►
From YouTube: Kubernetes SIG Storage Meeting 2022-10-20
Description
Kubernetes Storage Special-Interest-Group (SIG) Meeting - 20 October 2022
Meeting Notes/Agenda: https://docs.google.com/document/d/1-8KEG8AjAgKznS9NFm3qWqkGyCHmvU6HVl0sk5hwoAE/edit#heading=h.ro5v0bdbk14h
Find out more about the Storage SIG here: https://github.com/kubernetes/community/tree/master/sig-storage
Moderator: Saad Ali (Google)
A
All
right
today
is
October
20
2022.
This
is
the
meeting
of
the
kubernetes
storage
special
interest
group.
As
a
reminder,
this
meeting
is
public
recorded
and
posted
on
YouTube.
Today
on
the
agenda,
we
are
going
to
go
through
our
126
planning
spreadsheet
and
get
status
update
for
features
that
folks
are
working
on
and
as
a
reminder
for
timeline,
the
enhancement
freeze
has
already
passed.
So
if
you
have
a
feature,
that's
working
that
you're
planning
for
126
that
wasn't
already
declared
and
approved
as
a
cap.
That
deadline
has
passed.
A
The
next
upcoming
deadline
is
the
code
freeze
which
will
be
on
November
9th,
which
is
coming
up
fairly
quickly.
So
if
you
have
a
feature
that
you're
working
on
that
is
the
date
to
be
aware
of,
and
next
up
after
that,
we'll
go
over
any
PRS
to
discuss
designs
to
review
and
miscellaneous
items.
So
if
you
have
anything
that
you
want
to
discuss,
feel
free
to
add
them
here
and
we'll
go
over
them
after
going
over
the
planning
spreadsheet.
A
B
C
A
A
D
Right,
yeah
right
now,
basically
just
to
continue
review.
Well,
actually,
we
still
continue
to
review
the
CSS
flag
and
the
and
the
cap
yeah,
but
I
think
most
likely
it's
not
going
to
make
it
because
this
is
out
of
tree.
We
I
mean
in
theory
we
can
still
continue,
but
I
think
unlikely.
This
will
make
it.
C
C
E
E
A
C
D
A
A
A
D
A
C
D
E
C
A
D
D
Sunday
but
release
notes
is
already
there
right,
so
it's
basically
just
a
just
to
check
if
we
need
but
I
I
believe
even
documentation,
midnight,
okay,
I'll,
double
check.
I,
remember:
Javi
submitted
something
kind
of
updates.
All
the
plugins
but
I'll
ask
a
humble
to
double
check
and
just
steepness
anything.
We
need
to.
A
E
A
E
A
E
F
E
D
F
D
D
No,
that's
that's
for
next
release
right
now.
It's
just
getting
everything
ready
or
are
you
saying,
you're
updating
your
account
yeah
if
your
cab
says
better
than
that
that
cup
cannot
be
merged
yet
I
was
not
sure
about
it.
So
so
for
now
you
can
change
everything
else
by
the
stage
from
alpha
to
Beta.
That
part
has
to
wait
until
1.27.
D
D
E
A
A
D
D
D
A
D
This
is
a
like,
so
that
will
be
like
next
steps
got
it.
So
I'd
like
to
you
know,
discuss
that
so
actually
and
I
actually
have
a
session.
If
you
come
about
the
numbers
for
no
shutdown,
so
we'll
talk
about
next
steps,
so
the
fencing
will
be
one
thing.
So
there
were
a
few.
There
are
a
few
things
right.
There
is
the
there's,
no
fencing
and
story,
fencing
cluster,
fencing
sort
of
few
things.
C
D
D
Okay,
so
two
of
those
are
already
referenced
in
the
cap
in
the
alternative
section.
Why
is
the
safety
patch?
That's
already
that
was
in
earlier
version
of
the
cap?
The
the
second
one
is
the
note
fencing
this
was
a.
There
was
a
there
was
a
repo
by
you
know,
rude
FS.
He
has
that
one
right.
He
actually
has
no
code
there.
So
that's
another
one
and
then
the
the
and
then
the
third
one
is
yours.
D
Then
you
have
a
there's
a
PR
in
CSS
back
right,
that's
the
storage
thing,
and
then
there
was
a
fourth
one
which
is
I.
Don't
know,
if
is
Tom
here
today,
yeah
Tom
from
WMC.
They
actually
have
a
project
understood,
so
they
did
something
they
did
some
kind
of
like
storage,
fencing.
They
did
check
from
Storage
side.
D
D
G
D
A
B
A
C
C
A
A
A
H
Hey
folks
yeah
I'm
Richard
from
AWS
one
of
the
maintainers
of
the
EBS
CSI
driver,
so
yeah
so
I'll,
pretty
much
summed
it
up.
What
we've
heard
from
from
customers
is
like.
They
will
really
like
to
modify
volume
through
the
kubernetes
control,
plane
right
now,
they're
doing
stuff
like
using
the
CLI
or
the
console
to
update
the
volume
stuff
like
volume
type
iops
throughput.
But
there's
like
there's
like
a
bit
of
a
disconnect
between
what
the
volume
actually
is
and
what's
stored
in
the
kubernetes
control
plane.
H
H
So
the
admin
probably
has
like
a
bunch
of
volume,
types
that
they
expose
to
users
and
someone
can
just
choose
a
different
volume
type.
Like
a
storage
class
name
update
the
field
update
the
PVC
and
then
you
know,
that'll
trigger
the
the
modify
volume
endpoint
I
did
see.
Ben
and
sad
did
mention
that.
Maybe
that's
not
the
best
option,
so
yeah
I
just
wanted
to
discuss
with
you
guys
what
you
guys
think,
because
yeah
well.
G
G
G
If
we
wanted
to
have
kubernetes
be
able
to
update
your
Qs
policy,
which
is
a
great
idea.
I
think
we
would.
The
first
step
would
be
to
have
kubernetes
actually
have
a
way
of
finding
out
what
the
current
value
is,
so
that
it
could
compare
it
to
the
desired
value
and
have
a
reconcile
Loop
to
change
it
right.
That
would
be
step.
One.
G
Yeah
yeah
I
mean
something
like
that
or
or
a
new
field.
I
mean
I,
I,
guess
I'm
agnostic
as
to
where
exactly
you
hang
it
off
the
PV
object,
but
you
would
need
some
place
to
say
like
these
are
the
these
are
the
current
parameters
that
are
that
are
mutable
and
these
are
their
current
values
and
then
you
would
have
some
other
object.
That
would
say
what
the
desired
value
is
and
then
you
could
just
write
a
reconcile
Loop.
G
That
would
try
to
try
to
reconcile
it,
but
but
the
key
is
knowing
which
values
are
in
principle,
mutable
and
which
ones
aren't,
because
you
know
there's
there's
a
there's.
Some
of
both
and
storage
class
parameters
cover.
You
know
the
union
of
them
without
knowing
which
is
which
so
like
you'd
need
someone
to
come
and
say
like
this
is
an
immutable
parameter.
This
is
not
a
mutable
parameter.
G
This
is
the
current
value
of
this
mutable
parameter
and
then,
if
you
want
to
change
it,
go
write
a
reconcile
or
look
to
do
that
and
then
there's
the
whole
question
of
security
like
who
is
allowed
to
change
those
things.
You
know
with
Qs
policies,
there's
obvious
sensitivity
to
you
know
an
unprivileged
user
upping
their
own
qos
policy
in
a
way
that
might
damage
other
tenants
using
the
same
storage
system.
F
B
Will
this
require
only
control
plane
or
will
it
require
something?
On
node
side,
the
example
uses,
for
example,
AWS
EBS
and
the
modify
volume
call
Can,
for
example,
be
also
used
to
resize
the
volume
in
case
of
EBS,
for
example,
and
that
kind
of
who
worries
me
like
like
what
kind
of
changes
it
can
make
into
volume
and
if
it
has
an
interaction
with
like
available
disk
size
and
whatnot.
F
G
A
Yeah
so
I
think
tldr
on
this
is
having
a
generic
mechanism
by
which
to
mutate,
storage
class
parameters
is
kind
of
a
no-go
for
us.
The
recommended
path
is
if
there
is
a
specific
field
that
you
are
interested
in
mutating,
say
it's
iops
or
quality
of
service,
or
something
like
that.
First
step
would
be
to
promote
that
into
an
official
first
class
field
in
PVC
similar
to
capacity,
and
then
we
can
talk
about
what
mutation
for
that
looks
like
and
the
benefit
of
doing
that
is.
A
You
can
have,
for
example,
different
types
of
policies
and
how
you
represent
it,
meaning
you
know
things
like
what
Ben
was
saying.
If
you
you
know,
you
don't
want
a
storage
administrator
to
let
application
developer
modify
the
quality
of
service
at
will.
They
should
be
able
to
to
set
that
somehow
or
versus
like
they
don't
care
they
want
to.
Let
anybody
modify
it.
You
know
they
should
be
able
to
set
it
that
sort
of
thing.
H
G
Well,
you're
asking
for
a
very
generic
facility
and
I
think
we're
pointing
out
that,
like
that,
creates
a
lot
of
new
problems,
and
so,
rather
than
do
something
generic,
we
could
do
something
much
more
narrow.
That
would
just
scratch
your
itch
and
be
done
in
a
way
that
doesn't
cause
implementation
issues.
H
H
So
yeah
we've
we've
thought
about.
You
know
what
what
thoughts
I
just
did
was
having
some
fields
that
are
I,
guess
like
a
common
subset
of
among
all
vendors
and
promoting
them
as
like,
first
class
members
in
PVC,
but
that
I
guess
that
gets
us
like
I,
don't
know
like
60
of
the
way
there,
but
not
100
percent.
G
C
H
A
I
I
I
think
just
on
the
email
thread,
there
were
like
two
or
three
different
CSI
providers
who
also
expressed
interest
in
this
capability,
and
it
seems
like
this
would
be
utterly
useful
for
for
providers
to,
especially
in
terms
of
the
kind
of
automation
that
they're
trying
to
build
for
for
migration,
and
things
like
that.
So
I
mean
you
guys
probably
have
a
lot
more
background
than
I.
G
Yeah
I
I
totally
agree
in
my
experience
like
from
the
second
day,
I
was
working
on.
Csi
people
started
asking.
Well,
how
do
you
change
the
storage
class
right
because
they
they
want
to
do
this
and
unfortunately
my
advice
to
them
was
always
well
you
you
do
it
out
of
band
right
kubernetes
doesn't
let
you
do
that,
but
but
I
agree.
If,
if
we
could
find
a
way
to
do
it,
it
would
unlock
a
lot
of
useful
things.
Yes,.
I
Exactly
and
and
I
think
I
think
the
the
point
that
you
make
there
is
a
very
valid
one
like
there
are.
There
are
a
whole
bunch
of
other
use
cases
which
could
potentially
benefit
from
this
capability,
but
people
haven't
yet
really
thought
about
because
the
option
is
not
available.
So
so,
if
we,
if
we
could
come
up
with
something
that
that
works,
I,
think
that
would
that
would
really
help
drive
a
lot
of
interesting
activity
in
this
space.
So.
F
G
I
was
just
going
to
compare
it
to
like
volume
resize
like
initially
kubernetes,
would
not
let
you
resize
your
volumes
and
the
utility
of
being
able
to
resize
the
volume
is
quite
obvious,
so
we
embarked
on
the
volume
expansion
feature
which
has
taken
years
to
to
you
know
to
reach
GA,
and
there
was
a
lot
of
bumps
along
the
way,
but
it
was.
It
was
a
worthwhile
exercise
because
I
think
implementation.
We
have
is
very
good.
G
I
I
hope
that
a
qos
feature
would
not
take
as
long,
but
it
does
face
about
the
same
amount
of
difficulty
in
terms
of
getting
all
the
pieces
right
so
like
I
think
we
should
do
it,
but
we
shouldn't
assume
it's
going
to
be
a
quick
and
easy
thing.
That'll
be
done
in
a
month.
I
think
it'll
require
careful,
careful
thought
and
Design
to
get
right.
C
G
G
I
mean
it's
a
one,
can
imagine
a
qos
policy
that
includes
like
class
of
storage,
SSD
versus
hard
disk
and
then,
as
long
as
you
can
transparently
change
that
under
the
covers,
like
you,
could
just
do
it
right.
It
is
that
that
would
be
a
feature
of
what
the
Qs
policy
can
express
and
I
would
argue
that
that's
the
kind
of
thing
it
should
be
able
to
express
well.
C
G
C
G
So
there's
things
in
storage
classes
like
like
volume,
encryption
right,
you
typically
can't
start
encrypting
a
previously
unencrypted
volume
or
start
unencrypting.
Your
previously
encrypted
volume,
things
like
the
file
system
type
like
once
you
put
ext4
on
a
disk.
You
can't
change
it
to
xfs.
After
there's
a
bunch
of
data
there
there's
a
bunch
of
there's
a
bunch
of
storage
class
parameters
that
sort
of
get
stamped
on
the
volume
at
creation
time,
and
then
it's
understood
that
they're
not
going
to
change
ever
right.
C
G
G
A
Yeah,
the
challenge
is,
you
know,
with
storage
class,
the
parameters
are
completely
opaque
to
us
and
and
the
way
that
the
system
has
been
designed
is
you
know.
Storage
class
is
a
set
of
opaque
parameters
at
provision
time
after
provision
time
it
has
no
meaning
to
kubernetes.
You
know
the.
The
fact
is
that
it
shouldn't
even
be
on
the
pvpvc
object.
C
A
A
C
H
F
C
D
D
D
A
D
D
You
know
the
6
30
session
on
Friday
I
will
be
giving
the
session
with
Mauricio
Mauricio
will
be
getting
a
Windows
update,
I
mean
Windows
Deep
dive
actually,
and
we
will
also
be
going
over
the
six
storage
updates
for
1.25
and
1.26
release
and
also
Wednesday,
and
there
is
a
session
on
data
protection
group,
so
Xiang,
Chen
and
I
will
be
giving
that
session
yeah.
So
if
you
are
at
coupon
and
you're
interested
and
please
join
those
sessions
or
events.
I
I
had
one
quick
question
and
I
apologize
if
this
is
out
of
this
is
not
on
the
agenda,
so
not
sure
how
we
handle
these,
but
we
had
some.
We
had
it.
We
had
some
changes
in
the
mount
utils
repository
that
we
are
looking
to
get
some
reviews
on
haven't
really
had
any
traction
on
it
for
a
while
I
can
send
out
a
link
to
the
pr.
If
that
helps,
and
we
were
hoping
that
somebody
who's
an
official
maintainer
for
that
repository
may
be
able
to
provide
a
review
soon.