►
From YouTube: Quilkin Monthly Sync - March 2022
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
B
B
Yeah,
I
think,
actually
we've.
I
think
you
you
you
sold
me
on
that.
Anyway.
That's
all
fine!
There
was
more
than
not
a
roster,
but
like
we
talked.
B
But
I
was
just
actually
what
were
my
notes.
I
don't
want
to
say
if
I'm
recording,
but
I
spoke
to
a
few
metropolitan
friends
about
culkin
people
were
like
where's,
that
we
got
both
questions
about
like
where
is
it
at
and
I
was
like:
hey
we're
still
a
little
early
but
we'd
love
to
come
play
and
come
use
it
and
we're
using
it
and
stuff
too.
I
gave
out
a
bunch
of
stickers.
I
need
to
give
you
all
some
stickers.
I
don't
know
we
should.
We
need
to
organize
that.
B
I
need
to
get
april
on
that,
because
the
stickers
are
really
cute
but
yeah.
The
other
fun
part
about
that
was
actually
the
num
like
two
or
three
people
were
like.
Oh
yeah,
that's
really
cool.
I
built
that
also
at
triple
a
game.
Dev
company-
and
I
was
like
yes.
This
is
why
we're
building
it
in
open
source,
because
then
we
can
do
that
they're
like
yeah.
That's
awesome,
great,
that's
great!
I'm
like
come
play.
B
An
air
sandpit
like
come
play
with
us
because
you've
filled
this
before
so
I
might,
I
might
see
if
I
can
remember
who
those
people
were
again
and
start
pulling
them
in
we'll
see
what
we
can
do.
But
it
was
just
a
nice
validation
that,
like
hey,
what
we're
building
is
stuff
that.
A
B
People
need
and
stuff
so
that
we're
super
cool,
so
I
was
I
was.
I
was
at
gdc
making
sure
to
spread
the
good
word
of
quilkin
to
as
many
people
as
I
could
oh
yeah
I'll,
actually
so
april.
I
also
mentioned
it
in
a
talk
just
here
as
well.
As
I
said,
really
nice
things
about
embark
as
well.
So
I
think
there's
a
talk
that
I've
done
in
the
past
by
myself
about
being
successful,
open
source
in
game
dev.
I
did
it
with
april
who
you've
all
met.
B
I
think
a
lot
of
you
met
april.
We
did
that
talk
that
went
really
well.
We
had
really
good
attendance
for
that
50
or
60
people
I
mentioned
cooking
in
that
barely
just
is
a
representation
of
like
really.
B
Between
people
and
like
learning
new
stuff
from
each
other,
but
also
like
highlighted
a
lot
of
the
cool
stuff
that
embark's
doing
as
well,
because
you'll
do
nice
open
source
work.
So
I
like
to
highlight
it.
That's
really
good!
I
just
wanted
to
say
that
I
was
telling
good
stuff.
C
And
how
like
their
approaches,
worked
out.
B
Okay,
go!
Do
my
external
advocacy
stuff
work
so
we'll
do
that.
That
is
fun!
The
only
other
thing
I
had
probably
but
I'd
love
yeah.
We
should.
We
should
talk
about
the
next
year
stuff.
C
B
B
B
D
B
Easy
enough
to
do,
I
can
just
yeah
I'll
just
do
a
little
pi.
In
fact,
I
might.
I
think,
that'll
totally
do
that
today.
Yeah
I've
got
time
perfect.
D
D
C
Like
we've
been
talking
about
wanting
to
run
quilting
as
a
load,
balancer,
and
so
so
of
the
things
I
can't
like
was.
The
issue
was
like
trying
to
use
the
annotations
for
defining
the
qualcomm
filters
becomes
hard
too
impossible
when
we
introduced
like
control
flow
and
having
being
able
to
do
that
doing
that
solely
for
annotations
would
have
been
quite
annoying.
D
C
It's
it
like
yeah,
the
pr
the
pr
is
ready
to
be
reviewed
as
an
xds
management
server.
That
is
how
I've
it
is
the
basic
implementation
of
just
aggregated
state
of
the
world
xds.
We
don't
do
what's
called
delta
or
the
non-aggregated
version,
because
that,
like
I
also
just
don't
think
we
need
that,
like.
I
don't
think
we
need
to
have
four
or
five
different
endpoints,
and
even
if
it's
my
understanding
like
from
just
pouring
over
those
docs
is
like
you
could
have.
C
If
you
needed
non-aggregate,
you
could
just
have
an
aggregate
server
point
to
five
different
endpoints
and
like
because
all
it
is
is
just
like
it's
the
same
types
of
requests,
but
like
the
non-aggregated
one
just
always
expects
like
the
endpoint
service.
Oh,
that
always
sends
endpoint
type
discovery
requests.
C
Yeah,
so
I
don't
know
right,
yeah
yeah,
exactly
like
that,
was
also
the
thing
sort
of
going
through
this
and
like
learning
the
gold
control
plane.
Stuff
is
like
a
lot
of
the
some
of
the
advanced
stuff.
Doesn't
work
or
works
in
a
really
weird
way
anyway,
like
in
the
current
go
control,
plane
stuff
like
there's
still
a
lot
of.
Apparently
it's
like
not
a
perfect
solution,
at
least
just
from
the
blog
post
I've
been
reading
when
I
was
going
through.
C
B
C
Stuff
and
that's
like
yeah,
one
of
the
main
motivations
is
like
having
a
better,
more
integrated
experience
of
like
using
this
stuff,
like
just
even
having
like
to
me.
The
most
compelling
thing
is
that
having
the
quill
can
manage
command
and
just
having
that
work
out
of
the
box
rather
than
having
to
be
like.
Oh,
you
need
to
spawn
up
this
second
server
and
run
that
in
the
background
stuff
and
like
then
we're
getting.
D
C
Yeah
we're
producing
two
things:
we
have
to
manage
two
things
all
the
time
and
like
potential
for
them
to
get
out
of
sync,
and
that's
also
one
of
the
things
that
happened
was
like
cool
can
added
in
some
filters
and
those
filters
changed
and
stuff,
but
like
the
management,
server
did
not
catch
those
things,
and
so
the
management
server
is
incompatible
with
the
current
master
version
of
kulken
anyway.
So.
B
I
look
actually
yeah
when
you
work
down
like
cool,
can
manage
your
gunners.
I
was
like
that's
that's
awesome.
I
was
like
okay
yeah
that
that
definitely
sold
me
on
on
things
as
well.
C
Yeah
turned
out,
we
didn't
need
like
we
don't
need
a
lot
of
like
the
agonist
library
from
the
go.
We
just
literally
just
need
the
crd,
which
I
just
wrote,
and
that's
what
most
of
the
pr
actually
is.
It's
not
even
a
lot
of
logic
mo
like
nearly.
C
Just
me
adding
in
the
crd
with
all
the
documentation.
It's
also,
I
will
say
the
current.
It's
not
like
fully
fully
complete,
like
I
said,
there's
no
improvements
we
made
like
mainly
adding
in
an
administration
server
for
the
xds
and
like
we
can
probably
we're
probably
just
going
to
reuse
the
same
admin
server
we
use
on
the
token
side
just
over
for
the
management
server,
but
for
the
sake
of
review,
it
made
more
sense
to
be
like
here's.
C
Only
the
server
and
here's
only
the
server
logic,
and
then
we
can
have
follow-up
prs
for
adding
in
things
like
the
management,
server
and
metrics,
because
then
we
can
more
easily
go
through
those
things
and
like
discuss.
Okay,
is
this
a
metric
we
want?
Do
we
have
this
card
analogy
rather
than
that?
Getting
like
lost
in
a
much
bigger
pr,
but
like
does
this
thing
do
what
it
should.
A
C
On
the
build
server
can't
find
open.
D
C
Unless
you
like
the
thing
it
might
be,
at
least
from
what
I
think
is
like,
maybe
it's
looking
for
it
on
the
host,
even
if
the
build
image
has
it,
but
I'm
not
quite
sure.
B
It
could
also
be
we're
using
that
tool
for
cross
compilation.
C
Like
it's,
it's
in
the
build
image,
ssl
dev
is
in
the
build.
D
C
But
like
I'm
just
not
sure
if
like
where,
like
the
context
in
which
cross
is
running,
it
might
be
right.
C
No
oh
yeah,
so
yeah
yeah.
I
should
have
added
that,
but
there's
the
one,
the
one
other
additional
change
is
that
you
know
does
not
look
at
annotations
on
the
proxy
label.
Instead,
you
just
have
a
config
map
and
that
config
map
points
just
contains.
Essentially
a
ammo
string
for
the
configuration
of
the
filters.
Yeah.
C
It
just
reads
reads
that
decodes,
that
into
using
the
exact
same
configuration
that
we
use
on
the
cooking
client
config,
just
picks
that
up
decodes
that
converts
it
into
the
protocol
friendly
form
or
whatever,
and
then
sends
that
as
a
discovery
response.
C
C
Well,
well,
the
way
it
works
at
the
moment
is
so
is
that
the
next
time
someone
sends
a
request
right
that
will
pick
it
up
the
way
yeah
we're
not
doing
currently
we're
not
doing
snapchat,
caching
or
anything,
because,
like
it's
just
a
lot
in
my
opinion,
it's
a
lot
easier
for
the
refreshing
instead
of
having
watchers
and
stuff
on
the
kubernetes
side.
To
like
pick
up
those
changes,
it's
much
easier
to
be
like
okay,
the
moment
someone
sends
a
discovery
request.
C
That's
when
we
look
at
the
config
map
and
then
we
send
back
whatever
we
get,
and
then
that's
also.
That
is
also
cached,
because
we're
just
using
the
way
the
discovery
service
provider
trait
works
is
it's
intro
friendly,
so
we
actually
just
have
a
cache
implementation
in
there.
C
That
is
very
just
builds
on
top
of
it
opaquely,
so
that,
if
you
call
you
send
a
discovery
request
within
like
the
next
five
seconds
or
whatever
it
will
be
cached
and
they
won't
look
at
the
config
map,
so
we'll
just
use
the
cache
response,
got
it
I'll.
B
Have
it
I'll
dig
through
as
well,
so
how
I
guess
the
interesting
question
is,
then:
how
do
we
update
routing
tokens
for
players?
Are
we
is
that
future
work,
or
is
that
updating
the
config
map
or
how's
that
going
to
work.
C
But
I
have
to
double
check
because
we're
we're
literally
trying
to
like
right
before
this
meeting
me
and
simon
were
trying
to
deploy
this.
B
C
Work,
I
think
we're.
C
To
populate
those
things,
but
it's
still
just
using
like
token
writer,
like
the
way
it
just
says
it's
just
like.
C
Oh
yeah,
oh
yeah,
no
yeah,
it
still
gets
yeah.
No
sorry
that
has
not
changed
it
emulates
the
exact
same
behavior
as
the
go
code
where
we
go
through
every
agonist
server
that
is
ready
and
allocated
yep,
and
we
get
if
they
have
the
quilting.dev
tokens.
Annotation
right.
B
Got
it
okay,
so
the
the
annotations
for
for
routing
stay
the
same
because
they
stay
on.
B
C
C
And
like
like,
maybe
maybe
we
want
those
like
more
advanced
features
like
delta
and
stuff,
but
like
again,
just
from
looking
through
posts
and
stuff,
it's
like
there's,
not
a
lot
of
performance,
metrics
of
like
state
like
state
of
the
world
versus
delta.
It's
more
just
like
well,
if
you're
running
a
super
large
with,
like
thousands
of
instances,
spending.
C
B
Yeah
yeah.
No,
I
absolutely
agree
and
like
just
the
way
this
is
set
up
right
now,
we're
pretty
much
kind
of
there's
a
there's,
an
implicit
assumption
that,
like
one
cluster,
is
both
kulken
and
agonist
together.
D
B
Ultimately,
not
that
large,
a
number
anyway
you're,
not
like
doing
ten
thousand
or
twenty
thousand
or
something.
B
At
least
not
yet
yeah.
C
It's
definitely
an
improvement
that
needs
to
be
made.
That's
not
doing
at
the
moment.
It's
not
trying
to
do
like
you
can't
go
I'll.
Give
me
cluster
a
b
and
c
it'll.
Just
if
you
have
three
clusters,
it
will
just
send
you
all
the
information
for
those
clusters
yeah
but
yeah,
but
it's
not
that
hard
to
be
like.
Okay,
only
send
the
things
for
cluster,
a
and
plus.
B
B
Yep,
no,
that
I
think-
and
I
think
that's
like
the
best
place
to
start
too,
because
maybe
we
don't
want
to
or
we
need
to
yeah.
C
Exactly
it
should
be
that,
should
that
should
come
from
a
need,
rather
than
I
was
trying
to
guess,
like
maybe
people's
infrastructure
works
like
this.
B
B
C
Yeah
and
we're
going
to
try
to
deploy
this
to
replace
the
old
version
that
we
are
currently
using,
so
we're
going
to
actually
get
experience
with
it.
So
hopefully,
before
this.
B
B
B
A
B
B
Oh,
that
could
be
fun
totally,
possibly
answering
to
that
one
of
the
a
talk
you
probably
would
actually
enjoy
if
you'll
have
fault
access
is
ea
gave
a
talk
about
how
they
containerize
game
servers
and
host
them,
and
what.
D
B
Like
dev
pipeline
looks
like
they
don't
mention
the
tools
they
use.
It
looks
a
lot
like
they're,
not
strangely
enough,
but
they
did
one
thing.
I
actually
really
liked
and
I
was
like.
Oh
that's,
that's
a
clever
idea.
Is
they
used
an
unreal
fps
demo
and
they
so
they
have
their
own
sdk.
So
they
took
that
demo
and
they
integrated
the
sdk
into
that
and
used
that
as
a
demo
project-
and
I
was
like
I
should
do-
that.
That's.
B
Very
smart,
so
yeah.
C
D
B
C
I
just
don't
know
if
there's
one
for
like
unreal
five,
which
is
what
I
think
it
is
just
ue4
anyway,
but
like
you
want,
if
you
want
like
again,
if
you're,
just
talking
about
cool
it'll,
be
very
cool,
if
it
was
the
ue5
stuff
that.
C
B
D
C
B
C
Yeah,
that
would
also
be
great
for
for
development
not
just
for
showcasing,
but
like.
D
A
C
Test
that
we'll
test
the
things
we
want
without
having
to
go
through
all
of
our
infrastructure.
B
Sounds
great,
okay,
I'm
looking
to
see
there's
a
malt.
That's
I
know
four.
Is
there
a
five
out
yet
like
people
play
with
that.
C
C
A
certain
stage
of
development,
like
at
least
like
epic,
is
definitely
using
it
like
fortnite,
is
written
in
unreal
engine.
Five
yep.
C
B
A
B
B
What
was
I
going
to
say
with
the
xcs
stuff,
did
you
want
to
were
you?
Could
you
write
tickets
for
the
the
other
items.
B
The
the
demo
I
do
in
presentations
that
I
need
to
do
more
of
in
a
few
places
is,
is
the
the
example
we've
got,
of
which
one
call
it
of
it's
not
yeah,
I'm
running
it
as
a
sidecar,
which
is
just.
C
C
The
xcx
server
of
like
trying
to
spawn
this
go
service
and
trying
to
like
communicate
on
that
it'd
be
slightly
easier
because,
like
agonist
ships
like
pre-built
docker
images
that
we
can
very
easily
just
pull
and
we're
not
trying
to
develop
a
gunness
while
developing
thing,
but
it'd
be
nice.
If
there
was
a
simple
way
in
the
project.
To
just
be
like
here
are
here
here:
are
the
agonist
tests
for
the
agonist
integration
and
making
sure
that
that
works?
C
B
C
At
the
very
least,
I
would
want
to
be
able
to
do
it
locally.
That's
obviously
what
matters
to
me
is.
I
just
want
to
be
able
to
yeah
run
my
thing
and
just
know
how
that
works,
but
I
I've
I've
never
deployed
the
gunner,
so
I
don't
know
how
hard
or
easy
like
if
there's
a
helm
chair
that
I
can
just
go
for
right.
B
Yeah
there
is
a
home
chart,
you
can
just
kind
of
pull
down
the
way.
Well,
this
is
an
interesting
one.
So
the
way
we've
got
end
to
end
test
running
and
gone
is
you
can
deploy
them
locally
on,
like
for
almost
almost
all
tests,
mini
cube
can
be
a
bit
of
a
pain
for
udp
connectivity,
which
is
always
kind
of
hilarious.
One
works
fine
on
linux,.
B
Well,
so
on
windows
and
mac,
mini
cube
will
run
inside
of
vm.
So
like
it's
hard
to
reach
directly
and
so
like
when
you
go
to
mini
cube
and
you're
like
hey,
what's
your
ip,
and
so
I
can
connect
to
you
directly,
it
go.
It
gives
you
the
ip
of
the
node,
but
it
may
be
the
ip
of
the
node
inside
a
vm.
That's
running
inside,
like
another
network.
B
C
Fun
and
hilarious
is
that
just
minicube
or
like
does
that
also
affect
things
like
docker
desktop
or
what's
the
other
one
rancher
desktop
which
supposed
to
dude.
B
D
B
One
thing
and
make
sure
the
thing
works
it
can,
depending
on
how
you
do
it
or
like
kind,
some
of
them
just
work
a
little
differently,
and
it's
just
it's
all
about
how
the
virtualization
of
containers
works.
B
And
it's
just
pretty
fun
and
mini
cube,
has
a
bunch
of
different
drivers
as
well
so
depending
on
which
driver
you
use
and.
B
Or
docker,
it's
like
it's
just
messy,
but
we
do
have.
Actually
we
run.
B
It
should
be
fine
actually,
because
we
have
the
same
basically
have
the
same
cube
plant
and
rust
as
we
do
in
in
go,
but
basically
we
run
it
as
go
tests.
D
B
Picks
up
the
credentials
and
it
just
from
from
the
local
machine
and
talks
directly
to
the
cluster,
basically
that
we
have
steps
that
install
the
gunners
and
that
kind
of
stuff.
So
we
could,
we
could
potentially
replicate
some
of
that
it
works
both
locally
and
then
it
works
in
ci.
We
run
it
against
the
gke
cluster
and
then
we
just
do
some
locking
to
make
sure
like
no
two
people
are
accessing
the
cluster.
At
the
same
time,.
D
C
C
Well,
we're
just
like
talking
about
stuff.
I
don't
know
if
you
saw
that
the
docker
desktop
it's
now
like
90,
faster
for
reading
files
and
stuff
because
it
uses
fs.
They
ported
that
to
the
mac.
Now
rather
than
using
auc
is
like
grpc
calls.
Every
time
you
were
updating
files,
but
now
it
actually
uses
that
red
hat
file
system,
that's
designed
for
being
shared
between
a
host
and
a
virtual
machine.
B
I'll
write
up
a
ticket
on
that
and
point
to
the
relevant
stuff
that
we
do
in
a
gun
is
because,
once
you
can
talk
to
the
kubernetes
api
that
should
that
gets
us.
That
gets
us
a
lot
of
the
way
there
and
then
you
can
run
it
against
the
local,
local
kubernetes
instance
or
gk
instance,
or
I
I
have
cloud
privilege.
So
I
just
run
this.
B
I
don't
have
to
pay
for
it,
but
yeah
I
have
I
have.
I
have
done
it
on.
I
have
done
it
on
a
linux
machine
before
I
know,
there's
ways
of
doing
it
on
windows.
You
just
have
to
get
the
we
have
some
docs
on
that
and
how
to
connect
to
it.
It's
painful,
but
you
can
do
it.
Actually.
I
think
I
think
most
of
the
tricky
stuff
that
we
ran
into
for
local
issues
is
not
around
udp
connection,
but
some
other
things.
So
it
should.
B
I
think
that's
it
for
me
too,
all
right,
so
I
have
some
things
to
work
on
today,
which
is
good.
B
Oh,
you
can
answer
a
question
for
me
because
I
think
I
got
this
right
in
my
head.
The
the
the
thing
about
switching
from
certain
yamal
to
sir
jason
value.
Oh.
B
Which
makes
perfect
sense
the
what
was
it
you
can
go
you
can
create.
I
think
I
understood
this,
and
this
is
where
this
is
like.
This
is
my
best
knowledge.
It's
not
very
good.
D
D
C
Yeah
and
you
can
also
like
the
way
certainty
works
like
you-
can
actually
just
say
if
you
say
like
let
value
colon
certain
json
value
equals
certain
yamaha
that
will
that
will
decode
a
third
amol
value
directly
into
json.
B
C
Right
yeah
because
yeah,
it's
all
completely
generic
so
like
using,
you,
can
definitely
use
those
json
types
when
decoding
from
yaml
yep
it'll,
just
it'll
just
remove
the
one
edge
case
that
was
while
I
was
writing
the
conversion
code.
I
realized
was
pretty
annoying,
which
was
that
yaml
apparently
is
still
super
complicated,
even
though
they
tried
to
restrict
it.
It's
that
any
any
node
in
a
map
can
be
of
any
type.
So.
B
C
Found
I
learned
a
new
syntax,
the
other
day
of
like
having
an
array
that
then
points
to
an
array.
So
you
can
do
a
bunch
of
weird
stuff
in
yammel
that,
like
don't
make
a
lot
of
sense
or
I.
C
B
C
C
B
Cool,
no
that's
fair
enough.
I
had
actually
we
we
actually
could
as
well.
Although
my
computer's,
updated
yeah.
B
Yeah
something
I
was
looking
at,
how
kubernetes
does
it
where,
where
it's
like,
either
json
or
yaml,
and
it
pulls
stuff
in
it'll,
grab
like
the
first
256
kilobytes
like
it's
a
hard
code
value
or
something
has
a
look
at
it
to
see
whether
it
looks
like
jason
or
not,
and
then
does
the
decoding,
depending
on
which
way
it
looks.
It's
super
hacky,
but
it
works.
D
C
Yeah
good
don't
want
to
be
going
to
like
language
stuff,
but
definitely
going
through
some
of
this
go
code.
I'm
like
I
definitely
disagree
with
how
all
this
stuff
has
been
done,
very
hacky
just
and
stuff
with
like.
Are
you
any
bonus
that
you
have
to
make
your
own
list
types
because
it
has
a
crd,
so
it
needs
its
own
listeners
and
stuff,
and
I'm
just
like
I,
like
the
cube.
You
just
have
a
list
type
like
I
really
like,
like
you.
C
B
D
C
And
stuff-
and
we
were
already
using
that
to
do
the
document.
B
C
Same
unless
simone
orez,
if.
B
You
have
anything
you
want
to
bring
up
and
I
really
need
this
twitter
thread.
No.
A
B
Yeah,
so
you
you
on
a
gunners,
I
used
to
run
a
bunch
of
like
just
used
to
run
mini
cube
locally.
We
had
some
make
targets,
for
I
had
some
make
targets
just
for
creating
the
image
pushing
it
into
mini
cube
and
then
installing
it
that
way,
and
it
was
just
it
would
do
some
we
were
using
a
helm
chart
to
install
agonize
and
you
can
template
what
the
image
parameter
is.
So
you
could
we
just
use
the
hash
of
something
I
can't
remember
what
it
was.
A
B
B
A
B
But
you
could
use
pretty
much
anything
we
used
to
do
it
on
kind
as
well,
but
the
person
who
was
using
kind
all
the
time
just
stopped
doing
it.
What
am
I
looking
for?
Nope?
That's,
not
the
one
develop
testing
and
building
I
got
is.
B
Yeah,
we
even
have
we
even
had
make
files
for
just
like
make
mini
cube
test
cluster,
and
it
would,
it
would
create
like
what's
the
word
it
would
create.
A
B
Only
weird
thing
or
fun
thing
that
we
ended
up
doing
for
a
lot
of
the
tooling
we
had
was
making
sure
this
is
getting
into
the
weights,
but,
like
we
have
a
shell
which
has
like
kubernetes
in
it
and
stuff
like
that
and
doing
like
mini
q,
we
have
a
mini
cube
installed,
just
because
network
had
to
be
host
to
to
push
stuff
to
mini
cube
so
that
it
could.
Rather
because
the
instruction
between
the
two
networking
layers
was
fun
but
yeah.
We
just
we
just
wrote,
make
tests
for
it.