►
From YouTube: Kubernetes SIG Apps 20230306
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
Good
morning,
good
evening,
good
afternoon,
depending
on
where
you
are
today
is
March
6th,
and
this
is
another
of
our
bi-weekly
say
gaps
call.
My
name
is
Machi
and
I'll,
be
your
host.
Today,
one
quick
announcement.
We
are
roughly
a
slightly
over
a
week
from
the
127
code
freeze,
which
is
happening
march
to
March
14th
over
to
15,
depending
on
where
you
are
on
earth.
A
There
is
a
particular
dates
and
hours
when
the
code
freeze
happens,
outlined
in
the
link
that
I
put
in
the
agenda,
but
in
generally,
towards
the
end
of
the
day
on
the
March.
14Th
is
a
good
way
of
thinking
about
it
and
I.
Think
I
was
the.
B
A
A
So
that
was
all
when
it
comes
to
the
announcements.
The
main
two
topic
of
discussion,
the
first
one
was
already
handled
by
by
by
Ken
and
I
on
Friday
afternoon,
when
we
were
thinking
prior
to
this
call,
and
that
leaves
us
only
with
the
topic
from
Lucy's
sweet
about
scaling
up
and
down
deployments.
So,
let's
hear
from
let's
see
about
the
use
cases
and
the
idea
behind
it,
do
you
have
a
presentation
that
you
want
to
share
with
or
oh.
C
A
C
Yeah,
just
look
at
the
issue:
that'll
work,
okay,
so
here
I'll
give
you
the
quick
rundown
of
where
this
has
come
from.
I
work
at
Uber,
ubis,
kubernetes,
I,
know
big,
surprise
there
and
we
have
a
load
of
workloads
running
on
K8
right
now
and
while
many
of
them
are
designed
to
withstand
disruption
from
a
load
of
instances
being
dropped
all
at
once
or
a
load
of
instances
appearing
all
at
once.
Some
of
them
are
is
that
anti-pattern
yeah,
probably,
but
it's
an
anti-pattern
the
week.
C
What
I
was
looking
at
doing
is
allowing
end
users
to
basically
control
the
speed
at
which
some
so
right.
Now,
when
you
create
a
deployment
and
you
change
the
replicas
number
in
the
deployment,
that's
immediately
passed
the
replica
to
the
replica
set
and
that
immediately
just
instantly
scales
up
or
scales
down,
depending
on
what
you
change
the
number
to
there's.
No,
what's
the
word,
there's
no
there's
no
slow
roller
or
anything,
it's
just
instant
in
Behavior.
C
So,
in
parallel
effectively,
what
I
was
thinking
of
doing
is
making
it
so
that
if
you
opt
into
this
Behavior,
so
it
doesn't
break
people
who
already
ex
rely
on
the
parallel
behavior
that
you
can
scale
up
and
down
at
a
speed
that
you
control
I.
Guess,
like
my,
the
two
things
going
on
in
my
head
is
first
off.
C
Do
we
even
want
this,
because
that's
kind
of
important,
obviously-
and
the
second
thing
going
on
my
head-
is
because
there's
quite
a
few
approaches
we
could
take
to
actually
do
this
of
yet
how?
How
would
you
do
this?
But
I
guess
the
first
thing
before
I
say
anymore
is
yeah
first
off,
do
we
even
want
this?
Is
this
something
that
yeah
should
even
exist
in
the
API,
so
I'll
stop
there
for
now
laughs.
A
A
Deployment
and
grouplook
has
said
that
actually
prevents
from
scale
up
or
scale
down,
be
super
excessive
and
actually
I
was
checking
the
code
earlier
before
the
call.
What
we
do
when
we're
scaling
up.
We
are
slowly
speeding
up
the
scale
up
operation,
so
you
start
with
a
one
and
then
everything
goes
well.
We
double
the
scale
operation
with
every
every
next
iteration,
so
it
doesn't
scale
immediately
to
whatever
number
you
come
up
with,
but
it
will
be
one
two,
four
and
and
so
forth.
A
So
I
wonder
when
you
were
testing
it,
what
kind
of
scale
up
and
down
you
were
talking
about,
whether
the
the
numbers
were
significantly
big
that
this
was
I.
Don't
know,
debossing
your
cluster,
that
you
brought
this
this
topic
up
and
in
a
similar
fashion.
There
is
indeed
a
unfinished
work
that
I
was
actually
looking
and
I
just
spotted,
because
the
enhancement
was
closed.
Previously
we
had
the
notion
that
a
random
part
where
were
being
chosen
during
downscaling,
but
that
has
changed
with
this
enhancements,
which
is
currently
beta.
A
If
I
remember
correctly,
it
wasn't
promoted.
We
should
probably
pick
it
up
and
and
move
it
across
the
Finishing
Line,
which
would
allow
users
to
slightly
affect
the
scaled
down
operation.
So
maybe
before
answering
those
what
kind
of
problems
you
run
into
with
the
current
mechanism
that
we
have
that
you
are
proposing
this
enhancement.
C
Okay,
so
I
guess
very
quickly.
Yes,
so
when
I
was
testing
it
on
a
cluster
of
my
own,
it
rolled
out
in
parallel
I'm
wondering
I,
guess
my
initial
thought
would
be.
Does
that
wrote
Progressive
rollout
happen
if
you
change
the
Pod
generation,
maybe
but
I
don't
know
yeah
the
actual
problems
we
were
dealing
with
is
what's
the
word.
C
Let's
see
if
I
can
skirt
my
NDA
the
correct
side
it's
is
pop
is
basically
there
are
some
servers
Services
inside
Uber
that
service
owners
have
written
where
the
pods
communicate
with
each
other
I've
seen,
for
example,
people
before
where
they
like
run
FCD
on
the
service
itself
and
then,
when
you
suddenly
drop
a
load
of
them
or
introduce
a
load
of
them.
The
whole
thing
blows
up
in
their
face.
Is
that
an
anti-fattern
yeah
sure
it's
definitely
a
challenge
to
deal
with,
but
that's
that's.
C
To
give
an
example,
it's
mostly
basically
where
pods
talk
to
each
other
is
where
it
starts
to
become
painful
but
yeah,
and
when
I
tested
the
behavior
in
minicube
on
what
was
it
one
point:
I
can't
remember
the
the
latest
version
that
mini
mini
cubes
are
supporting
right
now.
It
it
instantly
took
them
up
in
parallel,
like
with
no
check
with
no
waiting
at
all
really.
A
Ron
I
would
have
to
dug
up
from
what
the
API
is
called,
because
I
can't
remember,
but
basically
there
are
options
which
allow
you
to
decide
how
fast
the
rolling
update
is
happening.
Have
you
tried
changing
those
values?
The
only
downside
is
that
they
are
actually
affected
during
a
rollout
and
not
during
a
scale
operation,
which
is
probably
one
of
the
issues
that
you
might
be
running.
C
Yeah,
so
to
be
clear,
yeah
this
happens
when
I'm
talking
about
here
isn't
a
rollout.
It's
just
increasing
the
replicas
number
up
and
down,
so
it
doesn't
create
a
new
pod
generation
and
it
doesn't
go
through
Max
surge,
Max
unavailable
or
anything
like
that.
It's
yeah
it's
completely
separate
from,
for
example,
changing
something
about
the
Pod
spec
and
then
the
deployment
goes.
Oh,
the
prospect
change,
I
need
to
roll
out.
If
that
makes
sense,
I
mean
one
of
the
proposals
in
this
issue
actually
was
to
yeah.
C
Have
the
option
for
users
to
extend
how
rollouts
work
to
the
scale
up
and
down
stuff,
but
anyway,
sorry
I'm
going
off
track.
A
B
A
So
in
general,
we
try
not
to
add
too
many
new
features
into
the
existing
controllers
and
would
probably
advise
folks
to
give
it
a
try
and
try
to
write
their
own
version
of
something
like
that
and
then
getting
back
to
us
with
with
some
kind
of
a
proof
of
concept
which
we
couldn't
better
understand
the
use
case
and
how
it
works
before
actually
deciding
on
accepting
such
a
feature.
C
Yeah
I
can
have
a
look
at
that
I'm,
quite
literally,
about
to
go
on
PTO
and
when
I'm
in
C
and
going
to
Seattle
and
when
I'm
in
Seattle
I'm
not
playing
through
any
K8
stuff
but
yeah.
When
I'm
back
I
don't
mind
having
a
look
at
that
I'm.
C
It's
yeah
internally,
that's
basically
how
we
solved
it
already
is
with
a
controller
on
top,
but
it's
what's.
The
word
would
be
nice
to
actually
integrate
that
into
the
main
line,
rather
than
just
hold
on
to
it
ourselves.
In
some
proprietary
code
base
somewhere
uh-huh.
A
Yeah,
so
if
you
could
probably
let's
scrape
that
out
a
bit
and
and
demo
a
set
of
proof
of
concept,
how
it's
working,
how
it's
solving
the
issue
you're
having
and
then
based
on
that,
we
could
make
a
little
bit
more
informed
decision.
A
Calls
I
I,
think
that
would
be
a
regional
approach
so
that
we
can
better
understand
how
you're
using
it
and
what
you
are
trying
to
achieve.
C
C
The
issue
we
have
is
these
Services
aren't
meant
to
be
stable
in
the
first
place
and
whenever
and
the
fact
that
they're
being
hosted
on
the
Clusters
they're
being
hosted
on
is
more
of
the
fact
that
the
service
owners
are
pretending,
they're,
stateless
when
they're
not
yeah,
but
that
just
gets
into
organizational
pain.
B
A
Any
questions
for
Lucy
about
this
particular
case.
A
Okay,
does
anyone
else
have
any
other
topics
that
I
want
to
the
group.
A
Hearing
none
with
that
I'm
gonna
close
the
call
ahead
of
time.
Thank
you
very
much
all
for
today
and
see
you
next
time.
In
two
weeks.