and
in
this
environment.
Actually,
don't
even
have
pipelines
installed
so
jafar.
If
I
remember
correctly,
I
need
to
go
to
operator
hub
and
just
install
the
pipelines
operator.
B
A
And
so
operator
hub
for
those
who
may
not
have
seen
it
before.
This
is
kind
of
our
convenient
user
interface
into
a
catalog
and
marketplace
of
certified
and
community
operators.
And,
if
you
don't
remember,
an
operator
is
just
a
piece
of
software
that
is
a
kubernetes
control
loop,
that
sort
of
makes
sure
that
something
is
running
in
a
kubernetes
environment
as
you
define
it
in
the
case
of
openshift
pipelines.
A
This
is
a
operator
that
provides
the
openshift
pipelines
feature
which
is
based
on
the
upstream
tecton
community
project,
and
so
installing
this
operator
is
going
to
add
the
tecton
features
into
this
openshift
environment.
So
I
will
go
ahead
and
click
the
install
button
here.
I
guess
I
want
stable.
We
want
pipelines
to
be
available
everywhere.
A
Approval
strategy
has
to
do
with
when
or
how
the
operator
gets
updated
if
there
are
updates
available
and
here's
all
the
apis
that
are
provided
by
it.
Unsurprisingly,
the
pipelines
operator
provides
a
pipeline
object
or
api
endpoint,
so
we
click,
install
and
what's
happening
under
the
covers
is
operator.
Lifecycle
manager
is
potentially
creating
a
namespace
and
it
is
deploying
the
operator
into
that
namespace
and
adding
some
artifacts
into
that
space.
And
so,
let's
see
oc,
get
project
grep
pipe,
I'm
using
a
pipe
to
pipe
into
grep.
To
look
for
a
pipe.
A
A
Oh,
I
was
asking
about
the
project
where
pipelines
lands,
but
I
guess
I
was
just
a
little
too
early.
Let's
see
you
get
pod
in
the
pipelines
project
we
see
there's
an
operator
and
then
there's
the
pipelines,
controller
and
the
pipeline's
web
hook,
which
I
guess
is
an
admission
controller
or
something
like
that.
And
now,
if
we
hard
refresh
the
openshift
user
interface,
we
actually
pick
up
this
new
pipelines,
tab
which
lets
us
look
at
the
pipeline
objects
in
our
various
projects
and
other
stuff.
A
D
B
B
D
B
D
D
B
C
A
D
B
Because
the
when
you
are
going
to
do
the
git
clone
so
basically,
when
you
generate
the
pipeline
you're
going
to
have
three
tasks:
you're
going
to
have
the
git
fetch
source
code-
okay,
okay,
and
it's
going
to
clone
the
code
in
the
working
there
like
it's
going
to
be
in
the
cloud
in
the
code
where
it
clones
your
your
your
source,
then
if
it's
going
to
do
the
docker
build
from
that
folder.
So
if
you
copy
that
the
the
dot
folder
to
slash
tmp
game
server,
it's
gonna
copy.
The
whole
git
ring.
B
A
A
B
B
B
A
A
B
B
B
A
A
A
A
Yeah,
I
wanted
to
see
the
actual
pod
where
the
the
list
of
the
pods,
where
the
things
are
happening.
So
what
we
see
here
is
the
game.
Server
is
trying
to
be
deployed
and
it's
failing,
because
we
don't
actually
have
an
image
yet
so
the
reason
it's
pulling
the
reason
it's
backing
off
from
trying
to
pull
the
image
is
because
it's
failing
to
pull
the
image
because
there
isn't
an
image
because
we're
in
the
process
of
building
the
image.
A
B
D
A
Here's
all
the
crazy
volumes
that
get
attached,
wow,
there's
a
lot
of
volumes
holy
crap
so
anyway
back
to
our
pipeline,
so
here's
the
build
which
hasn't
exploded
yet
so
that's
a
good
sign.
So
it's
using
fedora
33,
it's
installing
a
bunch
of
packages,
blah
blah
blah
that
takes
forever
looks
like
the
package.
Build
stuff
was
successful,
sorry
package
install
which
is
good.
F
F
A
E
A
B
F
A
A
F
B
C
C
A
A
A
Okay,
so
I'm
going
to
add
some
more
nodes
to
my
cluster
so
that,
hopefully
I
have
space
to
run
this
workload
because
I'm
sure
elasticsearch
is
being
hog
and
eating
all
my
ram.
I
guess
I
could
kill
logging,
but
that's
less
fun
jared.
Are
you?
Have
you
made
any
progress
or
do
we
want
to
just
do
it?
Live.
G
D
A
A
C
D
A
A
A
C
Yes,
yeah,
so
basically,
there's
broker
deployments,
which
is
the
amq
broker.
Crd
there
are
broker
addresses
so
that
people
can
define
their
cues
and
topics
and
amq
addresses
mqp
addresses
that
they
want
to
deploy
across
their
broker
cluster,
and
there
is
a
scaled-down
controller.
So
the
architecture
of
the
operator
is
such
that
there's,
actually
three
independent
controllers
inside
the
operator,
one
for
each
crd
makes.
A
C
The
scale
down
controller,
in
particular,
watches
just
for
pod
discrepancies
to
see
whether
or
not
the
size
you
have
specified
matches
the
number
of
pods
and
when
there's
a
discrepancy
it
starts
up
and
if
configured
to
do
so,
it'll
go
get
your
messages
from
the
pod
that
just
scaled
down
and
moved
those
messages
to
somewhere.
That's
still
available.
C
E
C
C
C
A
A
C
A
C
A
C
D
C
D
C
A
D
F
E
A
B
B
A
A
A
B
A
C
B
A
C
A
F
A
A
C
C
A
A
A
F
A
A
A
A
C
A
C
C
C
A
A
B
E
A
A
D
C
A
A
No,
no
hungry,
oh
damn,
all
right!
Well,
he's
busy
doing
stuff
all
right.
So
let's
recap
what
we've
done
so
far.
So
we
have
deployed
the
amq
operator
into
our
project
and
found
a
blog
all
right.
So
we've
got
the
amq
broker.
We've
got
our
sorry.
The
operator
we
deployed
a
broker.
We've
got
a
pipeline,
that's
in
the
process
of
doing
its
thing.
A
We
modified
the
route
for
the
broker
to
use
a
tls
edge
termination
with
the
self
sign
certificate
from
our
router,
and
then
we
locally
tested
the
game
client
talking
to
the
wss,
the
secure
websocket
for
the
broker.
So
that's
good,
so
we
validated
that
the
broker
is
doing
what
we
want.
We
fixed
our
limit
range
that
was
may
have
been
causing
us
problems
with
our
build
and
oom.
A
F
A
C
B
A
So
yeah,
okay,
perfect,
so
I
had
created
this
demo
app
name
space
earlier,
not
today,
some
other
day.
Oh
look!
I
can
tell
you
exactly
when
it
was
created
on
the
on
the
7th
which
is
not
today,
and
we
got
this
core
resource
limits
thing.
There's
some
there's
some
like
limit
range
controller,
web
hook
thing.
A
A
A
So
openshift
has
a
concept
of
a
project
request.
So
if
you
use
the
default
workflow
of
clicking
new
project
in
the
user
interface,
what
actually
is
happening
under
the
covers
is
it's
doing
this
project
request
workflow,
which
says
hey?
Is
this
user
allowed
to
create
a
project?
Oh,
they
are
great
use.
This
template
object
and
instantiate
a
bunch
of
stuff,
and
what
you
see
here
is
there's
a
network.
There's
a
default
network
policy.
A
There
is
the
actual
project,
object,
there's
a
bunch
of
role,
bindings
and
service
accounts
and
other
stuff,
and
so
what
the
demo
people
did
was
they
added
this
limit
range
to
that
template,
and
so
you
can
see
that
the
pod
memory
max
is
12
gigs
and
the
default
request
for
a
container
sorry,
the
default
limit
for
a
container
is
only
1.5
gigs,
which
is
not
very
much
right.
So
that
would
explain
why
we
were
getting
blown
up
and
oom.
A
A
F
B
D
D
F
A
Man
holy
crap,
all
right,
so
here's
the
service,
the
cool
thing
about
openshift
and
and
services
and
internal
and
all
that
fun
stuff-
is
that
you
don't
need
fully
qualified
domain
names
because
we
set
the
end
dots
on
the
on
the
resolver.
So
we
can
just
go
broker
your
is
it
broker
uri
broker
uri?
We
can
set
that
to
simply
that,
oh
variety.
Do
I
need
the
tcp
or
do
we
do
that?
C
A
D
B
A
It's
moving,
oh
god.
I
think
we,
oh
the
default.
The
default
cycle
sleep
cycle
is
really
bizarre
anyway,
yeah,
so
so,
whatever
spaceship
moves,
we
did
it.
So
what
did
we
do?
We?
We
only
did
we
did
it
in
an
hour.
That's
like
a
new
record
for
achieving
stuff
on
this
stream.
So
what
did
we
do?
I
should.
G
G
F
B
A
A
When
I
run
phaser
games,
especially
it
just
like
heats
up
the
cpu
really
bad
and
then
it
basically
goes
to
like
thermal
throttle
and
so
the
processor
speed
will
get
throttled
down
to
like
200
megahertz
and
then
the
load
average
shoots
through
the
roof.
And
then
it's
basically
just
like
whatever
the
linux
scheduler
decides
is
important
at
that
moment,
which
usually
isn't
zoom
so
yeah
it
takes,
takes
a
few
moments
for
for
it
to
all
come
back,
but
I
think
I'm
back.
A
Okay,
so
in
a
new
record
for
this
stream,
in
about
an
hour,
we
actually
got
cool
stuff
working
and
so
to
recap
again
we
have
a
pipeline
that
builds
a
c
application
from
a
docker
file
that
speaks
amqp
over
tcp
to
the
broker
and
we
deployed
the
broker
using
the
amqp
operator.
So
that's
kind
of
cool,
it's
pretty
vanilla,
simple,
config
right,
there's,
no
redundancy,
there's
no
persistence
of
the
message
queue,
but
it
does
work
and
so
now
potentially
jared
we
could
deploy
the
web
client
for
the
game.
A
A
G
G
A
D
G
So
so
we've
got
a
gulp
file
yeah,
so
I
added
a
gulp
file
added,
which
basically
adds
like
you
know
it.
It's
it's
build
tasks
basically
like
copying
stuff
to
the
dist
directory
running
running
babel,
which
running
babel
and
browser
fi,
which
takes
all
of
the
all
of
the
like
dependencies
that
are
required
in
the
javascript
and
builds
them
all
like
package
them
all
into
a
single
javascript
file
that
gets
included
at
the
got
it
at
the
top
level.
Does
a
bunch
of
magic.
G
G
So
the
top
level
we
have
like
assets,
yes,
sjs
and
let's
see
what
else
did
I
have
to
do.
I
had
in,
and
I
had
to
go
into
each
file
pretty
much
because
before
we
were
including
everything
globally
by
just
using
the
script
tag
on
the
index.html
file
and
just
every
dependency,
we
would
just
import
it
in
the
global
window
scope.
So
all
the
javascript
files
had
a
scope
into
everything.
G
So,
since
we're
not
doing
that
anymore,
I
have
to
require
those
dependencies
directly
in
the
files
themselves.
So
if
you
look
at
the
one
of
the
files
like
look
at
like
the
amqp,
client
or
or
the
main
scene
or
something
yeah
there,
we
go
yeah.
So
you
see
where,
if
I
got
protobuf
equals
require
protobuf
and
uuid
four,
you
could
require
uuid
those
those
didn't
used
to
be
there,
because
you
could
see
where
they
just
used
to
be
in
the
global
scope,
because
they
were
scripted
in
the
html
file.
G
G
G
It
is
yeah
there
you
go
so
we
got,
we
got
build,
dev,
build
prod
start
by
default,
runs,
build
dev
when
you're,
just
running
it
locally,
and
the
only
difference
between
build
dev
and
build
prod
is
build.
Prod
will
include
minified
versions
of
the
minified
version
of
phaser,
which
is
like
like
the
unminified
version
of
phaser,
is
huge,
is
giant
is
so
yeah,
so
it
you
know,
makes
it
makes
it
smaller.
It
also
adds
like
a
global,
like
environment
variable
for
prod.
G
So
if
you
want
to
switch
on
stuff
in
the
code,
if
it's
prod
you
can
you
can
reference
that,
but
other
than
that,
oh
and
then
the
other
thing
is
the
prod.
Build,
will
include
source
maps.
So
when
you,
when
you're
needing
to
debug
something
in
prod,
even
though
like
without
a
source
map,
you
wouldn't
be
able
to
debug
it
because
everything's
just
like
c
c
dot,
one
two,
you
know
it's
like
you
can't
everything's
obfuscated
right,
but
the
source
map
lets
you
lets
you
debug
stuff,
even
if
it's
minified
it's
kind
of
weird.
A
E
A
B
G
And
then
it's
installed
yeah.
If
you
do
an
npm
install
like,
and
I
think
the
s2i
will
do,
an
npm
install,
it
will
be
there.
The
question
is,
the
question
is
is
like
when
it
when,
if,
like
the
difference
between
the
building
and
the
running
like
it's
going
to
build
it's
going
to
output
the
stuff
to
a
disk
directory
which
is
static,
html
and
javascript
files
that
doesn't
need
to
run
under
node.js,
it
just
needs
to
run
under
apache
or
something
so
we
could.
C
A
A
So
if
we
look
at
the
node
script
for
the
build
the
s2i
image,
so
for
those
who,
who
don't
understand
some
of
these
acronyms
so
source
to
image,
is
a
feature
that
open
shift
takes
advantage
of
to
build
container
images
directly
from
source
code.
So,
instead
of
having
to
use
the
docker
file,
you
can
basically
just
point
it
at
your
gate,
repo
and
so
right.
Now,
I'm
looking
at
red
hat
software
collections
provides
a
source
to
image
container
for
node.js
and
the
build
process
that
gets
invoked
is.
Is
this
a
symbol
script?
A
And
so
I'm
looking
in
the
assemble
script,
to
see
what
happens
so
we
move
the
source.
We
do
some
permission,
z,
stuff.
We
do
some
npm
configure
stuff
if
it's
dev
mode,
we
set
that
to
false
if
node
end
is
defined,
we
set
node
end.
If
it's
production,
we
do
an
npm
install.
Otherwise
we
do
node
and
dev
npm
install
and
then
npm
run
build
so
jared.
If
we
changed
our
npm
tasks
to
make
the
default
behavior
of
build.
Do
the
gulpy
stuff.
G
G
G
A
G
A
Target
so
right
and
we
do
and
and
look
if
I
remember
correctly,
the
oh
man,
I
lost
my
browser
window,
so
many
browser
windows,
yeah
it'll-
do
an
npm
install
prior
to
if
it's
then
otherwise,
actually
so
we're
gonna
act.
So
here's
the
here's
the
caveat
right.
If
the
node
environment
is
not
production,
it
will
do
an
npm
install.
A
G
A
A
B
G
B
B
A
B
B
G
A
A
A
A
A
D
A
B
B
A
Well
so
jafar,
once
the
build
is
theoretically
complete,
so
the
default
pipeline
is
not
going
to
be
particularly
useful
right,
but
we
couldn't.
We
take
the
output
of
the
node,
build
and
then
shove
that
into
the
pvc
for
the
pipeline
and
then
have
another
task
that
builds
an
httpd
image
and
just
copies
the
junk
from
the
from.
B
G
G
F
B
G
D
G
A
A
Yeah
so
it
looks
like
we
also,
it
says:
if
noted
is
not
set
by
the
user,
then
nodem
is
determined
by
whether
the
container
is
run
in
development
mode.
So,
if
note
end
is
set
otherwise
export,
but
this
is
the
dumbest
thing
I've
seen
so
you
have
not
only
do
you
have
to
set
node
end
view.
You
also
have
to
set
dev
mode
yeah,
because.
A
B
A
A
So
if
even
if
we
didn't
set
dev
mode,
this
is
like
the
weirdest
thing.
If
we
set
node
nv
equals
anything,
it
will
do
all
the
other
stuff,
so
we
actually
didn't
even
need
to
set
dev
mode.
Like
I
do,
I
don't
understand
whoever
wrote
this,
I
I
I'm
sorry,
I
don't
understand
what
you
were
thinking,
but
this
is
just
the
I
don't
understand
here,
set
dev
mode
to
false
by
default,
export,
dev
mode.
False
then,
if
note
end
exists,
if
dead
mode
was
true,
then
set
node
end
to
development.
B
E
B
B
B
B
A
D
A
A
B
G
G
B
You
can
use
also,
which
is
in
the
assemble.
You
have
pre,
assemble
and
post
assemble
stuff
that
you
can
do
also
like.
If
you
don't
want
to
to
to
copy
the
overall
assembled
file
there.
There
are
hooks
that
you
can
use
to
do
stuff
before
or
after,
and
then
just
call
the
assemble
script
wherever
you
want.
D
D
G
G
G
G
A
G
G
A
G
A
B
F
G
A
G
A
E
A
G
F
G
A
A
G
B
A
F
B
B
G
B
B
A
G
A
G
G
A
A
B
B
A
A
A
B
A
D
A
G
B
G
G
G
E
A
G
G
G
G
A
C
A
G
D
B
A
All
righty:
well,
I
guess
we'll
call
it
a
day,
then
stop
my
screen
share
here.
So
thanks
much
to
our
special
guest
jafar
for
helping
us
lose
our
sanity
with
node.js
and
pipelines
and
and
building
c,
plus
plus
applications
with
om
killer,
I
mean
with
openshift
but
yeah.
That's
it
for
this
month,
and
so
maybe
next
month
we
will
go
back
to
doing
more
sort
of
actual
game
dev
stuff.
A
What
I
was
hoping
to
get
to
today
was
to
actually
get
the
web
client
running
enough
so
that
we
could
share
it
to
have
a
bunch
of
people,
try
to
play
it
since
it'd
be
running
on
openshift.
At
that
point
we
could
all
interact
with
it,
but
we'll
have
to
save
that
for
next
month.
Yeah
we
do
a
load
test
of
like
I
will
all
of
our.