►
From YouTube: Kubernetes SIG Storage - Bi-Weekly Meeting 2022-07-14
Description
Kubernetes Storage Special-Interest-Group (SIG) Bi-Weekly Meeting - 14 July 20122
Meeting Notes/Agenda: -
Find out more about the Storage SIG here: https://github.com/kubernetes/community/tree/master/sig-storage
Moderator: Xing Yang (VMware)
A
A
A
A
So
I
think
yesterday
iman
said
he's
going
to
have
someone
work
on
this
one.
That's
that's
what
I
heard
last
time
so
I'll
just
say.
A
B
There
has
been
some
discussion
on
in
the
pr
I
think
it's
still
not
merged,
but
it's
getting
close,
that's
my
impression.
C
Yeah,
sorry,
I
was
thinking
csi
inline,
yes,
so
I
so
I've
not
made
much
more
progress
since
last
time,
but
I
still
think
we're
on
track
for
this
release.
I
just
have
some
manual
testing
to
do
and
then
I
need
to
promote
some
of
the
e-2e
tests,
the
conformance
tests,
so
I
need
to
get
that
started.
So
I
can
get
the
pr
out
for
review.
C
A
Oh,
I
actually
should
actually
look
at
our
timeline.
It's
actually
a
cold
freeze
is
coming
august,
2nd.
So
it's
actually
less
than
two
weeks.
A
A
A
Next
one
is
a
local
ethmoid,
storage
resource
management,
so
jean,
I
think
jean
cannot
join
today.
She
did
add
a
issue.
She
did
add
something
in
this
in
the
stock,
so
we
can
talk
about
that.
A
little
later.
A
What
in
group
api,
so
I
think
we
talked
about
this
last
time-
I
think
we
still
have
not
decided
on
whether
we
need
to
have
like
two
object
or
just
one
object
is
enough.
Yes,
I
think
we
still
need
to
need
to
spend
some
time
looking
into
this
one.
So
still.
A
Now
michelle
do
you
have
a
strong
opinion,
but
whether
we
need
to
have
the
steel
object
to
support
the
you
know
the
time
kind
of
static
provisioning
that
we
always
do
with
this,
like
ppvc
model
or
with
the
wording
group,
do
we
do
you
have
a
strong
opinion
on
this
one
that
we
have
to
have
the
same
specify
this?
What
a
group
handle
has
to
be
in
this.
D
At
least
for
me,
I
think
it's
gonna
really
depend
on
the
use
cases
right.
I
think
that's
what
we
need
to
evaluate.
E
A
A
Yeah,
so
I
think
what
I'm
going
to
do
is
probably
maybe
I
because
I
know
there
are
some
implementation
of
this
already,
so
we
also
started
to
do
a
poc,
but
I
think
there's
already
some
indentation
out
there,
maybe
I'll
just
see
what
they
did.
A
So
those
are
just
to
see
what
other,
if
people
really
require
this
specific
use
case.
A
A
Okay,
so
I
think
I
think
ryan
is
working
on
e3
tests,
but
I
have
not
seen
a
pr
yet
so
still
he
tweeted
her
spending.
A
But
I
did
submit
a
pr
and
to
promote
the
csi
stack
ga,
because
because
there
are
a
few
owning
house,
related
csi
spec,
like
capabilities
and
rpcs
that
are
experimental
right
now
are
in
alpha.
So
I
submitted
a
pr
to
promote
that.
A
And
next
one
cozy
do
we
have
sit
here.
A
So
so,
for
this
one,
since
the
cups
merged
there
are
a
few
new
contributors
joined
the
meeting,
the
cozy
meeting,
which
actually
it's
going
to
happen
right
after
this
meeting.
So
it's
thursday
every
thursday
at
this
time,
so
right
now,
basically
folks
are
trying
to
update
the
the
apis,
the
closest
back
and
the
controller
based
on
the
emergency
cap,
so
just
say:
well,.
F
Yeah
I
talked
to
sid
about
this
a
few
days
ago
and
he
said
yeah
now
that
the
new
apis
are
merged.
It's
a
matter
of
bringing
all
the
sidecars
and
controllers
up
to
date
with
the
spec
and
they
have
some
pr's,
but
it's
still
a
work
in
progress.
F
Also,
some
pdrs
are
out.
Some
pr's
are
not
even
ready
for
review
and
I
think
there
will
be
maybe
half
a
dozen.
I
don't
know,
I'm
not
sure
how
many
total.
A
F
A
There's
also
ci
right,
so
we
need
to
cut
a
release.
I
think
we
need
to
bring
that
up
to
speed.
I
think
there
is
some
basic
there,
but
it
should
be
updated.
So.
A
Okay,
so
next
one
is
the
node
expansion
secret.
I
think
this
one
is
the
implantation
is
done.
Basically,
just
I
think
the
dot
pr
is
not
out
yet
so,
basically,.
A
Actually,
there
is
a
pr
long
time
ago,
because
we
didn't
make
it
last
time
so
just
need
to
update
the
stock.
B
Well,
I'm
waiting
for
api
review
for
quite
some
time
michelle.
If
you
are
here,
it
could
help
with
that
and
giving
the
deadline
like
early
august,
I'm
thinking
it's
going
to
sleep
with
an
extra
list
again
really.
B
G
The
think
that
with
the
api
review
is
done,
the
rest
of
the
appears
like
we
are
behind
a
feature
gate,
so
I
mean
yeah.
D
H
Did
we
have
any
open
bugs
on
this
because
so
we're
we
should
be
all
set
in
terms
of
like
the
cap
and
that
kind
of
thing
so.
A
It's
basically
the
is
the
the
feature
gate
I
mean
there
should
be
a
code.
Pr
right
is
that
one
merged.
I
do
not
remember
yeah.
D
A
A
Awesome,
so
it's
probably
just
to
do
is
just
a
dog.
I
think,
and
also
can
jelly
write
a
blog
about
this
again.
D
A
Can
you
can
you
help
ping
him
about
this
yeah
yeah
I'll
pick
that
thank
you
all
right,
that's
good
and
okay.
So
next
one
c
is
a
migration
v
sphere,
so
the
so
the
main
pr
to
flip
the
feature
gate
is
also
merged.
I
think
come
on.
Have
some
comments
on
the
mainly
the
e3
test
right.
So
I
think
the
team,
the
vsphere
team,
is
working
on
thus
providing
the
tests.
A
D
A
I
think
so
so
we
have
some
eq
tests
ourselves
in
our
sisa
java
repo.
I
think
those
are
all
tested,
so
I
think
I
believe
they
may
have
already
posted
results
there,
but
I
think
how
many
you
you
point
out
to
this
and
the
entry
test
so
they're
going
to
look
into
that
next
and
then
there's
also
some
etw
test
entry,
which
is
under
the
vc
repo.
That
was
also
that
was
around
last
time.
Let
me,
I
think,
first
bring
the
feature
to
beta
so
well.
G
Yeah,
I
guess
the
those
tests
like
the
entry
tests
passing
with
the
migration
apple
is
kind
of.
Like
I
mean
yeah,
those
are
like
the
overarching
requirement
for
the
driver.
Migration.
A
G
G
A
Because
of
the
there
is
a
couple
of
issues
with
the
csr
proxy,
I
think
we
opened
the
issues
there
of
god
what
was
performance
or
something
I
forgot.
What
was
the
issue?
They
were
like.
I
remember
the
two
things
right,
so
those
are,
but
those
are
not
this
specific
issue,
basically
at
the
least
to
those
open
issues,
but
because
of
that,
I
think
we
didn't
make
it
ga.
G
So,
okay,
I
think
we
also
need
to
previously
we
deprecated
like
older
than
six
seven
u3,
seven,
zero
u2,
all
this
version.
Now
we
need
to
in
this
release.
We
need
to
just
we
have
to
cover
that.
Okay,
the
support
is
completely
removed
from
for
those
versions
of
uv
sphere,
and
that
needs
to
go
like
loud
and
clear.
I'll
I'll
make
a
note
in
the
in
the
pi
that.
A
Is
already
it
is
already
in
the
document
and
is
already
in
the
release.
Note
last
time
right,
I
think
it
said
we
added.
We
divided
last
time
in
last
release.
1.24.
A
We
said
that
you
should
we
in
the
in
the
in
the
kubernetes
document.
We
already
said
that
you
need
to
update
to
seven.
Oh
you
two,
that's
that's
the
dot
in
the
kubernetes
dock
website,
kubernetes
website.
Basically,.
A
Okay,
you
can
take
a
look
yeah,
you
can.
If
you
have
more
yeah,
definitely
you
can
add
it
to
the
to
there
and
then
yeah
and
also.
I
think
we
also
need
to
update
our
dock.
We
have
our
own
driver
dock
so
that
one
needs
to
be.
We
will
update
that
one,
basically,
that
one
still
has
a
67
there,
so
we'll
remove
that
one
as
well
from
that
that's
the
official
dock,
I
mean
that's
our
official
dock.
A
G
G
A
Can
you
can
you
maybe
okay?
So
maybe
you
can
add
that
in
that
in
the
pr,
so
we
can
make
sure
that
exactly
what
what
is
the
requirement
there.
I
A
H
Yeah,
this
should
be
all
set
again.
I
don't
know
if
we
need
any
doc
or
blog
thing
or
if
that's
subsumed
by
the
you
know
the
overall
stuff
you're
always
working
on.
I
I
think
actually,
I
think
it
may
need
a
the
feature
gate
code.
Actually,
oh.
H
Yeah
yeah
yeah
yeah,
I'm
sorry,
I've
been
out
on
vacation
the
last
week
or
so
and
still
getting
caught
up.
A
And
then
I
think
after
I
think
this
one
definitely
needs
a
dog
change,
but
I
think
for
the
blog
we
can
just
have
that
comment.
Blog.
A
H
A
Is
that
is
that
enough?
They
can
have
a
section
talking
about,
like
particular
drivers.
Each
driver
can
have
a
session
if,
if
they
want.
A
A
This
is
sad
michelle
you
remember
about
this
one.
This
is,
we
know
the
cap
is
merged,
but
is
there
a
pr
merged
already
or
this
is
aws
csm
migration?
I
do
not
remember.
D
I
have
not
seen
anything:
oh
okay,.
A
A
And
it's
about
next
year
seth-
I
is
humble
here
today
so
so
I
I
pinned
him.
He
did
to
get
back
to
me
today,
but
he
said
the
basis
f
rbd.
He
need
to
work
on
the
tests
need
to
show
the
need
to
publish
the
test
results
so.
A
A
And
the
next
two,
I
think
we
are
just
tracking
the
e2
test
for
the
next
two,
because
they're
saying
after
in
this
release,
always
owner
reclaim
policy
see
deepak
is
probably
not
on
call
today,
so
he
has
not
got
a
chance
to
work
on
the
csf,
but
he
said
he
will
work
on
this.
One.
A
Yes,
and
then
this
next
one
is
ronald,
is
looking
into
this
one,
basically
the
we.
We
talked
about
this
in
the
csi
implementation
meeting
for
convert
volume,
mode
control,
volume
mode
conversion,
so
we
will
add
those
tests
in
the
external
snapshot
of
ripple.
A
But
for
that,
since
this
is
the
first
time
we're
doing
this,
there's
a
lot
of
work
involved
just
to
getting
that
set
up.
So
he
said,
as
I
said,
it
will
take
more
time
obviously,
but
he
will
be
working
on
that.
A
J
H
Yes,
so
the
the
next
steps
are,
the
alpha
beta,
you
know,
feature
gate,
code,
change
and
then
docs
need
to
be
updated.
A
So
now
does
this
already
have
eq
tests.
H
A
H
Yeah
the
the
change
to
mark
the
feature,
gig
beta,
okay,.
A
J
A
G
Yeah,
I
think
we
we
just
waiting
like
we
are
waiting
for
the
design
and
like
follow-up
meetings
on
this
one,
we
have
not
made.
There's
been
no
updates
since
last.
Actually
so,
okay.
A
Next
one
is
the
battery
for
sweaty
class.
The
moment.
B
E
E
F
F
That
would
give
kubernetes.
The
ability
to
you
know
distinguish
between
different
types
of
volumes
generated
within
a
csi
driver,
but
there's
still
some
snags.
It
doesn't
solve
all
the
use
cases,
so
we're
trying
to
see
if
we
can
sort
those
out
and
maybe
make
a
proposal
that
would
be
acceptable.
F
If
we
can't
we
don't
have
any
better
ideas.
So
that's
where
that
stands.
A
Okay,
thank
you
ben
okay.
So
that's
all
we
have
on
this
brush
here
now
go
back
here,
we'll
look
at.
A
This
one
has:
this
is
deep
eddie.
This
one
is
deep
here:
hey
deep.
E
Hey
yeah,
this
is
a
this
is
a
quick.
I
guess
reminder
for
a
pr
submitted
by
ebo.
This
was
around
doing
like
basically,
I
think
the
sca
the
scheduling
sig
they
introduced.
E
This
new
way
of
filtering
results
called
pre-filter
result,
and
so
basically,
this
is
an
enhancement
in
the
volume
binding
plug-in
to
use
the
pre-filter
result
to
kind
of
better
basically
shrink
down
the
number
of
nodes
that
should
be
configured
for
scheduling,
and
this
has
a
sort
of
a
beneficial
effect
for
local
pvs,
because
they're
just
bound
to
one
node,
so
it
doesn't
make
sense
to
evaluate
all
the
nodes
and
the
new
prefer.
The
result
mechanism
allows
you
to
do
this.
E
I
think
we
discussed
this
once
before
and
I
think
maybe
jan
had
some
context,
but
basically
this
is
like
ready
for
another
review.
A
Okay
has,
but
I
don't
see
any
review
comments
from
seed
storage.
It
looks
like
it's
all
from
honoring,
some
okay,
so
we
have
a
cold
feet,
but
mostly
it's
from
the
scheduling
side
right
so
yeah
so
yeah
and
okay.
So
you
look
at
them
yeah
michelle!
Maybe
you
do
you
want
to
take
a
look
of
this.
A
B
Oh
well,
I
just
wanted
to
say,
like
I
don't
know
much
about
scheduler
and
okay
and
the
topology
is
screwing.
A
A
A
There
is
a
concern
from
this
this
other
than.
Basically
he
was
saying
there
are
on
some
systems
like
kind.
A
You
can't
really
use
this
feature,
so
she
was
saying
that
because
it
can't
detect
the
route
file
system,
disk,
u
usage,
so
they
actually
disable
this
feature
with
the
feature
gate.
So
now
that's
the
problem:
if
we
bring
this
to
ga,
then
they
don't
have
the
safety
gauge
to
disable
anymore.
A
So
ching
is
thinking
that
she
can
maybe
add
a
flag.
It's
like
a
cubelet
configuration
flag
called
enable
local
3d
capacity
isolation.
A
A
No,
this
sounds
reasonable,
but
I
think
definitely
we
need
to
go
to
signal
right,
because
this
is
a
cube,
lit
flag
need
to
ask
them,
but,
like
we
don't
have
any
concerns
from
our
side.
It
sounds
okay
to
me.
D
D
A
So
this
one,
I
think
it's
a
yes,
because
it's
they
will
have
to
improve
those
changes
using
kubelet
configuration
all
right.
Thank
you.
So
that's
that's
the
one
and
then
second
one
is
this.
I
think
someone
submitted
this
pr.
A
So,
okay,
so
I
think
it's
probably
it's
easy
just
to
read
this
one
here.
Should
I
do
this
on
here.
So
basically,
what
happens
is
that
windows
containers
are
created
with
the
default
volume
size
of
20
gb,
so
I
think,
does
it
make
sense?
So,
basically,
add
a
new
field
to
windows.
Container
resource
cri
object,
specify
the
size
of
your
fs
scratch
space
used
for
windows
containers,
so
there's
a
pr
submitted
here
for
for
this.
A
Does
it
make
sense
for
for
this
change
and
then
there's
the
thing?
There's
a
comment
here:
oh
did
the
city
preach
attitude.
Was
that
different
one
or
oh
okay,.
A
All
right,
okay,
so
I
think
gene
actually
added
some
some
comments
here:
okay,
yeah,
okay,
so
this
is
just
let's
just
look
at
this
one.
A
A
I
think
the
this
person
was
suggesting
to
you
to
use
this
fm
storage
request
limit
for
for
this,
for
this
windows
size
the
windows
steeper
size.
A
Now
I
think
gene
has
some
concerns
to
overload
that
feature,
and
basically
it's
good
for
someone
who
has
you
know
who
is
familiar
with
windows
to
take
a
look
at
this
get
some
feedback.
So
I
don't
know
if
deep,
if
you
have
a
chance
to
look
at
this
issue,.
E
And
talk
about
this
yeah
thanks,
so
this
is
something
while
so
so
this
came
up
while
we
were
experimenting
with
the
csi
provisioning
capability,
that's
now
available
for
basically
local
based
pvs,
the
the
distributed
provisioner
capability.
So
what
we
were
trying
to
do
was
you
know
basically
have
a
csi
plugin
that
creates
local
pvs
instead
of
using
the
static
local
provision
that
we
used
to
have-
and
that's
still,
you
know
supported
completely.
E
So
what
we
found
is
that,
like
one
of
the
typical
usage
patterns
and
something
that
many
of
the
operational
tools
on
local
pvs
depend
on,
is
this
host
name
name
label?
That's
applied
to
the
pve
right
after
creation
by
the
static
local
provisioner
and
when
going
to
the
csi
path,
there's
no
way
for
the
csi
plug-in
to
tell
the
provisioner
that,
instead
of
the
stuff
in
the
volume
context,
that's
applied
to
a
certain
field
within
the
pv
object,
also
apply
like
a
special
thing,
which
would
be
a
label
instead
of
a
volume
attribute.
E
I
was
just
curious
if,
if
such
a
use
case
has,
you
know,
risen
up
in
other
scenarios,
where
there
would
be
a
preference
to
add
something
as
a
label
instead
of
something
in
the
volume
context,
and
if,
if
this
is
something
folks,
things
folks
think
might
be
useful
overall
to
kind
of
have
a
csi
plug-in
be
able
to
specify
to
the
provisioner
that
when
you
create
the
pv,
I
want
like
a
certain
set
of
key
value
pairs
to
be
applied
as
labels
instead
of
contexts.
L
I
think
we
looked
at
something
similar
in
the
past
and
the
challenge
was:
do
these
labels
get
applied
only
once
at
creation
time
or
are
they?
You
know
something
that
is
updated.
So,
for
example,
if
you
have
like
the
host
name
of
where
the
disk
is
attached.
A
L
Would
need
to
be
updated
every
time
the
the
disk
moved
something
like
that.
Ideally,
it
would
be
the
latter
and
that
becomes
more
difficult
and
more
challenging,
but
seems
useful.
E
Okay
cool,
so
if,
if
this
is
useful,
then
I'll
probably
try
to
come
up
with
some
kind
of
a
csi
spec
enhancement
and
follow
up
there.
D
I
think
I
think
also
when
we
originally
implemented
the
topology
feature.
This
was
also
something
we
considered,
I
think,
to
just
automatically
like
apply
the
topology
and
the
volume
as
labels.
I
think
the
the
main
problem
was
that
topology
could
be
represent.
The
topology
representation
could
be
more
complicated
than
just
a
single
key
value
label.
So
I
think
that's
why
we
kind
of
left
it
off
originally,
but
it
is
sort
of
a
it
is
sort
of
a
regression
compared
to
the
entry
case
right.
E
G
G
Previous
for
at
least,
I
don't
know
if
how
much
same
thing
is
useful
in
other
like
for
for
your
case,
but
first
like
entry,
local
pvs,
it's
it's
a
useful
capability,
but
in
industry
we
are
creating
pvs
like
inline.
Are
we
calling
our
own
functions?
So
it's
not
as
important,
but
for
csi
yeah
through
external
provisions
will
be
useful.
A
So
deep
surprise,
so
you
are
planning
to
basically
propose
a
change
in
css
back,
and
this
is
like
when
we
create
a
volume
again
create
volume.
This
is
that
this
is
yeah.
E
Basically,
like
in
the
volume
response
like
have
a
way
for
the
csi
plugin,
to
maybe
specify
that
you
know
beyond
context
like
have
like
another
set
of
key
value
pairs
that
indicate
like
it'd,
be
treated
separately.