►
From YouTube: SIG Apps Bi-Weekly Meeting for 20201020
Description
SIG Apps Bi-Weekly Meeting for 20201020
A
Okay,
welcome
everyone.
This
is
the
gaps
and
it's
monday,
19th
october.
My
name
is
adnan
abdullah
and
I'll.
Be
leaning
us
through
this
meeting
today
I'll
share
the
link
to
the
agenda
in
chat,
so
people
can
follow
along.
A
A
A
The
other
announcement
here
is
that
gcr
will
eventually
be
going
away.
It's
going
to
be
replaced
by
this
new
service
from
google
called
artifact
registry.
I
don't
know
that
this
affects
the
gaps
too
much,
but
there's
an
issue
here
in
case
you're
interested
tracking.
The
move
from
gcr
to
artifact
registry
for
kubernetes
so
feel
free
to
check
that
out.
If
you're
interested.
A
Cool
those
are
all
our
announcements
going
into
discussion.
So
one
thing
that
we've
added
here
is
that
starting
in
120,
beta
apis
only
have
three
releases,
which
is
nine
months
to
graduate
or
increment
beta
apis
in
sig
apps.
We
have
two
apis
that
are
still
in
beta
controls
and
pdbs,
which
I
believe
con
job.
We
have
we're
on
plan
to
move
from
beta
and
pdb.
There's
some
kept
open
here.
B
C
No,
we
will
be
able
to
go
cron
jobs
without
bumping
the
api.
We
are
targeting
120
for
the
alpha
controller
and
with
121,
when
the
controller
will
be
promoted
to
beta,
which
basically
means
it'll,
be
the
default
controller,
with
a
fallback
to
the
old
one.
If
you
desire
at
that
point,
we
will
be
promoting
cronjob
api
to
ga.
C
This
will
allow
us
not
to
introduce
that
additional
step
of
api
for
pdb,
because
I
was
also
talking
with
martin
and
vojtek
and
couple
other
folks.
It
looks
like
the
api
will
not
require
additional
changes
and
we
will
be
able
to
promote
it
within
the
next
release
or
two
so
we're
targeting
for
121
to
do
the
initial
work
that
will
be
required
to
promote
the
api,
so
both
should
be
on
track.
D
Matthew,
so
the
only
thing-
that's
not
clear
to
me
like
you-
can
go
for
for
crime
job
based
on
the
cap.
The
current
status
and
the
plan
moving
forward
is
super
clear
for
pdb.
I
think
we
might
need
to
update
the
cap
to
indicate
for-
or
I
don't
even
know,
if
it's
a
cap
at
this
point,
I
think
it's
actually
just
the
status
item.
D
We
just
probably
need
to
update
it
to
communicate
a
little
better
that
that
plan
is
to
take
it
to
ga
with
the
current
api
and
then
resolve
any
other
issues
or
features
that
we
want
after
the
fact,
because
we
have
a
confidence
that
the
api
won't
change
by
adopting
anything
else,
we
want
to
do
later.
C
C
D
C
Yeah,
that's
that's
perfectly
clear.
We
just
missed
the
deadline
I
was.
I
was
focused
on
the
cron
job
and
the
deadline
was
so
clear
that
before
I
realized,
I
ran
out
of
time.
Yeah
yeah
it
just
yeah.
It
was
too
close
to
get
both
at
the
same
time
but
yeah,
I'm
I'm
in
sync,
with
morton
and
we'll
make
sure
to
to
get
updated
and
in
in
the
form
of
a
cap,
so
that
that's
for.
E
Get
out
for
you.
A
A
A
A
Sounds
good,
then
we
can
move
on
to
a
bug.
Triage
did
someone
wanna
lead
us
through.
F
F
F
D
A
E
D
She's
I
mean
if
she
doesn't
want
it
back
to
me
I'll
talk
to
her
online
she's
in
taiwan
right
in
the
asia
she's
in
the
apac
time
zone.
So
it's
just
kind
of
hard
for
her
to
be
here.
E
F
Well
there
there
is
the
back
of
endemons,
that's
right,
which,
if
you
start
fighting
with
the
note,
there's
a
backup
and
a
delay.
Maybe
it's
related.
D
D
F
C
F
D
F
C
F
C
I
can't
remember
the
exact
code
currently
off
top
of
my
head,
but
maybe
there's
a
bug
in
howie
how
we
decide
when
the
pot
is
running
and
when,
in
a
multi-container
part,
send
them
away
and
tag
it
at
6
cli
as
well.
F
E
C
F
C
F
F
D
D
D
D
C
C
Yeah,
it's
it's
more
of
a
misconfiguration
rather
than
debug
in
the
cli
or
controller
or
anything
else
yeah,
because
the
controller
properly
reacts
creating
only
one
part
and
working
as
expected,
yeah
send
it
my
way
I'll
I'll
have
a
look
and
I'll
leave
a
comment
explaining
what's
the
problem
there.
A
A
Sorry,
what
did
that
makes
you
missed
out
a
sign
when
I
signed.
F
B
D
The
new
pod
revision
created
at
once,
but
for
some
reason,
the
deletion
request
for
the
whole
pod
is
refused
by
validating
web
hook
for
a
long
time.
What's
the
val,
I
don't
know
which
validation
by
book
like
we
don't
ship
any
yeah,
so
you
can
sign
to
me
I'll
ask
him:
what
did
you
put
on?
What's
in
your
cluster.
E
F
D
The
only
thing
a
deletion
should
create,
or
do
we
delete
soft
deletes
for
a
pi.
It
says
the
termination
timestamp
right
so
like
it
is
a
yeah.
I
don't
know
this
doesn't
make
sense.
A
D
Like
we
do
sort
of
have
a
standard
way
to
do
this
right,
like
if
I'm
going
to
build
a
custom
controller,
for
instance-
and
I
use
kuby
builder
or
I
use
operator
kit
or
any
of
those
things
it
does
like.
It
basically
demonstrates
the
exact
workflow
for
configuring,
a
pod
to
have
the
correct,
irvac
and
service
accounts
to
perform
any
requests.
You
want
against
the
api
machinery
and
inject
it
directly
to
your
cluster
right.
B
D
I
mean
like
it's.
What
I'm
wondering
is
if,
if
the
guy
is
asking
for
network
coordinates,
if
he's
like,
I
want
a
way
to
standardize.
The
injection
of
network
coordinates
for
the
api
server
directly
into
a
pod,
so
I
can
consume
it
as
an
environment
variable
and
then,
like
I'll
figure
out
all
of
the
rest
of
the
things
that
I
need
to
talk
to
the
api
server.
I
guess.
G
C
Probably
so,
there's
a
bunch
of
environment
variables
that
we
are
injecting
on
for
every
pod,
that's
being
run
in
a
cluster,
and
that
includes
the
the
name
space
you're
running
in
the
cluster
api
right.
So
that's
it.
D
Just
we
already
inject
the
coordinates,
that's
the
thing
you
can
get
in
kubi
builder
and
operator
kit.
They
build
their
infrastructure
out.
They
leverage
that
the
harder
part
is
setting
up
the
r
back
to
be
able
to
talk
to
the
api
machinery
and
setting
up
the
service
account
and
service
tokens
to
allow
your
pi
to
be
able
to
actually
communicate
with
the
api
community
and
have
the
correct
credentials
to
be
able
to
perform
the
operations.
D
What
that
you
want
to
do,
and
there's
already
a
lot
of
automation
for
building
that
out
on
two
different
projects
that
are
leveraging
the
same
infrastructure,
that
being
controller
runtime
and
controller
tools
under
the
hood
so
like
it
seems
sort
of
solved.
Maybe
he
just
needs
information.
B
That
might
be
my
guess,
because
a
lot
of
what
happens
in
kubernetes
folks
you
get
in
these
calls
and
things
like
that.
No
and
he
may
just
not,
and
a
lot
of
this
may
not
have
intuitive
documentation,
and
it
may
just
be
pointing
about
saying:
hey:
have
you
does
this
work?
You
know,
here's
how
you
can
get
to
it
and
point
him
to
something
that
may
just
solve
his
problem.
F
F
E
D
Unless
you
are
in
the
same
things
anyway
yeah,
but
you
can't
hear
that
doesn't
work
like
yeah
assumptions
about
we
try
specifically
as
not
to
make
assumptions
about
the
underlying
network.
Right
like
there
are
very
many
different
underlying
software-defined
network
configurations
that
you
can
oh
yeah
work,
so
we
can't
like
the
api
machinery
doesn't
know
like
you,
you
have
to
kind
of
configure
that
and
then
run
your
cluster.
D
B
I
was
just
gonna
say
and
even
if
you
want
something
like
this
exposed,
it's
not
a
sig
apps
problem,
it
would
be
a
sig
network
issue
and
then
they
can
reject
exposing
that
information
because
of
those
issues
too.
But
it
sounds
like
he
wants
access
to
sig
network
things
related
things,
not
so
much
at
the
application,
workload
level
and
or
maybe
it's
a
cross.
But
it
sounds
like
it's
it's.
He
wants
those
lower
level
details
and
we're
not
going
to
give
it
to
him,
which
makes
sense.
C
I
mean
basically
he's
touching
the
deployment
strategy
that
he
is
using
for
his
own
deployment
of
the
entire
cluster
and
that
will
vary
from
provider
to
provider
from
one
user
building
their
own
cluster
to
a
different
user
and
both
data
you're
gonna
inject
and
how
you're
gonna
inject
or
consume.
Those
will
differ
so
that
I
think
there's
a
reason
that
we
stayed
away
from
this.
F
Well,
I
think,
and
it
it
may
not
only
be
possible
to
inject
it
like
he
wants
this
data
for
cubelet,
but
cubelet
is
the
one
injecting
anything
into
ports
right.
So
you
come
back
to
only
using
localhost
or
a
old
balancer,
but
I
don't
see
really
how
we
could
do
it,
and
I
agree
with
matt
that
sig
ups
is
probably
not
the
seek
to
tackle
this.
F
C
Well,
that's
surprising,
that's
the
most
basic
feature,
although
something
tells
me
there's
knitting
greater
details
about
how
completions
work,
when
you
don't
specify
them
in
combination
with
parallelism,
eric
eric
documented
that
actually
pretty
well
I'll.
Send
it
my
way
I'll
have
a
look
and
I'll
I'll,
eventually
point
him
to
the
docks
or
try
to
figure
out
what's
wrong.
There.
C
Yeah
but
I'm
pretty,
I'm
pretty
sure
that
he
just
doesn't
know
that
they're
there's
an
option
to
say
conclusions
and
that
it
will
run
indefinitely
until
we
say
to
stop.
F
C
E
C
You,
but
the
new
controller
is
pretty
it's
pretty
slick.
F
F
C
C
C
The
garbage
collection
can
be
stuck
because
you
put
a
third-party
finalizer
on
a
pod
for
whatever.
E
C
Which
basically
means
that
the
garbage
collection
will
not
remove
that
particular
pot
until
all
the
final
finalizers
will
be
removed.
This
is
to
ensure
that
if
you
need
to
need
to
perform
any
any
manual
cleaning
that
the
pod
will
not
be
removed
or
basically
any
resource
will
not
be
removed
until
all
the
finalizers
are
cleared.
C
Multiple
options,
I
would
need
to
see
the
parts
back
I'll
I'll,
read
through
it
carefully
and
see
what
it
might
be,
causing
this
particular
case.
F
All
right
next
ones
are
demonstrating,
update,
taking
longer
I've
just
paired
it,
and
it
looks
like
he
wants
to
download
the
image
before
starting
terminating
the
demo
set,
which
we
won't
do
right,
because
you'd
have
to
create
a
bot
actually,
and
you
can
just
stop
it
before
it
starts.
But
clayton
had
the
proposal
for
them
on
sets
to
make
the
search.
So
unless
everyone.
B
D
B
I
can
know
you
know
what
okay
here
go
ahead,
ken
or
you
can
do
it.
B
F
F
D
I
mean,
but
it
wouldn't
be
a
partition
right,
like
you,
can
partition
staple
set
because
you
can
think
of
them
as
like
an
ordered,
sequential
set
of
things.
Deployment's,
not
not
like
that.
There's!
No
there
it's
groups
of
fungible
replicas
right,
that
is,
that
are
statistically
likely
to
have
a
certain
number
of
replicas,
depending
upon
like
what
you
set.
Your
your
surge
and
your
admin
available
to
during
an
update.
D
I
I
feel
like
they're
they're,
if
you
really
want
to
do
canary
analysis
nowadays,
there
are
better
ways
to
do
it,
like
it
kind
of
depends
on
what
your
cluster
looks
like
like.
If
I
was
using
istio,
I
would
do
canaries
one
way
if
I
was
not
using
istio,
and
I
was
trying
to
do
capacity
based
like
just
set
up
one
deployment
with,
let's
say
10
of
my
capacity
and
one
with
90.
I
would
do
it
a
different
way,
but
there
are
different
ways
to
do
a
canary,
so
pause
to
me
is
more
like.
D
I
want
to
stop
it
in
the
middle,
because
I
don't
like
the
way
it's
looking,
but
I
don't
want
to
roll
it
back,
and
the
short
answer,
though,
is
no
they're
asking.
If
they
want
to
prepare
a
pr.
I
would
point
them
to
you
can
assign
to
me.
I
would
say:
please
submit
a
cap
like
if,
if
they
got
something
that
looks
good,
I
wouldn't
say
like
we
shouldn't
take
it,
I'm
just
saying
like
before
you
go
and
implement
it.
D
It's
not
like
a
quick
patch
like
it's
an
api
change
at
a
minimum
and
probably
very
complicated,
so
I'd
want
to
see
a
cap
for
it.
F
C
Yeah
just
send
in
my
way
I'll
have
a
look
if
it's
a
feature,
I'll
I'll
reply
that
sure,
but
you
can
you
have
to
work
it
out
from
scratch.
C
F
F
F
F
D
B
D
D
Yeah,
as
far
as
I
know
like,
if
you
set
your
m
in
a
particular
way
like
when
you
print
your
m
you're
gonna
get
it
in
the
order
that
you
specified
it
up
there.
That
could
be
that
could
have
changed,
but
it
it
is
fundamentally,
like
you
said
signo
and
it's
not
saying
actually,
there's
not
much.
We
can
do
the
implementation
here
is
going
to
be.
B
C
The
yama
conversation
was
recently
brought
up
and
when
we
were
trying
to
bump
young
version
for
cobra,
there
was
an
issue
with
line
length.
Yes
and.
C
G
F
F
E
F
Then
there
is
the
case
where,
if,
if
you
actually
specify
an
invalid
image
to
a
stateful
set
and
that
starts
rolling
out,
you
can't
go
back
because
statefulset
actually
waits
for
unavailable
bots.
First,
I
was
already
working
on
a
fix,
so,
okay,
I
I
will.
I
will
take
it
and
there
is
probably
a
dupe.
C
C
F
F
D
So
we
implemented
originally,
it
didn't
respect
the
skills
of
resource
at
all
and
then,
like
maybe
a
year
year
and
a
half
ago,
we
implemented
support
from
the
skill
set,
but
I
believe
that
inside
of
the
code-
even
if
you
don't
obviously
also
resource.
A
E
A
Cool
yep:
that's
all
we
had
on
the
agenda
and
we're
out
for
time.
So
unless
anyone
has
any
anything
that
they'd
really
like
to
discuss,
we
can
meet
in
two
weeks,
good
cool,
all
right
thanks.
Everyone
for
joining
thanks,
bye.