►
From YouTube: Kubernetes SIG Storage - Bi-Weekly Meeting 2022-06-16
Description
Kubernetes Storage Special-Interest-Group (SIG) Bi-Weekly Meeting - 16 June 2022
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
B
Yeah
so
fab,
you
didn't
have
unnecessary
time
frame
cycles
to
do
this
in
125
time
frame,
but
I
have
started
a
csi
spec
change
proposal.
I
have
a
pr
open
to
to
move
the
feature
from
alpha
to
ga,
so
I
think
once
that
is
done,
it
will
be
pretty
simple
to
just
flip
the
feature
kit,
but
so
I
think
we'll
target
this
in
126
at
this
point,
but
we'll
just
try
to
get
the
csi
spec
change
done.
This
quarter.
B
Yeah,
so
the
I
originally
wanted
to
do
this,
but,
as
you
know,
they
have,
the
requirement
has
been
changed
to
have
a
e2e4
alpha
to
beta,
and
I
don't
have
any
e2e
for
this
one,
and
I
will
need
a
review
like
bunch
of
release
of
some
side
cars
to
be
able
to
test
it
so
that
I
can
simulate
a
test
failure
like
the
guidance
we
got
from
sig
release
and
like
lava
lamp
and
other
people
who
do
the
work
on
this
just
to
use
this
time
to
add
al
lucky
to
e,
rather
than
try
to
move
it
to
beta.
A
Right
right,
okay,
so
all
right,
so
pretty
good
yeah.
So
we
should
just
trap
it
right.
Otherwise,
yeah
we
still
don't
have
time
in
next
release
right.
So
all
right!
Okay,.
A
A
C
Yeah,
so
there
are
two
pr's
like
the
item:
four
four
and
a
five
where
there
are
two
pr's
related
escape
months,
point
checking
during
cleaning
up
mount
point
and
the
other
is
improved
like
a
tracking
one
point
with
a
new
like
a
library
the
implement
they
added
the
api
there
like
open
ads
too
and
both
pr
like.
I
have
a
few
more
comments
and
I
think
they
mostly
addressed
it.
It's
basically
yeah
almost
ready.
There
are
also
other
reviewers
involved,
so
I'm
basically
waiting
for
the
final
approval.
A
B
D
B
Left
by
it,
so
that
code
actually
like
it
first
checks
in
the
actual
state
of
the
world,
whether
it
is
possibly
mounted,
and
then
it
checks
if
the
the,
if
the
volume
has
left
any
mount
point
so
like
that,
seems
problematic
to
me.
That
checks
explicitly
for
mount
points
because
for
drivers
that
don't
create
mount
points.
C
Right
so
yeah
this
is
a
issue
is
not
addressed
by
these
two
pr.
Okay,
I
think,
after
this
two
pr
I'll
go
over
the
logic
again,
because
they
are
also
changing
some
of
the
logic
related
when
doing
the
cleanup
one
point,
and
then
we
can
work
on
that
substitution
issue
after
yeah,
these
two
merged.
B
And
one
more
issue
I
found
out
was
was
what
was
it
sorry.
A
E
B
So
I
have
to
yet
I
think
I
should
file
issues
and
one
more
check
was
like
when
we
are
reconstructing
the
volume
during
the
during
the
reconstruction.
I
I
don't
know
if
anybody
else
saw
this,
but
the
problem
there
is
a
problem
and
the
problem
is
that
there's
a
code
that
checks
if
the
volume
exists
volume
exists
in
operation
and
it
checks.
If
the
it's
like
it's
it
has
a
code
like
reconstruction
fails.
B
B
D
B
F
Because
I'm
trying
to
think
of
an
example
of
how
you
could
have
a
a
non-raw
block
volume
without
a
mount,
can
you
can
you
give
an
example
of
how
that
could
happen?.
B
Like
certain
csr
driver
would
just
probably
create
like
place
like
like,
for
example,
entry
host
host
path
driver
doesn't
create
mind
points.
The
csi
host
path
does
create
because
in
the
node
publish
it
it
creates
a
bind
mount,
but
I
could
see
cases
where,
like
do
they
have
to
like.
If
the
css
spec
doesn't
say
that
the
note
published
path
has
to
be
a
mount
point
right.
F
F
B
You're
right,
I'm
just
yeah,
I'm
saying
that
I
was
thinking
that
it
could
like.
It
may
not
be
a
mount
point,
because
obviously
the
csi
spec
is
operating
system
agnostic,
so
it
it
may
like
my
reading
of
it
doesn't
say
that
it
has
to
be
one
point,
but
it
just
says
it
has
to
be
accessible.
So
it
will
be
the
the
path
at
which
the
volume
will
be
published
and
one
can.
C
C
Right
so
we
can
talk
more
details
offline
so
but
basically
the
long
point
check
is
for
like
safety
right.
We
don't
want
to
delete
the
parts,
results,
cleaning
up
the
mountains
and
yes
because
they
have
corruption,
so
yeah.
B
C
Yeah
so
like
imagine
so
there,
those
three
hours
change
some
logic
in
the
clean
at
one
point
and
so
and
then
after
that,
we'll
have
more
like
checking
on
the
the
the
reconstruction
like
that
part.
C
B
Maybe
two
separate
issues:
one
about
yeah,
the
bar
deletion
and
one
about
reconstruction.
G
H
And
start
if
you
are
here,
you
are
assigned
as
approver
of
dead
cap.
So
if
you
could,
please
approve
it.
C
Yes,
so
the
cap
and
the
pr
file
is
submitted
and
yeah
waiting
for
review.
A
Oh
okay,
yeah
I'll,
take
a
look.
A
A
All
right
thanks
and
next
one
warning
group,
so
that
so
this
I
basically
I
have
that
cap
updated,
I
think,
need
to
be
reviewed.
But
probably
I
think
we
talked
about
this
one.
A
little
bit
in
the
csi
implementation
meeting
looks
like
patrick,
actually
has
a
proposal
that
has
that
has
a
different
design,
which
is
not
using
the
two,
not
using
two
objects
like
the
pdtvc
type
of
model,
but
it's
using
one
object
for
dynamically
provisional
sources.
A
So
I
need
to
take
a
look
and
see
if,
if
we
can
use
that,
because
right
now
in
this
updated
cap,
I
have
like
one-
google
and
volume
content.
It's
basically
two
two
resource
similar
to
the
pdpvc
and
the
one
snapshot,
one
snapshot
content
model.
So
I'll
take
a
look
at
that.
A
Okay,
so
next
I
think
the
next
two
markers
is
done.
So
I
think
I
don't
have
to
talk
about
those
next
one
is
pvc
volume
snapshot,
namespace
transfer,
hey
masaki,.
I
Yes,
we
closed
this
transfer
cap
and
oh.
I
I
Yeah,
I'm
waiting
for
approval.
Thank
you
for
your
review.
A
Did
you
get
it?
Did
the
team
get
a
chance
to
review
that.
B
A
I
think
he's
familiar
with
that.
Maybe
just
talking
that
you're
using
that
you
like
to
see
what
he
think
of
that
is
that
that
that
itself,
it's
still
offer
right.
That's
the
one
one
uncertainty
right:
we
are
using
something
that
is
still
alpha,
yeah
all
right.
Thank
you,
and
next
one
see
the
warning
house
right,
so
we
are
so
we're
targeting
beta
yeah.
Actually,
I
think,
homogeneous
one
I
forgot
to
mention
earlier.
A
Basically
there
is
this
new
requirement
that
we,
if
we
want
to
move
a
cap
to
beta,
then
we
need
to
have
to
show
merge
e3
tests
in
that
cap
for
the
capture
merge.
So
now
this
was
not
what
we
expected
before
now
we
enjoy
that
the
test
will
come
later.
So
so
we
are
trying
to
add
a
testing,
so
we
have
one
week
see
if
we
can
get
tests
added
and
merged
before
the
cap
merged
that
line
yeah.
So
that's
the
status
of
this.
A
In
the
next
one
voting
populator
so
ben,
are
we
doing
anything
for
this?
One.
F
J
A
All
right,
thank
you,
oh,
and
also
actually
the
this
one
right,
one
that
this
one,
the
provision,
warnings
on
course,
namespace.
That's
one
potential
adoption
of
this
next,
one
cozy,
so
actually
the
the
cap
for
cozy
just
got
merged.
So
that's
actually
really
good
news.
A
A
A
Yeah,
that
may
be
a
good
idea.
We
could
schedule
another
one.
D
A
Good
yeah,
the
only
thing
yeah,
I
think
that
yeah,
I
think
it's
very
too,
but
it's
too
tight.
This
is
a
big
big
cap.
That's
just
one
week
but
yeah.
I
should
have
a
pin
due
boxer.
Yes
before
yesterday's
meeting
so
yeah.
So
let
me
let
me
check
with
them
just
to
see
if
we
want
to
schedule
a
one-off
meeting
just
for
this
one.
Otherwise
we
probably
can
use
the
csr
meeting
or
next
to
the
production
of
the
meeting.
J
Mounting
hey
so
we
had
a
meeting
last
week
with
ben
yan
and
michelle
and
got
some
feedback
so
initially
the
kept
in
its
current
state.
It
kind
of
tries
to
focus
on
a
model
where
every
csi
plugin
knows
how
to
interact
with
every
special
micro,
vm
runtime,
which
was
not
desirable.
J
So
the
feedback
was
mainly
to
come
up
with
a
more
generic,
potentially
a
proxy
based
mechanism.
So
this
quarter
the
plan
would
be
to
kind
of
explore
that
prototype
it
with
at
least
cara
and
then
integrate
that
into
the
cap
as
well
as
work
on
the
csi
site
changes.
A
E
Sure
is
there
a
meeting
note.
A
Let's
try
to
see
the
part
c
for
windows
transition
to
prevent
these
containers.
This
is
a
design.
Is
my
ratio
here
or
a
gene
deep?
What's
the
status
of
this?
It's
a
sesame.
C
He
has
not
yet
got
time
to
work
on
that,
but
we
might
consider
to
find
someone
like
the
intern
to
to
do
this.
But
if
anyone
like
from
community
want
to
help,
that
would
be
also
great.
A
Okay,
all
right,
so
if
there's
anyone
who
are
interested
in
one
canvas,
please
like
jane
d
for
my
region
now.
A
Yeah
yeah,
sometimes
he
occasionally
do
it,
so
I
just
want
to
check
yeah,
so
this
one
is
actually
already
merged
and
then
he
also
just
updated
the
cap
with
the
this
new
test
plan
section.
So
I
think
this
should
be.
This
is
actually
done.
I
think
the
only
thing
would
be
documentation
so.
A
K
I
need
to
double
check,
but
he
did
update
the
cap
and
I
think
that
merged,
but
yeah
I'll
double
check
the
status
of
it.
Okay,.
A
And
v
sphere
yeah,
so
we
feared
the
dvr
submitted
pr
and
updated
the
there's
this
new
test
section
that
is
merged,
review
command.
You
have
a
comment.
I
think
he
submitted
another
pr
for
that.
B
A
Yeah
so
yeah
can
you
just?
Can
you
review
his
the
cap.
A
B
A
A
B
Yeah
the
concern
he
had
was
like
when
we
drop
alpha
notation
from
the
css
from
the
jrpc
protocol.
Does
it
cause
any
backward
compatibility
issues?
I
think
it
does
not,
but
I
have
not
verified
it.
I
think
we
did
something
similar
for
the
capacity
tracking
alpha
fields
and
nobody
complained,
so
it
should
be
okay,
oh.
D
C
K
A
A
And
next
one
is
the
sev
rbd,
basically
moving
to
beta
off
by
default,
so
I
think
humboldt-
probably
still
not
here
yet
so
he
so
the
cap
is
merged.
So
he
added
the
test
section.
There.
A
And
system
migration
forward
works
also
moving
to
beta
off
by
default,
so
this
one
is
the
same.
The
cap
is
also
matched
basically
all
updated
with
that
new
test
flat
pi
section.
A
So
this
one,
basically,
there
was
a
there's,
a
there's,
a
question
about
this:
the
cluster
entry
plugin
slack
by
dims.
A
Right
so
there's
this
this
hecate,
so
with
this
library
that
it
it
is
used
by
glasgfs,
but
this
one
has
some
some
problems
and
he
submitted
pr
trying
to
fix
that,
but
they're
just
not
not
not
merged.
Yet
it's
been
there
for
very
long
time.
So
so
humble
is
saying
that
he's
checking
with
this.
B
K
If
we,
if
the
gloucester
community
you
know,
is
not
does
not
want
to
update
their
dependencies-
and
you
know
we
need
to-
I
think
we
need
to
clean
up
our
dependencies
is
the
first
thing,
but
if
they're
not
willing
to
sort
of
keep
it
updated,
then
yeah.
I
think
that's
a
sign
that
we
need
to
deprecate
this.
A
D
Yeah,
I
think
we
just
need
to
send
out
the
right
emails
and
communications.
J
D
D
Is
there
anybody
on
the
call
that
cares
about
this
gloss,
rfs
or
weather.
A
All
right
thanks
next
one,
always
owner
reclaim
policy.
So
this
one
we
we
are
basically
moving
this
one
to
1.26,
so
we're
actually
not
doing
not
not
moving
to
beta
in
this
release
and
control
volume
mode
conversion.
This
one
is
the
same
thing:
we're
going
to
wait
until
1.26
to
move
to
beta
and
next
one
is
a
secret
protection.
I
Yep,
it's
the
design
for
this
release;
okay
and
yeah
for
in-use
protection.
It
is
tracked
for
this
release
to
vr
firm.
A
A
Okay,
thanks
and
non-greece
from
shadow,
so
yeah,
so
the
sound
is
the
same
thing,
so
we
are
trying
to
move
to
beta,
but
then
this
because
of
this
requirement
for
the
e3
tests.
So
we
are,
we
are
trying
we're
working
on
e2
tests,
I'm
trying
to
get
it
in
by
the
cab
merge
that
myself.
A
So
are
we
doing?
Are
we
doing
anything
for
this
one,
it's
matt
here
michelle
do
you
know
what
are
we
we're
staying
off?
Our
aren't
you
doing
anything
for
this.
B
A
This
is
the
the
cap.
Merge
is
actually
next
next
week
and
also
it's
already
past
the
the
pr
deadline
right.
So,
if
he's
not
targeting,
then
it's
probably
too
late
already
and
also
this
is
a
target
team.
If
it's
moving
to
a
beta
now,
we
also
have
this
new
new
requirement
for
the
e3
test,
I'm
not
sure
if
he
has
usually
has
in
place
yeah,
so,
okay
and
the
next
one
is
a
warning
extension
for
a
staples
that
I
have
not
heard
of
financially
about
this
one.
So
I
don't
know.
B
A
A
The
last
time
I
heard
was,
you
said
basically
said
you
chatted,
but
there
are
a
few
issues:
kind
of
kind
of
hard
to
resolve
or
unresolved
issues,
still
basically
right.
Yeah.
A
And
then
the
the
so
the
person
who
submitted
the
poc
code
did
this
person
try
to
address
those
under
observed
issue.
B
No,
I
think
no,
I'm
not
sure
I
don't
think
so,
but
I
think
I
haven't
like
he
has
addressed
answered
it,
but
there
are
some
issues
that
I
think
even
not
captured
in
the
cab
comment.
Section:
okay,.
A
E
A
Okay,
and
also
the
next
one,
is
the
better
defaults
with
the
class.
A
F
Yes,
yes,
I
do
so.
We've
had
two
meetings
on
this
subject
so
far.
Wait
am
I
off
mute,
yeah,
okay
and
we're
slowly
gravitating
towards
the
idea
of
some
kind
of
sub
volume
type
in
addition
to
the
csi
driver
type,
that
would
be
persisted
on
the
co
side,
in
addition
to
the
csi
driver
type
and
then
act
as
a
key
to
all
kinds
of
other
features
that
that
seems
to
be
the
the
least
painful
way
to
actually
solve
these
problems.
It's
it's
still
fairly
unappealing.
F
It
would
require
communication
between
cubelet
and
the
scheduler
to
take
account
of
this.
You
know
for
things
like
the
node
volume
limit
and
it
would.
It
would
mean
a
lot
of
necessary
changes
on
the
kubernetes
side
to
take
advantage
of
that
particular
enhancement.
But
it's
the
only
way
that
we
could,
in
a
reasonable
time
frame,
actually
address
the
underlying
problem,
which
is
csi.
Drivers
have
real
difference
volume
types
within
them
that
just
it's
just
a
reality.
F
That's
not
captured
by
the
current
abstraction,
and
so
we
can
either
choose
to
continue
living
in
that
world
and
suffer
the
pain
or
you
know
explicitly
track
the
some
concept
of
a
volume
subtype
and
just
have
that
carried
around
with
the
volume
for
its
lifetime.
So
so
so
the
you
know
we're
having
the
meetings
every
tuesday.
That's
that's
where
the
discussion
stands
today,
if
other
people
have
other
ideas,
please
come
to.
F
A
A
You're
going
to
write
down
some
details
on
the
option,
two
or
three
right
that
we
talked
about
in
last
meeting
and
then
we
can
go
over
that
in
next
meeting.
F
Yeah,
so
I
I
wrote
down
notes
on
tuesday
after
the
meeting
that
captured
the
discussion
and
the
plan
going
into
the
meeting
next
week
would
be
to
have
a
some
kind
of
a
straw,
man
and
I'll
have
to
come
up
with
that,
but
but
yeah
the
the
combination
of
options,
two
and
three
that
that
we
seem
to
be
steering
towards
is,
is
some
formal
concept
of
a
volume
subtype
that
would
exist
at
the
csi
layer.
The
sp
would
generate
it.
F
F
It
would
help
with
just
about
every
use
case
I
can
think
of,
except
for
potentially
the
available
capacity
ones,
because
when
you
know
the
available
capacity
ones,
you
don't
have
a
volume.
Yet
so
you
can't
talk
about
the
subtype,
but
but
there's
I
think
we
already
have
a
fairly
good
implementation
for
it.
For
that
problem,
we
can
talk
about
it
as
well,
so
yeah,
that's
where
that's
where
things
stand.
Please
come
to
the
tuesday
meeting.
If
you
want
to
contribute
to
that.
B
I
just
want
to
like
say
that,
like
at
least
for
me,
I
have
not
mentally
ruled
out
option
a,
but
we
can
talk
that
in
there.
Well,.
F
I
mean
option
a
is
still
an
option,
but
I
think
we
agreed
that
it's
very
expensive
up
front
and
you
would
seen
absolutely
no
progress
for
approximately
a
year
after
doing
a
bunch
of
work
and
even
then
it
might
still
not
solve
the
problem,
because
vendors
might
just
choose
not
to
do
it.
So
it's
we
can
consider
it.
But
you'd
have
to
convince
the
rest
of
us
that,
like
it's
worth
doing
all
that
work
for
potentially
no
payoff.
B
No,
I
mean
okay,
we
can
go
in
the
the
details,
but
the
basically
like
make
it
easier
to
run
different
volunteers.
Yeah.
F
D
B
F
F
Anyways
in
particular
find
a
way
for
node
plugins
not
to
consume
a
lot
of
memory
on
the
node,
if
they're,
not
in
use
so
like
some,
some
mechanism
for
cubelet
to
spin
up
and
spin
down,
node
plugins
as
needed
would
be
a
big
memory
saver
in
in
situations
where
you
had
a
large
number
of
node
plug-ins,
which
you
would
definitely
have
if
you
went
down
option
number
one
and
things
like
economizing
on
side,
cars
and
making
it
easier
to
you
know,
deploy
multiple
csi
drivers
without
a
bajillion
side
cars,
that's
a
valuable
whether
or
not
we
do
option
one.
F
The
migration
yeah,
I'm
curious
to
know
what
your
thoughts
are
on
that,
because
the
ability
to
we
have
heard
of
vendors.
Coming
in
saying,
like
hey,
I
want
to
change
my
csi
driver
name
because
my
company
got
acquired
or
something
like
that,
and
it's
like
well
that's
something
that
we
should
probably
have
a
way
to
do,
but
we
don't
so
that
yeah
that
would
fall
under
the
category
of
migration.
F
But
but
I
guess
my
point
is
we
could
pursue
all
of
those
and
and
we'd
spend
a
year
doing
so
and
then
people
could
still
say.
Well,
I'm
not
going
to
redesign
my
csi
driver
to
do
that
to
solve
the
subtype
problem,
and
then
you
still
haven't
solved
the
problem.
So
we
should
do
those
things.
But
if,
if
we're
interested
in
solving
the
the
problems
related
to
volume
subtypes,
I
think
we
need
to
directly
solve
that
problem.
A
Okay,
thanks
ben,
so
we
can
talk
more
in
next
tuesday's
meeting.
A
Okay,
so
that's
all
on
that
spreadsheet
and
we
have
a
few
things.
Let's
see,
let's
just
I
copied
this
from
last
meeting,
so
I'm
not
sure
if
the
person
wrote
this
down
is
in
this
meeting.
A
Is
the
so
whoever
wrote
this
down
to
you
here,
if
not
then,
or
just
to
come
to
the
next
next
meeting.
A
And
next
one
is
there's
a
note
from
gary
siegmann.
It's
gary
here.
A
Because
that's
not
also,
I
don't
know
if
you
just
want
to
let
people
know
about
this
or
if
you
want
to
oh
okay,
so
it
looks
like
this
is
a
this
is
actually
today.
Okay,
so
I
think
we
said
as
a
co-present.
Co-Presenter
of
this
talk
he's
a
goldman
sachs
engineer
at
a
workshop
on
june
16th,
which
is
today
so
there's
a
link
here.
A
If
you
are
interested,
it's
about
goldman
sachs
using
all
restore
fs,
csi
and
openshift
as
our
to
deploy
software
to
a
hundred
thousand
machines
globally,
okay
yeah,
so
I'm
not
going
to
say
move
to
next
meeting,
because
this
is
actually
happening
today.
So
if
you
are
interested,
you
can
join
this
talk,
you
can
listen
to
this
talk.