►
From YouTube: 20210218 SIG Arch Code Org
Description
GMT20210218 190500 SIG Archit 1942x1092
A
Hi
hi.
Okay,
so
today
is
feb
18th.
It's
just
the
two
of
us
me
and
narsh,
hoping
that
alina
and
jordan
would
join
at
some
point.
If
not.
A
Thanks,
jordan,
okay!
So
just
to
give
you
an
introduction
of
sick
architecture.
A
Okay,
I'm
gonna
share,
share
my
screen
and
just
say
a
few
things
to
to
give
our
you
go
by
ours
right,
yeah,
yeah,
okay,
our
perspective
of
what
what
we
do
here.
So
there
is
this
thing
called
sig
architecture.
Can
you
see
the
screen.
B
Is
it
just
me
or
is
it
a
single
line
instead
of
a
screen,
I
think
someone
else
can
from
because,
like
I
can't
see
it
properly,
I
just
see
a
white
line.
I.
A
How
about
how
about
now,
yep,
okay,
okay,
so
kubernetes
community
has
a
sig
architecture.
Sig
architecture
is
responsible
for
architecture,
so
there
is
five
sub-projects
in
music
architecture
so
enhancements.
You
can
read
there
so
well,
four
plus
the
main
meeting.
So
currently
we
are
talking
about
the
code
organization
part
here
and
switching
back.
This
is
our.
A
You
know
talk
that
we
take
notes
on
jordan
thanks
for
joining,
so
I
don't
have
anything
specific
jordan
other
than
let's
go
through
old
issues,
and
pr
is
that
okay,
do
you
have
something
else.
D
D
Stefan
took
a
look
at
what
I
had
put
together
as
a
test,
and
it
worked
well
for
him.
I
think
the
one
remaining
question
so
just
to
set
context.
D
Our
client
go
repo
has
some
weird
versioning
issues,
and
because
of
that,
if
someone
just
says
go,
get
client
go,
they
get
like
a
super
old
version
of
it
that
doesn't
actually
work
well
with
latest
versions
of
other
stuff.
Go
116
gives
us
a
way
to
fix
that
by
basically
marking
those
old
versions,
as
ones
that
shouldn't
be
preferred
anymore,
so
that
the
latest
current
one
current
versions
are
preferred.
D
F
D
In
order
to
mark
these
versions
as
bad,
we
actually
need
to
tag
a
new
version
that
includes
those
directives
saying
these
versions
are
bad
and
that
version
has
to
be
the
latest
valid
semver
tag.
So
it
would
be
1.5.2
question
is
what
should
the
content
of
the
1.5.2
tag
be,
which
will
be
used
by
go
versions
older
than
go
116.
D
like
that
that
will
become
the
new
default
version
for
go
versions
older
than
116.?
Should
it
be
the
same
as
1.5.1,
or
should
it
be
the
same
as
11.0.0.
D
It
doesn't
actually
work
well,
but
that's
what
you
get,
which
is
which
is
more
surprising
or
which
is
which
is
less
surprising
for
1.5.2
to
match
1.5.1
with
these
extra
retraction
statements
or
for
1.5.2
to
match
sort
of
the
the
latest
bad
version
you
got
before
we
made
this
change.
D
A
D
A
D
D
A
D
D
Yeah
like
pushing
a
152
that
retracts
that
stuff
is
all
we
have
to
do
to
get
and
then
wait
for
people
to
start
using
one
go
116.
A
Okay,
so
I
so
let
the
plan
of
action
be
exactly
what
you
said:
152
copied
from
151
with
the
extra
retracts
and
then
pushing
it
manually
on
a
specific
date
right
and.
A
D
It's
thursday,
I
will
summarize
this
tag,
stefan
for
comment
and
plan
to
do
it
next
week,
sometime
assuming
like
a
friday,
no,
maybe
mid-week,
so
that
if
there's
an
issue,
we
don't
go
to
work
in
the
weekend
time
to
address
it
before
the
weekend.
Yeah,
let's
say
next
wednesday.
Just
I
just
pulled
that.
A
Data
out
of
the
air
but-
and
we
have
the
approve
and
lgtm
on
this-
there
is
a
hold
on
this
yeah,
so
you
can
unhold
this
and
then
you
will
have
privileges,
and
then
we
can
file
a
revert.
A
Okay,
yep
lubomir
did
you
have
anything
specific
to
talk
about
today.
A
So,
let's
open
up:
let's
do
the
pr's
first,
smaller
asset,
jordan.
I
think
we
need
to
review
this
once
again.
The
first
one,
the
other
three
are
not
ready
by
any
means.
I.
A
What
do
I
remember?
I
I
yes,
this
was
the
last
thing
that
I
did.
I
I
looked
at
it.
I
looked
at
the
files.
There
was
only
one
file
and
the
file
had
didn't
have
too
many
things,
the
the
directory
looked
scary,
but
you
know
when
you
look
looked
at
the
directory.
There
was
literally
nothing
there
except
for
this
one
file.
So,
and
you
know,
I
think
we
should
just
go
ahead
with
this
all
right.
A
A
D
A
Yeah
because
it's
not
maintained
by
anybody
properly,
so
yes
did
you
look
at
the
api
for
the
embed
package,
jordan.
I.
D
A
A
Doesn't
seem
too
bad,
okay,
okay,
so,
and
for
this
one,
I
think
I
have
one
person
who
wanted
to
do
this
siddhartha
chaudhary.
A
I
am
actually
talking
to
him
tomorrow
evening.
I
think
so.
I
set
up
some
time
I'll
walk
him
through
what
needs
to
be
done,
and
hopefully
he
can
help
with
this.
The
next
other
one,
oh
seriously,
somebody
filed
one
after
me.
A
Oh
no,
this
is
an
old
one.
Then
oh,
okay,
never
mind
it's
a
dupe,
either
way,
it's
fine
tools
for
evaluating
dependency
updates
to
kubernetes.
This
is
our
this.
U
we
can
talk
about
it
later.
D
A
That's
I
should
probably
run
this
again.
The
vmware
is
nowhere
near
done
and
I
don't
think
they
are
even
paying
attention
to
it.
Yeah.
D
I
I
mean
at
this
point
I
think
the
cloud
provider
integration
has
a
target,
I
want
to
say
124
yeah,
if
there's
a
an
actual
target
release
for
dropping
that,
and
that
would
resolve
both
of
the
two
remaining
items.
I
would
probably
close
this,
like
nothing
specific
is
going
to
happen
on
those
two
libraries
between
now
and
124
is
my
guess.
A
D
Wasn't
that
related
to
what
the
storage
team
just
asked
about
in
terms
of
deprecating
the
built-in
vmware
support?
Yes,
so
isn't
that
still
targeting
124,
because
that's
a
year,
okay,.
A
Yeah,
that's
a
year,
yeah!
Okay!
This
is
like
maybe
I'll
update
the
current
situation
here
and,
and
then
I
mean
you
can,
if
you
want
okay
I'll
I'll
close
it
out.
I
do
want
to
talk
to
the
vmware
folks.
Once
before
I
close
it.
Let
me
do
that.
A
We
got
rid
of
one
of
we,
we
got
it
off
golan
right
yeah.
We
still
have
govet
and
static
correct,
so
what
I
ended
up
doing
was.
I
went
down
another
path
which
is
to
see
I
wanted
to
introduce
gold,
golang,
ci
lint
in
some
fashion,
and
what
I
ended
up
doing
was
checking
if
we
could
run
just
these
three
dead
code
unused
and
var
check
right,
so
dead
code
unused
and
var
check
basically
pop
up
things
that.
A
Unused
code,
so
so
a
new
I
I
I
messed
around
with
struct
check
and
I
decided
okay,
we
are
not
going
to
do
struct
check
because
it's
finding
stuff
that
is
actually
valid
users
and
which
you
know
you
stuff
in
something,
and
then
you
serialize
it.
You
know
and
it's
not
understanding
that
it
is
still
being
used.
So.
A
Right
yeah,
the
point
here
also
was
to
gently:
introduce
it
so
so
definitely
aircheck
and
enough
assign.
I
will
go
next,
but
the
idea
here
was
to
do
at
least
these
three
then
see
how
many
places
we
need
to
fix
up,
and
with
these
just
so,
we
found
like
a
two
dozen
places,
which
are
all
simple
delete
line,
delete
a
line,
delete
a
bunch
of
lines,
kind
of
thing.
A
So
I
think
we
are
good
here,
I'm
working
with
another
student
who
wants
to
file
a
pr
for
this
when
she
files
a
pr
for
the
set,
will
ask
for
your
approval,
jordan
and
then,
when
this,
when
that
pr
gets
in,
then
I
will
check
in
this
script,
but
I
will
not
run
it
by
default
right.
A
So
yeah,
which
is
what
I've
been
doing,
I
was
actually
using
my
intel
my
go
land
to
do
this
periodically
so
now
I
don't
know
how
to
do
this,
so
I'm
happy.
Okay,.
D
My
guess
is,
it
doesn't
catch
public
things
because
it
doesn't
know
if
they're
used
or
not
so
it
looks
like
it's
only
flagging
private
things,
because
you
can
actually
reason
about
it.
That's
right.
A
A
I
was
like
going
through
each
line
by
line
and
I
made
sure
that
it
doesn't.
That
is
exactly
the
problem
we
had
lubomir
when
we
were
telling
people
to
fix
golan
and
people
who
are
changing
public
stuff
right.
So
we
don't
want
to
repeat
the
same
mistake
here.
I
agree.
Okay,
so,
and
also
this.
This
will
help
get
more
people
into
trying
things
out
with
kubernetes
filing
prs
and
doing
actually
things
that
we
would
like
them
to
fix.
So.
D
Go
back
there.
There
was
a
public
method
that
it
flagged
must
or
was.
It
must
parse
something
versions.
Yeah
I
mean
that's
under
cluster
images
of
cd,
so
that
actually
would
be
fine
to
remove.
But
if
it's
flagging
public
functions
then
it
would
probably
flag
them
under
staging
as
well
is
my
guess.
A
That's
right,
so,
okay,
how
do
we
do
this,
then.
D
I
don't
know
I
mean
we,
we
can
know
that
that
is
like
we
shouldn't
remove
public
stuff
or
we
can
look
at
and
say.
Okay,
that's
not
actually
that's
a
file
that
we're
not
publishing
in
a
module
for
consumption.
It
would
be
nice
if
you
could
say
like
unused
private
or
unused
public,
but
I
don't
know
if
it
has
that
level
of
granularity
well,.
A
Another
reason
not
to
switch
this
on
in
ci
right,
okay,
good!
We
are
in
sync
on
that
yeah.
F
I
think
this
litter
supports
pragmas,
so
we
can
flag
some
of
them.
A
Yeah
the
problem
lubomir
is
when
we
tell
people
who
are
just
calling
kubernetes
to
run
the
script
and
fix
the
errors
that
it
shows
right.
They
will
end
up
just
deleting
the
method
because
they
don't
understand,
they
might
not
see
the
problem
that
this
is
like
a
public
method.
It
might
be
called
by
somebody
else
in
the
ecosystem.
Right.
D
I
mean,
alternatively,
if
you
have
a
public
method,
that
you
have
no
tests
around,
that's
pretty
questionable,
so
I
think
even
just
adding
a
test
would
make
that
be
used
so
yeah,
if
you're,
if
you're
exporting
a
public
api
and
not
testing
it,
that's
really
questionable.
Actually,
but
the
right
thing
to
do
there
is
probably
test
it
not
remove
it.
A
Okay
sounds
good
anyway.
We
won't
have
to
worry
too
much
right
now,
because
this
is
going
to
be
like
a
controller
environment.
We
are
not
telling
we're
not
opening
it
up
to
the
ci
okay.
So,
but
the
point
here
is
over
a
period
of
time.
A
Let
me
see
how
many
things
we
can
fold
into
this
script
and
delete
duplicate
scripts,
and
the
other
problem
to
think
about
here
is
like
how
much
time
this
takes
by
itself
and
how
much
time
does
the
other
ones
take,
and
then
we
have
to
figure
out
whether
we
need
to
run
another
job
and
how
to
divvy
up
where
to
run
the
scripts
and
stuff
like
that.
Anyway,
that's
far
down
the
line,
so,
but
you
know,
started
the
process
here,
I
don't
think
there
was
any
update.
A
F
F
A
A
Anyway,
I'll
take
a
look
at
this
one
later,
I
think
that's
about
it.
The
rest
are
all
old
stuff.
Oh
grpc,
can
we
do
something
for
122.
D
Jordan,
the
discussions
around
that
are
continuing-
I
don't
know
yet
it
it
hasn't
stalled
out
like
there
are
still
people
paying
attention
to
it.
So
peter
is
working
on
trying
to
force
a
decision
there
in
terms
of
like
migration
or
maintenance
or
fixing
or
patching.
Like
he's
continuing
to
push
on
that,
I
don't
have
a
good
answer
yet
for
what
we'll
need
to
do?
D
Okay,
which
peter
is
this
works?
He
works
on
the
ncd
side,
but
he
let.
D
A
T-A-V-R,
like
that,
yes,
okay,
I'll
just
see
him,
so
I
remember
who
it
was:
oh
theater,
okay,
okay,
so
that
that's
all
here
now,
jordan,
if
you
remember
some
time
ago,
we
ended
up
writing
this
in
our
meeting
notes.
A
So
I
turned
that
into
a
proposal
to
the
lfx
mentoring
program
and
quite
a
few
people
got
interested
alina.
A
Alina
alina,
what
never
mind
elena
katz
said
that
she
might
be
help
interested
in
helping
a
student
with
this
program
as
well.
So
we
ended
up
trying
to
expand
on
the
things
that
we
were
looking
at
as
background
research
and
a
couple
of
people
chimed
in
saying
that
they'll
be
interested,
including
harsh
harsh.
B
Yeah
hi
everyone,
so
what
I
was
like
discussed
with
emanum
was
like
the
blog
you
wrote,
which
used
like
grep
and
wn
w
gate
and
all
these
scripts.
So
basically
we
could
have
a
custom
cli
where
we
try
to.
You
know,
have
a
single
command
for
all
of
these,
whereas
internally
we
are
using
something
similar
to
crep
and
duplicate.
I
actually
posted.
F
D
B
It
yeah,
so
this
is
like
what
the
venom
proposed
like
we
get
the
patch
file
of
a
pull
request,
and
then
he
wrote
some
shell
scripts,
which
was
like
combining
commands,
mostly
grep
and
w
gate
so
like
first,
this
is
what
you
these
are
like
the
original
dependencies
you
get
when
you
follow
all
this,
but
I
mean
the
end
goal
would
be
like
not
having
to
do
all
this
and
you
know
like
make
it
more
customizable
and
something
which
maybe
we
could
add
in
a
pipeline,
or
something
like
that.
B
A
So
I
would
twist
it
around
a
slightly
different
way,
so
this
basically
helps
you
start
with
golang
and
it
helps
you
understand
our
requirements
and
stuff
like
that.
That's
for
sure,
but
I
want
to
start
from
the
other
end,
the
other
end
being.
A
To
see
and
then
walk
backwards,
because
what
you're
doing
is
implementing
the
hackmd
that
I
put
together
for
what
are
the
new
dependencies
and
who's
calling?
What
kind
of
thing
right?
Let
me
just
show
jordan
that
part.
B
A
Okay,
so
this
is
jordan
going
back.
If
you
remember
the
customized
problem
that
we
were
working
on
the
customize,
you
know
there
was
a
new
version
that
came
out
and
monopole
jeff
updated
the
pr.
So
I
ended
up.
A
I
needed
this
right
like
I
wanted
to
see
what
are
the
new
packages
that
are
getting
added
and
what
is
the
call
stack
basically
for
for
the
for
the
ones,
so
I
ended
up
putting
together
this
hackmd,
which
is
what
ash
was
talking
about
right,
but
I
want
to
flip
the
thing
back
to
the
original
problem
that
we
were
trying
to
discuss.
Is
this
one
right?
What
is
the
information
that
we
need
to
print
in
a
report
for
every
pr
that
gets
logged,
and
that's
where
I
want
to
start
jordan?
D
Yeah,
I
think,
looking
at
pull
requests
where
we
manually
noticed
or
caught
things
and
like
saying
what
what
was
it
that
we
caught
there
was
it
that
there
were
a
bunch
of
new
dependencies
or
that,
like
we
had
a
lot
of
version,
disagreement
on
existing
dependencies
or
like
those
are
the
two
that
are
the
most
obvious
and
easiest
to
catch
manually,
like
so
being
able
to
say
like
before
this
pr,
we
had
300
things
in
our
dependency
tree
and
after
this
pr,
we
have
400
things
in
our
dependency
tree
like
that
just
number
of
dependencies.
D
That's
a
pretty
easy
one,
but
useful
depth
of
dependencies
like
before
this.
The
deepest
dependency
tree
we
had
was
whatever
seven
and
now
it's
16.
D
One
that
we
see
a
lot
that
I'm
not
sure
how
easy
it
would
be
to
pull
out,
but
we
see
dependencies
that
we
don't
actually
use
like
they
don't
get
pulled
into
our
vendor
tree,
but
they
do
get
pulled
into
the
dependency
tree.
So
dependencies.
D
D
It
might
depend
on
stuff
for
like
release
engineering
reasons
like
they
use
particular
tools,
and
they
just
happen
to
include
those
in
their
module,
even
though
they're
not
required
to
build
or
test
or
ship
their
thing,
or
it
might
be
things
they
require
from
other
packages
which
we
don't
use.
D
So
those
those
are
the
three
sort
of
categories
of
transitive
dependencies,
and
so
what
that
looks
like
is
those
don't
show
up
in
our
vendor
directory,
but
it
does
show
up
in
our
go
mod
files
and
it
makes
those
relationships
way
more
complicated
and
so
like
dependencies
that
we
don't
vendor,
but
do
that
are
still
in
the
dependency
tree.
Like
that's
an
interesting
metric.
D
So
before
you
know,
before
this
pull
request,
there
were
100
things
in
our
dependency
tree
that
we
don't
vendor,
and
after
this
there's
200
things
in
our
dependency
tree
that
we
don't
vendor.
That's
a
smell
that
we
probably
don't
want
to
take
this
update
without
at
least
going
to
the
upstream
component.
Saying
can
you
like
carve
some
of
those
off
or
drop
some
of
those
release?
Engineering
tools
move
those
to
a
tools,
sub
module
or
yeah.
D
So
we
we've
seen
that
in
c
advisor
c
advisor,
the
c
advisor
binary
had
a
ton
of
dependencies
that
we
didn't
even
use,
and
so
we
got
them
to
split
that
off.
The
customized
thing
is
a
good
example
of
that,
where
we're
not
actually
using
this,
but
some
other
bit
of
it,
depended
on
it
so
number
dependencies
depth
of
dependencies
dependency.
Is
we
don't
vendor
but
do
on
in
our
go
mod
file?
Now
those
are
the
three
most
obvious
ones
that
tend
to
get
noticed
manually
and
spur
like
pushback
or
discussion.
B
So
but
this
would
still
be
for
like
whenever
a
new
pur
introduces
dependencies
right
and
not
for
the
existing
ones.
D
I
mean
being
able
to
evaluate
it
on
the
existing
tree
is
useful
so
that
we
can
track
improvement
over
time
right
like
ideally,
we
would
cprs
that
make
these
things
go
negative,
like
fewer
dependencies
and
a
shallower
dependency
tree
and
fewer
things
in
our
tree,
that
we
don't
actually
vendor
so
being
able
to
run
this
just
against
head
would
be
really
useful,
and
then
you
can
do
that
on
a
particular
pull
request
by
running
it
on
head
before
the
pull
request
and
diffing
against
the
results
of
the
pull
request.
A
D
A
Yeah,
so
also
one
of
the
benefits
here
is
when
people
are
doing
this,
they
will
be
able
to
spot
it
themselves
without
us
having
to
go.
Tell
them
look.
This
is
bad.
D
Yeah
I
mean
these
three
numbers
would
be.
D
A
Okay,
does
this
give
you
enough
to
write
your
request
for
the
lfx
thing.
B
There's
actually
no
request
of
the
portal
just
had
us
like
submit
a
cover
letter
and
a
copy
of
our
resume,
which
I've
already
done.
So
that
is
why
I
started
working
on
the
poc
and
link.
I
know
more
about
the
project.
A
Okay,
so
what
I
would
then
say
is
start
a
google
doc
write
your
thoughts
about
you
know
what
we
discussed
just
now
and
have
lou
bomer
me
and
jordan,
pika
and
alina,
of
course,
right
give
examples
of
give
examples
of
exactly
what
you,
what
end
user
is
going
to
see
and
then
we'll
work
backwards,
to
figure
out
how
to
get
that
information
and
how
to
collect
what
code
to
write.
What
you
can
reuse,
that
kind
of
stuff.
B
A
Yeah,
so
what
I
would
say
is
it
would
run
against
the
current
so
start
with
a
go
mod
gomain,
which
will
generate
a
text
file.
B
A
So
start
small
right
just
do
a
minimum
yeah
main.go,
which
collects
these
three
data
points,
puts
it
into
a
json
file
and
we'll
figure
out.
You
know,
after
that.
A
A
Yeah
try
to
get
some
sleep.
Lubomir
is
in
the
west
coast,
jordan
is
in
the
east
coast,
I'm
on
the
east
coast,
so
we
we
can
probably
at
some
point,
revisit
the
timing
of
this.
If
you
show
up
regularly
to
this
meeting,
okay.
F
A
F
I
had
a
one
question
about
this.
The
previous
item
is
this
related
to
the
lfx
work
for
mentoring
at
the
cncf.
F
F
A
Exactly
there's
at
least
two
or
three
other
people
who
are
interested
in
the
same
thing.
A
Yeah
we're
going
to
follow
the
process
that
lfx
sets
out
for
sure,
and
this
is
also
like
a
welcome
to
the
community
kind
of
thing
for
us
here.
F
This
whole
thing
is
not
versioned
with
the
kubernetes
correct
this
release
they
just
they
are
versioned
with
some
arbitrary
attacks
which
are
in
the
kubernetes
release
repository,
and
this
is
obviously
a
kubernetes
sorry,
a
sick
release
topic,
but
we
have
been
discussing
recently
bringing
back
because
they
used
to
live
in
kkk,
bringing
back
the
system,
the
specs
in
the
package
specs
to
kk,
because,
according
to
ben's
opinion
and
mine
as
well
this,
basically
we
have
to
version
them
with
the
source
code.
F
A
Yeah
yeah:
what
do
you
think
so
sig
release
folks
for
good
or
bad?
They
started
carving
out
things
that
they
can,
that
they
need
to
touch
and
putting
their
themselves
in
owner's
files
and
things
like
that
and
like
moving
them
to
places
where
it's
convenient
to
them.
A
This
is
one
of
those
I
don't
have
a
very
strong
opinion,
but
I
do
respect
the
boomer
you
and
ben
what
you're
thinking
about.
I
would
differ
it
to
jordan.
What
do
you
think
jordan.
F
To
buy
the
setting
originally
the
system
d
and
a
package
specs
were
in
kk.
At
some
point,
we
moved
we
duplicated
the
the
specs
in
the
k,
release
repository
and
bazel
built.
The
basic
build
were
producing
from
kk
was
producing
artifacts
in
one
way.
The
other
one
was
the.
What
was
the
other
build
like?
The
other
group
built
was
producing
from
k,
release.
D
Release
tooling,
living
in
k
release
makes
sense
if
there's
anything
that
would
be
version
specific,
like
component
invocations,
then
that
should
be
in
kk
or
in
a
repository
which
matches
kk
versioning
so
has
a
release,
118
1920
branch-
I
don't
have
a
strong
opinion
about
which
approach
to
follow.
I
think
it's
easier
for
the
cubelet
invocation
to
live
where
the
cubelet
code
is
defined.
D
I
think
the
only
reason
the
current
system
has
worked
as
long
as
it
has
is
that
the
cubelet
invocation
changes
so
rarely,
but
that
doesn't
mean
this
is
a
good
approach,
so
either
the
cubelet
invocation
piece
should
live
in
kk
or
the
place
it
lives
in
k.
Release
should
be
versioned
by
branch
or
by
folder,
along
with
the
kubernetes
releases,
I
will
defer
to
the
node
and
the
release
folks
about
which
approach
they
want,
but
I
agree
that
a
single
unversioned,
cubelet
systemd
spec
in
some
other
repo,
is
not
workable.
Long
term.
F
Yeah
completely,
basically,
one
of
the
problems
we
have
seen
is
that
we
want
to
make
a
change
in
the
spec
and
then
secretly
it
pushes
the
button
they
release
packages
for
all
the
binaural
versions,
like
patch
pressure
releases,
and
this
can
break
in
terms
of
you
know,
rpm
dependency
somewhere
or
it's.
D
Yeah,
I
would
gather
the
examples
of
that
happening
in
the
past
and
open
an
issue
saying
the
spec
needs
to
be
per
kubernetes
version
and
point
to
the
problems
that
have
been
caused
in
the
past
and
then
kind
of
lay
out
those
three
options
like
keep
it
where
it
is,
but
make
a
folder
for
each
version,
keep
it
where
it
is,
but
make
a
branch
for
each
version
or
pull
it
from
kubernetes
kubernetes
as
the
source.
I
would.