►
From YouTube: 2022-02-17 Kubernetes SIG Scalability Meeting
Description
Agenda and meeting notes - https://docs.google.com/document/d/1hEpf25qifVWztaeZPFmjNiJvPo-5JX1z0LSvvVY5G2g/edit?ts=5d1e2a5b
A
A
Yeah
yeah
so
yeah.
I
would
like
to
understand
a
little
bit
better.
What
do
you
have
in
mind
because
it
depends
so,
for
example,
if
if
someone
is
listing
from
watch
cache,
then
basically
we
don't
have
currently
option
to
to
just
send
part
of
the
list
and
basically
we
always
send
the
whole
whole
result.
A
Where,
instead
of
like
listing
something
you
can,
we
will
have
a
streaming
api
for
for
getting
the
list
of
things.
So
this
will
be
actually
much
better
solution
for
the
api
server,
because
then
you
don't
need
to
provide
any
any
limit
and
also
it
will
be
less
invasive
for
for
api
server,
because
currently,
when
you
try
to
list
some
huge
huge
objects,
huge
amount
of
objects,
then
it's
possible
that
the
api
server
will
run
out
of
memory.
So
actually
we
had
discussion
about
it
some
time
ago.
I
think
it
was
yeah.
A
A
No,
actually
it's
it's
mostly.
I
think
it's
mostly
api
server,
and
but
the
difference
is
that
okay,
so
and
now,
for
example,
what
what
what's
happening
is
that,
like,
let's
say
that
you
want
to
list
everything
and
it
comes
from
watch
cache
and
when
it
comes
from
watch
cache.
What
happens
is
that
it
copies
the
whole
list
of
objects
and
it
keeps
this
copy
till
till
the
receiver
receives
it
receives
it.
A
So
so
that's
what
was
very,
not
well
optimized
for
memory
of
api
server
and
basically,
instead
of
it,
we
will
have
streaming
api,
which
will
have
kind
of
constant
memory,
consumption
on
api
server,
and
if
you
are
making,
if
you
are
listing
something
from
the
api
server
and
you
put
resource
version
0,
then
basically
it
comes
from
the
api
server
and
you
don't
even
need
to
ask
it.
Ask
a
cd
for
for
the
result.
B
B
So
our
proposal
here
was
to
never
send
unpaginated
requests
to
cd,
so
one
option
we
explored
was
to
have
this
pagination
in
hcd
client
and
have
the
api
server
say
that
I
want
to
honor
that
option,
but
ciao
found
a
problem
with
that
and
it
because
there
are
salt
options
that
sometimes
are
passed
in.
So
these
two
together
may
not
work.
B
A
Okay,
honestly,
I'm
not
very
familiar
with
this
part,
so
I'm
I'm
a
bit
yeah.
It
would
be
nice
if
wojtek
was
here
he's
way
more
familiar
here.
So
I
think
what
we
could
do
is.
I
can
follow
up
with
wojtek,
because
I
I
don't
feel
competent
enough
to
answer
this
question.
What
kind
of
issues
paginating
those
list
calls
to
at
cdd
would
have.
B
Yeah
sounds
good,
so
what
what
we
could
do
is
we
can
study
the
streaming
that
you
mentioned
the
details
of
that
9th
of
december,
we'll
check
that
out
and
we
should
have
a
follow-up
discussion
on
if
that
is
good
enough
for
us
or
do
we
need
something
else,
because
we
don't
necessarily
have
problem
only
with
the
api
servers
workings,
but
the
workloads
as
well.
Sometimes
the
workloads
drive
it
like
that,
so
that
can
also
be
an
issue.
A
B
A
C
Probably
one
thing
I
want
to
add
is
for
the
still
same
topic,
but
I
have
seen
some
streaming
proposals
yeah
previously,
so
it's
but
it's
for
informers
and
maybe
another
option
is
to
enable
it
for
all
the
requests
not
just
for
informers.
I
remember
in
that
proposal.
There
are
some
options
you
need
to
specify
like
the
most
recent
revisions
and
using
some
watch
instead
of
list.