►
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).
A
I
think
from
from
outside,
and
one
of
the
most
important
parts
is
also
upgrade
testing,
which
I
guess
we're.
Gonna
talk
a
bit
about
today
and
yeah,
probably
some
planning
for,
like
the
next
cycle,
at
least
drafting
something
I
know.
Tim
said
yesterday
we
should
talk
about
AJ
I,
agree
that
salsa
there.
B
A
C
C
C
C
I
I
cut
off
the
end
of
that
link.
My
bad,
oh.
A
We
I'm
not
sure,
what's
up
with
the
cubelet
there,
but
yeah
we
definitely.
This
has
to
get
get
that
anyway.
I'm
I
think
it's
okay
to
like
merge
the
things
into
kubernetes
anywhere
and
testing
front
and
like
say
that
they
are
with
a
reason
red
right
now.
We
have
other
just
use.
That
also
are
red
because
of
this,
which
is
partly
why
I
found
it
when,
when
investigating
that
one,
does
that
sound
good
to
you,
I.
C
Suppose
I
could
do
that
I
think
I'm,
because
I
don't
have
a
fully
successful
and
to
end
with
the
cube
admin
upgrade
command.
I,
don't
know
if
I
have
done
everything
correctly.
On
my
end,
so
let's
say
this
bug
gets
text
and
maybe
I
forgot
to
do
something
else
at
some
other
variable
or
runs
another
command.
I
won't
know
until
after
this
gets
fixed.
That
I
need
to
do
some
more
changes
to
actually
make
it
work.
You
can
use.
A
C
C
A
C
C
A
D
D
D
So
basically,
like
the
current
test
helpers
in
I,
think
it's
like
e
to
e
directories
and
stuff
like
that,
this
that
we
need
to
add
a
like
another
switch
statement
so
that
the
the
tests
can
actually
invoke
communities
anywhere,
because
right
now
the
only
providers
are
supported
are
like
GC
g
ke,
and
I
think
that's
it.
So
we
need
to
add
another
thing
with
kubernetes
anywhere
to
be
invoked
from
from
the
tests.
Is
that
a
fair
representation
jessica?
Do
you
want
to
add
more
content?
Man?
Maybe.
C
A
Question
like
from
what
I
understood
this
is
for
the
upgrade
or
feature
master
upgrade
test
ETA
test
right.
They
are
using
this
code.
Can
we
initially
like
for
the
first
testing
facility
bit
of
it?
Just
do
conformance
tests
after
upgrade
is
made
and
then
like.
If
we
get
this
into
one
eight
chair,
attics
and
all
that
I
mean
we
definitely
should
but
not
block
the
initial
test
coverage
on
this
feature.
Master
or
cred
edie
test
tag
reasonable
well,.
C
So
the
upgrade
logic
is
in
this
upgrade
tests,
which
is
in
kubernetes,
kubernetes
and
so
right
now,
if
we
don't
add
that
fork
to
actually
call
kubernetes
anywhere,
it's
gonna
try
to
call
the
coop
up
script
likely
that's
called
and
that's
not
gonna
work
for
kubernetes
anywhere
set
up,
so
it
won't.
It
won't
work
at
all.
So
we
do
need
this
if
we
want
to
have
the
upgrade
tests
working,
I.
A
Thought
that,
like
I,
thought
that
first
a17
cluster
would
be
spun
up,
then
we
run
conformance
tests.
Then
we'd
exit
like
testing
for
the
cube
test
binary
somehow
after
doors
are
finished,
like
upgrades,
cube
anywhere
and
then
when
the
upgrade
is
done,
conformance
tests
are
like
run
again.
That
was
my
mental
model,
but
yeah.
C
A
B
Know
like
because
you
can
switch
on
provider
if
we
can
somehow
like
link
through
provider.
You
know
through
this
is
a
bigger
change
right,
but
eventually
link
through
provider
and
do
the
switch
of
the
right
thing
in
the
right
spot,
because
right
now
that
the
upgrade
test
jiggery
is
a
little
riddle
me
for.
B
F
F
E
E
We're
still
there
working,
which
is
the
new
upgrade
framework
that
they
Kris
Rousey,
wrote
to
automate
that
those
tests,
I
think
have
actually
been
relatively
stable
and
and
sort
of
passing
the
ones
that
have
been
really
hard
to
get
working
and
get
a
signal
out
of
be
partly
because
I
really
slow,
and
then
they
run
all
of
our
tests,
and
many
of
which
are
flaky
are
their
operating
tests.
Where
we
like
provisional
cluster,
we
run
e
two
E's,
we
upgrade
the
cluster,
we
run
ETS
again.
E
If
anything
fails
during
that
whole
process,
and
then
it
takes
12
hours
to
repeat
so
like
ironing
out
flaky
signal
is
also
really
slow
and
the
other
problem
with
those
is
that
those
tests
are
sort
of
conflating
multiple
things.
They're
conflating
like
versions,
q
testing,
+,
upgrade
testing.
You
know
versions
hue
in
both
directions,
plus
upgrade
testing,
and
I
would
like
to
start
to
break
that
apart.
E
So
I
think,
for
for
this
release
cycle,
the
plan
with
the
cube
and
operate
tests
is
to
verify
that
cube
admin
grades
actually
work,
but
not
to
verify
that
all
of
kubernetes
works
in
an
upgrade
scenario,
because
we
have
the
old
test
framework
to
do
that.
I
would
love
to
at
some
point,
throw
it
away
and
replace
it
with
something
different,
especially
as
we
want
to
get
rid
of
cube
and
the
corresponding
great
shell
script,
which
is
pretty
terrible,
but
we're
not
quite
there
yet
I.
Don't
think.
A
I.
Don't
know
like
that.
Cube
we
specify
like
need
us
anywhere
upgrade
provider,
cube,
ATM
and
testing
for
the
cube
test
will
run
the
conformance
test
once
it
will
accept,
like
whatever
thing
inconvenience
anywhere,
we
specified
as
an
argument
and
then
it
will
run
them
the
conformance
test
again
then,
for
1/9
we're
gonna
look
into
how
to
make
the
master
upgrade
EDS
work
more
generally,
like
not
switch
on
provider
like
in
core
in
in
in
core
code
and
somehow
improve
that
I.
E
If
we
do
that,
then
we're
basically
doing
the
old-style
testing
where
we're
running
two
sets
a
conformance
tests
and
not
verifying
that
the
upgrade
process
itself
doesn't
break
you
and
that's
the
thing
that
we
use
to
do
manual
testing
for
so
I.
Think
if
we
don't
run
the
new
during
a
parade
part
of
the
upgrade
tests,
we
are
not
actually
getting
coverage
that
cube
admin
upgrades
work
like
don't
break
you
during
the
upgrade
like
I,
think
would
you'd
be
verifying
there.
A
E
It's
not
just
that
your
costs
are
still
healthy.
It's
that
an
object
you
created
in
your
cluster
at
the
old
version,
still
exists
and
still
works,
like
I.
Think
we've
had
bugs
in
the
past,
where,
like
the
upgrade
process,
will
rejigger
stuff
in
SED
and,
like
you
might
like
lose
objects.
Right
is
a
potential
scenario.
We've
been
concerned
about
so
you
know
you
know.
E
Obviously
this
is
sort
of
anticipate
with
alpha
objects,
but
say
you
create
an
alpha
cron
job
in
your
1/7
cluster
and
you
upgrade
to
1/8
and,
like
it's
possible
that
that
thing
just
gets
erased
from
at
CD
or
you
can
no
longer
access
it
right
and
we
want
to
make
sure
that
doesn't
happen
for
things
that
are
beta
and
above
objects
right,
and
so
that's
what
this
this
creates
something
beforehand
and
upgrade
and
verify
that
it's
still
there
and
still
working
it's
not
just
that.
The
upgrade
is
not
disruptive
because
it
often
will
be
disruptive.
E
It's
the
things
you've
created,
don't
get
destroyed
during
that
upgrade
process,
because,
if
all
you
do
is
run
conformance
test
afterwards,
you're
verified
it's
a
new
cluster
new
things
you
created
after
you've
upgraded
work.
It's
not
that
things
you
created
beforehand,
still
work
right
and
I
think
that's
the
key
part
that
would
be
missing.
A
Ok
come
so
regarding
a
timeline
I'm
just
wondering.
Is
it
better
to
try
to
push
the
the
full
feature?
Master
upgrade
thing
in
the
core
at
this
stage
or
I
mean
I
I
thought
that,
like
what
we
supported
for
this
version
was,
like
conformance
test,
upgrade
conformance
test,
not
the
like
what
you
described
a
minute
ago.
I
mean
it's
more
a
question
about
like
this
is
bida.
A
A
Yeah
I'm
just
wondering
whether
we
should
build
in
companies
anywhere
code
at
this
stage
in
the
release
to
really
test
either
a
test
right.
I
can
just
imagine
that
it
can
be
somewhat
heavy
like
should
we
then
block
the
release
on
this
thing
right?
You
bet
him
up.
Criteria
tests
aren't
working
because
there
was
a
bug
in
the
test
binary.
I
yeah,
I'm
just
wanna
I,
just
wanna.
Ask
these
questions
in
beforehand
before
we
do
and
I
mean
I
think
it
sounds
good,
but
we
really
have
to
make
sure
yeah.
E
So
yes,
I,
guess
I'm,
not
sure
I
understand
your
concern
about
modifying
q
Brady's
anywhere.
It's
not
actually
tied
to
the
cure.
Aids
release,
I'm,
not
even
sure
it
has
its
own
release
cycle
like
at
this
point
or
sort
of
using
it
as
test
infrastructure
for
Q
Badman
I'm,
not
sure
if
it's
actually
being
used
for
anything
else,
so
trying
to
sort
of
get
that
updated
so
that
we
can
use
it
as
operate
test
infrastructure
to
I.
E
E
I'm
just
saying
like
we
did
some
manual
tests,
like
you
said
it's
beta
and
that's
maybe
good
enough
for
now,
but
I
think
we
should
try
to
get
automated
upgrade
tests
working
because
we're
gonna
want
to
have
those
on
the
master
branch
anyway,
as
more
than
one
nine
development
cycle
to
verify
that
we
catch
breaks
as
early
as
possible.
Yeah.
A
We
can
update
that
really
quickly
and
and
also
like
get
get
new
runs,
but
we're
depending
I'm
just
saying
that
we're,
depending
on
the
release
team
to
fast
forward
like
until
we
have
no
until
the
master
and
release
one
eight,
is
starting
to
diverge
and
then
quench
are
picking
and
the
CI
jobs.
That's
wrong.
E
Yeah
I
think
it's
might
be
starting
to
diverge
today.
Right
I
think
we're
already
at
the
point
where
they're
not
fast-forwarding
and
I-
think
I
heard
yesterday
that
they
might
be
opening
the
master
branch
for
one
line
stuff.
So
already
at
the
point
where
it'd
be
cherry
picks
to
get
into
what
eight
I
I'm
not
sure
we
should
let
that
stop.
E
Us,
though,
like
if
we
think
it
should
be
put
in
and
cherry
pick
to
validate
the
one,
eight
release
and
I
think
we
have
an
argument
to
send
that
PR
and
if
they
say
no,
it's
too
risky,
then
we
do
a
little
bit
of
manual
testing
and
try
to
get
it
in
the
in
the
master
branch.
So
we
get
testing
it
head
right
where
we
want
it
in
the
mouth.
French
either
way
right.
If
they
reject
the
cherry
pick,
then
they
reject
the
parrot
terrific,
but
we
should
put
it
in
the
master,
birch.
E
A
E
E
A
C
A
B
Have
comments
but
they're
not
there
related
to
this
to
testing,
but
not
the
automation.
I
mean
should
I
go
with
that
now
or
should
we
tell
sure
sure
so,
I
started
doing
some
skeet
esting
on
my
own
to
validate
the
upgrade
process
and
I
came
across
some
bunch
of
thorny
issues
that
some
of
which
have
been
lingering
for
a
while,
but
I
was
never
able
to
reproduce
them
or
anniversa
at
the
time
to
dig
into
it.
B
We
believe
it's
definitely
related
to
some
race
condition
in
C
and
I.
This
is
a
hypothesis.
We
don't
have
solid,
definitive
proof
and
one
of
the
questions.
I
had
when
I
was
evaluating.
This
is
that
I've
uncovered
that
we're
still
running
really
old,
C
and
I.
We're
planning
on
updating
that
it's
really
in
the
game
to
do
that,
but
we've
never
actually
had
real
test
cycles.
B
B
A
B
A
B
Seems
it
seems
walkie
to
be
releasing
these
things
and
there's
known
revisions
which
have
known
fixes
and
other
modifications
to
them
right
we
should
be
the
node
team
should
be
tracking.
You
know
the
versions
they
support
and
updating
the
list
in
the
test
coverage
to
support
these
other
versions.
Right,
I,
don't
think
the
onus
should
be
on
us
and
this
one
because
they
own
the
integration
there
yeah.
B
The
the
code
for
all
to
spin
up,
though,
is
on
the
node
but
see
I,
see.
I,
see
your
point.
The
the
second
issue
that
we
had
as
I
mentioned
was
the
crash
new
pack
off
one
I,
don't
know.
If
Craig
wants
to
talk
about
this
a
little
more
cuz,
he
was
looking
at
one
condition:
I
get
a
much
nastier
version
of
that.
Where
my
whole,
my
whole
machine
goes
down.
B
G
I
mean
I
gets
picked,
Oh
real,
quick
I'm,
actually
looking
at
as
we
speak,
but
I
what
I
can
see
it
seems
to
be
somewhere
within
the
cubelet
or
the
CNI
I.
Don't
think
it's
in
CNI
itself
there's
a
chance
that
it
could
be
in
calico
just
a
plugin
that
we're
particularly
using
but
I
I'm,
suspecting
more
on
the
cute
side.
But
I
am
in
the
midst
of
that
code.
Right
now,
but
we
are
not.
G
The
only
good
thing
I'll
say
is:
we
are
the
only
users
with
this
particular
problem
and
from
what
it
looks
like
it's
as
Kim
described
it
earlier
as
namespace
leaking.
Is
that
two
of
these
odds
actually
come
up
on
the
wrong
in
space,
very
ruling
on
and
then
they're
unable
to
communicate
with
API
server?
If
you,
if
you
delete
them
where
you
scale
them
down
to
zero
and
then
back
up,
they
actually
come
back
on
the
right
namespace.
So
this
awesome
might
have
security
conditions.
B
Yeah
and
I've
also
witnessed
I
think
what's
occurring
on
my
environment,
because
I'm
really
like
a
fedora
kernel
versus
like
some
older
version
of
it
I'm
sure
they
probably
did
some
optimizations
and
I
get
a
hard
crash.
When
I'm
running
I
was
running
the
SKU
tests,
the
intent
test
with
you
know
as
many
as
I
could
so
like
600
tests
against
Oh
1/8
tests
against
one
7ku
BDM
install
it
was
gonna
flip
it,
but
I
never
got
past
that
point
because
it
kept
on
exploding.
B
A
B
A
A
A
Should
we
then
like
say
the
users
like
upgrade
Calico
on
your
one?
Seven
cluster
first
then
do
qubit
and
upgrade
because
otherwise
it
will
totally
break
or
what's
the
preferred
way,
I
mean
we
when
we
upgraded
C
and
I,
as
we
all
know,
in
one
six,
everything
broke
is
cubelets.
Had
this
new
notion
of,
like
oh
I,
should
send
this
error
to
see
and
I
hear
and
it
wasn't
compatible
with
the
season-high
Worsham
we
were
using.
We
had
to
bump
that
and
got
into
the
node-red,
no,
not
ready
so
yep.
B
There
was,
I
I,
think
what
I'm
gonna
probably
start
doing
more
of
now,
despite
my
my
desire
not
to
is
talk
with
the
go
to
the
Signet
workgroup
meeting,
just
to
verify
what
versions
they
plan
to
support
and
make
sure
that
their
revision
a
and
have
some
level
of
integration
testing
across
these.
So
many
X
providers,
because
we
have
no
coverage
right.
We
have
cout
medium,
which
basically
says
we
have
all
of
these
different
providers.
We
have
no
test
metrics
of
support
across
these
providers
at
all,
so.
B
A
A
Yeah
I
mean
we
have
to
somehow
choose
I
mean
I,
guess
we're
gonna
do
self
hosting
by
default
as
soon
as
possible,
in
one
nine
timeframe,
I
like
enable
that
in
in
the
feature
gate
which
will
automatically
get
picked
up
by
by
the
CI,
but
then
we
should
also
do
the
like
static
ports,
only
thing
to
still
do
not
drop
coverage.
For
that
then
I
noticed
Jessica.
As
we
said,
the
implicit
upgrade
of
cube
ATM
also
operates
the
cubelet.
This
is
a
bug
cuz.
A
Right
now
we
run
like
one
where
we
want
to
use
cubed
m18
and
deploy
a
one,
seven
cluster,
but
this
this
currently
means
that
we're
getting
one
eight
cubelets
as
well,
which
should
which
shouldn't
be
the
case.
If
we
want
because
then
cubelets
are
higher
than
the
master
api
version,
which
is
not
supported
so
I'll
file
an
issue
on
this
as
well,
but
it
will
have
to
be
done
anyway.
A
A
A
H
H
A
H
A
B
I
need
to
go
back
to
it
like
as
soon
as
I'm
done
with
this
skew
version,
testing
stuff.
So
I
wanted
to
get
like
a
batch
or
round
it
out
of
done,
because
we
planned
a
version
sonobuoy.
It
seems
like
after
the
one
eight
release
and
I
want
to
make
sure
that
it
didn't
explode.
Goob
ATM,
you
know,
but
it
explodes
it.
B
E
B
E
So
as
much
as
we
thought
that
we
had
agreement
on
what
that
would
look
like
it
turns
out
that
maybe
we
didn't
and
that
might
take
a
few
more
rounds
back
and
forth
before
we
can
actually
get
the
daemon
says,
search
upgrades
implemented,
but
we
I
mean
we
should
start
now
because
it
I
take,
might
take
us
a
month.
Yeah
at
least
I
definitely
want
to
give
it
four
one.
Nine
right
so
I
think
we
we
should
not
wait.
E
B
More
than
review
bandwidth
is
consensus.
Right,
like
that's.
The
biggest
problem,
so
I
think
is,
is
long
as
we
just
keep
pounding
the
pavement
as
quickly
as
we
can
in
the
one-line
cycle.
I
think
we
should
be
okay
for
things,
but
I
think
I
think
optimism
be
got
the
best
of
us
on
a
couple
of
fronts.
Last
week
cycle.
Yes,.
E
A
Yeah,
do
we
know
if
Diego
is
like
gonna
own
this
or
like
that
was
part
of
the
promise
for
for
1-8?
Do
we
know
if
he's
still
assigned
to
do
that
in
1-9
like
since
it
didn't
make
one
eight
or
like
it
was
anyway
I
mean
it
was
a
little
too
little
time
left
when
the
PR
was
like
opened
less
than
two
weeks
before
code
freeze,
so
this
is
for.
A
C
I
A
I
The
other
option,
if
let's
say
that
that's
just
not
accepted
the
proposal,
isn't
and
Damon
said
upgrades
in
that
style
aren't
option.
What
we've
been
doing
is
if
you
can
switch
the
deployments,
it's
a
little
bit
gross,
but
you
can
do
you
can
do
required
anti
affinity
and
then
also
do
some
like
label
shuffling
so
that
with
the
selector,
so
that
what
happens
is
if
you're
on
a
single
master,
the
anti
affinity
during
an
upgrade
is
allowed
because
you
technically
have
two
copies
with
different
labels
on
the
same
machine,
but
in
a
multi
master
case.
I
A
I
have
implemented
upgrades
using
self
hosting
using
the
up
kubaton
upgrade
command
using
the
temporary
manifests
kind
of
what
you
said
like.
First,
we
have
the
one
seven
version.
We
duplicate
that
one
make
sure
it
starts
running
and
rename
it
to
something.
Temp
then
upgrade
the
real
many
demon
set
once
it's
upgraded.
The
backup
takes
takes
like
ownership
and
then,
when
we
see
everything's
up
and
running
one
a
this
is
okay
again
we
just
remove
the
one
seven
copy,
and
this
works,
but
I
mean
rollback
is
super
hard.
Here.
I
No
I
mean
yeah
no
I,
just
the
I
think
the
daemon
things
a
upgrade
option
is
probably
the
best,
because
in
that
style
you
don't
get
rolling
updates
either,
which
technically
doesn't
really
matter
with
an
elite
leader
elected
component.
But
yes,
I
think
we
just
try
and
revisit
this
after
the
one
eight
release
and
see
how
far
it
goes.
A
C
A
E
Another
time
there's
a
couple
of
people
from
Google
and
be
interested
in
talking
about
AJ.
That
I,
don't
think
made
it
today,
because
I
tell
them
until
yesterday
that
they
should
come
to
this
meeting
and
I
would
love
if
they
could
read
over
the
current
proposal.
I,
don't
know,
I
know
you
and
Tim.
Both
are
proposals.
I,
don't
know
if
those
are
sort
of
both
still
considered
current
or
if
one
supersedes
the
other
I'd
be
nice.
If
we
had
at
least
a
single
proposal,
we
could
point
people
to
I
think.
B
A
Yeah
so
basically,
my
proposal
was
written
on
top
of
things
in
some
ways:
just
I
just
removed
the
HDD
operator
and
added
some
more
alternatives
and
the
security
aspects
like
ohmic
into
account
other
than
that
it's
pretty
much
insane
anyway.
I
mean
we
can
I,
haven't
updated
this
dark
since
July
and
it
wasn't
I
started
diving
in
to
upgrade
then
so.
I
didn't
even
have
time
to
like
answer
all
the
comments,
but
once
1-8
is
out,
that's
gonna
be
my
top
priority.
A
E
Okay,
that
sounds
perfect.
If
we,
you
say
two
weeks
from
now,
we'll
talk
about
IJ,
let's,
if
we
can
try
to
get
the
docs
sort
of
updated
and
freshened
up,
maybe
on
top
of
Tim's,
we
can
put
a
link
to
Lucas's
and
say
superseded
by
this
other
rock
over
here,
so
make
it
really
obvious
and
then
I
can
point
some
people
add
those
docs
and
we
can
start
talking
about
it
in
a
couple
weeks.
A
Good
to
me,
as
well
cool
due
to
lack
of
a
soon
room
for
the
newly
started
out
of
three
cloud
providers:
things
I'm,
gonna,
reuse,
this
one
for
the
next
call,
which
is
in
four
minutes
again,
I,
it's
if
I
understood
it
correctly.
Iago
was
about
asked
for
I,
don't
know
from
Zara
or
some
other
in
the
community
that
has
access
to
soon
admin,
something
anyway
that
until
that
will
reuse.
This.