►
From YouTube: Kubernetes SIG Storage 20200910
Description
Kubernetes Storage Special-Interest-Group (SIG) Meeting - 10 September 2020
Meeting Notes/Agenda: https://docs.google.com/document/d/1-8KEG8AjAgKznS9NFm3qWqkGyCHmvU6HVl0sk5hwoAE/edit#heading=h.77yv0dly2k90
Find out more about the Storage SIG here: https://github.com/kubernetes/community/tree/master/sig-storage
Moderator: Saad Ali (Google)
A
A
A
Code
freeze
is
going
to
be
november
12th
and
then
dox
freeze
is
going
to
be
shortly
after
that
november
30th
and
we're
targeting
a
120
release
on
december
8th.
So
please
keep
those
dates
in
mind
if
you
are
working
on
a
feature,
so
let's
jump
right
into
it
and
start
getting
status.
Updates.
First
item
we
have
is
from
hamanth
a
csi
online
offline,
resizing
volume,
expansion
fix
issues.
B
All
right,
so
we
had
a
call
with
them.
We
are
basically
blocked
on,
like
we
have
fixed
most
of
the
bugs.
We
have
two
or
three
outstanding
tickets.
One
is
that
read,
write
many
and
which,
for
which
there
is
beer
out
it
needs
refactoring.
Then
there
was
a
fix
for
volume
pvc
getting
deleted
while
resizing
spending.
Kk
has
like
volunteered
to
fix
that
one
then
recover
from
resize
fieldwork.
B
We
had
a
call
about
that,
one
with
kim
and
jordan,
and
we
agreed
to
look
into
the
part,
sizing
enhancement
and
follow
it
closely
and
also
team
was
just
tim
suggested
that
we
should
consider
giving
quota
back
for
terminal
failures
on
pvc
expansion
rather
than
the
control
says
that
the
pvc
quota.
B
B
And
I'm
working
on
updating
the
design
of
the
recovery
from
recess
failure
proposal
and
I'm
planning
to
schedule
a
call
tomorrow
to
discuss
the
what
our
next
steps
will
be.
D
Yeah
so
so
I'm
trying
to
cut
a
release,
so
we
got
the
the
validation
web
hook
and
the
moving
api
to
a
separate
gold
package.
Those
all
got
merged,
there's
one
pr
outstanding,
which
is
to
change
the
api
extensions
from
linux.
D
Right
now
couldn't
get
hold
of
the
pr
author.
So
maybe
I'll
just
cherry
pick
his
change
and
trying
to
get
that
in
and
then
there
are
a
couple
of
small
issues
that
we
need
to
fix
before
cutting
release.
So
it
should,
it
should
be
soon.
Cool
sounds
good.
B
I
don't
have
any
new
update,
but
I
have
a
proposal
done
by
next
week.
A
Man
next
up
is
sc
linux,
recursive
permission
handling.
C
I
may
not
have
enough
time
this
quarter,
so
what
I
would
will
try.
I
will
try
to
reduce
the
design,
but
I'm
not
sure
I
can
do
the
implementation.
Okay,.
A
A
A
Is
this
an
item?
Anybody
on
the
call
is
interested
in
helping
with
so
this
would
require
you
to
understand
what
the
existing
design
is
understand.
What
the
shortcomings
of
the
existing
design
are
and
then
propose
a
new
design
for
file
permissions
if
you've
ever
found
that
the
existing
design
is
painful,
this
may
be
a
good
opportunity
to
contribute
anyone
on
the
call.
F
F
A
Yeah,
the
answer
may
be
that
it's
not
or
that
we're
gonna
have
to
do
some
fun
dances
to
make
sure
backwards.
Compatibility
works
properly,
maybe
introduce
a
you
know,
a
whole
new
set
of
fields
that
you
can
use,
but
those
new
fields
may
be
exclusive
to
the
old
set
and
may
not
be
used
at
the
same
time.
Something
like
that.
G
F
G
I
think
also
this
will
depend
a
lot
around
the
investigation
for
windows.
I
think
that's
where
this
idea
came
in,
because
the
current
apis
are
very
linux
focused,
but
now
that
we
have
windows
that
might
throw
a
wrench
into
it,.
F
Yeah,
I
I
personally
don't
know
enough
about
what
happens
down
at
the
the
docker
layer
in
the
kernel
layer,
with
the
mapping
between
file
system,
uids
and
gids
to
process
uids
and
gids,
and
I
think
that
that
there
might
be
a
missing
facility
somewhere
or
I
don't
know
something-
that's
badly
designed
but
yeah.
I
just
it's
just
a
like
a
spidey
sense.
I
don't.
I
can't
point
to
it.
A
Yeah,
that
makes
sense,
so
if
anybody
on
the
call
is
interested
but
doesn't
want
to
speak
up
right
now
feel
free
to
reach
out
to
reach
out
to
me
michelle
schenger,
any
other
leads
and
we'll
be
happy
to
put
you
in
touch
or
reach
out
to
ben
even
and
we'll
get
started
on
this,
especially
if
you
work
closely
with
windows
or
or
with
the
with
microsoft.
This
might
be
an
important
project
for
worth.
Taking
a
look
at
related
is
file.
Permission
handling
for
windows
looks
like
we're.
A
H
H
A
Sounds
good.
Thank
you.
Kk,
for
the
update
next
item
is
csi
entry
read-only
handling.
I
think
this
was
a
to
do
to
check
if
humble
would
be
able
to
take
on
this
task.
A
Humble
are
you
on
the
line
by
any
chance?
Let's
take
a
look.
G
I
think
I
saw
humble
send
out
a
pr
to
external
provisioner
around
this
okay,
but
I
think
we
need
some
more
discussion
on
that
pr.
B
Yeah,
I
think
ben
and
we
are
having
a
discussion,
but
it
might
require
taking
a
step.
A
A
A
So
we're
gonna,
hopefully
set
up
a
meeting
for
that.
Can
someone
reach
out
to
humboldt
to
ask
him
to
set
up
a
meeting.
A
A
Okay,
next
item
here,
issues
related
to
assuming
volumes
are
mount
points.
Issue
number
seven,
two,
three
four
seven
andy
agreed
to
work
on
this
for
the
quarter.
The
last
status
update
was
pr
is
ongoing,
not
sure,
if
going
to
fix
everything
anything
new
here,
michelle.
G
G
A
A
All
right
next
item
is
storage
capacity
trend.
Excuse
me,
storage
capacity
tracking
by
patrick.
A
A
A
Yon,
do
you
know
what
the
status
of
these
two
are.
A
Next
item
is
spreading
over
failure
domains,
so
this
was
a
design
that
I
think
was
closely
related
to
volume
groups,
if
I
remember
collect
correctly
xing.
Do
you
want
to
talk
about
this.
D
Yeah,
so
we
haven't
really
get
to
talk
about
the
placement
part,
yet
we
are
still
looking
at
the
the
first
part:
how
to
create
a
group,
and
how
do
you
take
group
snapshots,
so
we
had
a
review
meeting
to
go
through
the
design,
so
I
think
there
were
some
in
last
meeting.
There
are
some
concerns
regarding
this
api
that
we
are
proposing
right
now,
which
allows
the
volume
membership
to
be
changed
one
by
one.
D
I
think
there's
some
concern
about
the
scalability,
so
I'm
still
thinking
about
that
because
I
look
at
the
the
later
part
of
the
design,
which
is
the
the
csi
step
changes
there.
I
actually
have
a
csi
spec
css
spec,
that
is
making
the
change.
D
That's
yeah
yeah,
but
I
mean
if
we
don't
have
that
at
the
kubernetes
api
level,
then
we
have
to
do
the
way
that
you
suggested
to
to
kind
of
watch
like
a
list
of
right
volumes
right.
So
we
can
look
into
that
yeah.
So
maybe
maybe
we
can
maybe
we
can
have
both.
Maybe
we
can
just.
We
can
add
that
in
the
css
back
first
and
then
yeah,
I
think
yeah
so
still
need
some
thought
and
then
I
need
to
schedule.
Another
meeting
to
follow
up.
A
G
Yeah,
nothing
has
changed
since
last
time
and
I
haven't
been
able
to
take
a
look
at
this.
Okay,
no
worries.
Thank
you.
A
A
G
I
believe
they've
they've
all
been
my
all.
The
code
has
been
migrated
to
the
new
repos,
so
I
think
the
next
thing
he's
looking
at
is
setting
up
the
ci
infrastructure
for
it
right.
Okay,.
A
A
Okay
does
shane,
do
you
remember
if
we
got
confirmation
from
mike
if
you
wanted
to
continue
your
work
on
this.
A
Okay,
so
let's,
let's
see
what
happens
at
the
next
cycle,
we
may
need
to
find
a
new
owner
for
that.
One
next
item
is
csi
volume,
health
shane.
Do
you
want
to
talk
about
this?
One.
D
Yeah,
so
we
have
cut
the
first
release
zero
point,
one
of
these,
so
next
step.
So
right
now
we
only
implement
this
in
the
mock
driver.
Next
step
is
to
implement
this
in
the
hostpass
driver,
so
we
can
add
e3
tests
now
regarding
to
whether
we
should
bring
this
to
beta
or
not.
I'm
thinking,
I
think
I
mentioned
this
last
time
as
well.
Should
we
be
looking
at
whether
to
add
this
information
into
the
like
pvc
status
so,
but
that
would
be.
D
D
That
probably
will
be
a
different
it's
going
to
require
a
if
we
want
to
add
that
in
the
status,
then
that
wouldn't
require
feature
gate
anyway.
So
yeah,
it's
very
different
from
this
one
now,
but
this
one
we
we
do
need
to
add
e3
tests
and
trying
to
make
it
more
stable
fixing
dogs,
any
tests.
F
Okay,
I
am
going
to
work
on
it.
This
cycle,
I
have
the
design
in
my
head,
I'm
going
to
write
it
down
and
I'm
going
to
implement
it.
A
I
A
Well,
thank
you
for
that
update
and
yeah.
If
anyone
on
the
call
is
interested,
this
is
a
fairly
large
new
effort
to
come
up
with
a
csi
like
interface
for
object.
Storage.
If
you
are
interested,
there
is
a
meeting
right
after
this
meeting
on
the
same
call
or
on
the
same
zoom
channel,
so
if
you're
interested
feel
free
to
join
that
and
I'll
provide
feedback
and
shape
this
new
project.
J
A
Sound
good
sounds
good.
Can
you
jog
my
memory
as
to
what
this.
J
A
A
So
we
have
the
core
component,
which
is
responsible
for
migration,
and
then
the
per
entry
volume
plug-in
component
for
migration
for
the
csi
core
migration
piece,
we're
still
looking
for
an
owner,
is
anyone
on
the
call
interested
in
contributing
this
is
going
to
be
a
critical
piece
of
code.
It's
already
coded.
The
effort
here
is
to
move
it
to
ga.
It's
already
been
moved
to
alpha
and
beta,
so
the
requirement
here
would
be
to
understand
how
this
code
works.
Look
at
the
existing
implementations
understand
what
the
test
coverage
looks
like
and
see.
A
A
Okay
and
if
anyone's
too
shy
to
speak
up
on
the
call
feel
free
to
reach
out
to
any
of
the
leads
offline
and
we'd
be
happy
to
help
get
you
bootstrapped
on
this.
If
not
we'll
continue
to
look
for
an
owner
for
this,
this
might
become
a
blocker
for
the
larger
migration
effort.
If
we
don't
find
an
owner
soon.
A
Speaking
of
the
larger
migration
effort,
let's
get
a
status
update
from
the
different
cloud
providers,
so
first
up
is
vsphere
moving
to
or
staying
in,
beta
divian.
You
want
to
give
an
update
or
shang.
If
you
have
any
updates
here.
D
It
looks
like
davian
is
not
in
a
meeting,
so
we
are
still
working
on
trying
to
backport
some
of
the
fixes
to
673,
but
we
have
not
finished
that
yet
so
I
can't
confirm
on
that.
Yet
so
it's
working
progress.
A
Good
all
right,
thank
you
for
that
update.
Shane
next
item
is
azure
disk
and
azure
file.
I'm
guessing
andy's,
not
on
the
call
michelle.
Do
you
know
what
the
status
of
these
are.
G
I
need
to
sync
up
with
him,
but
last
time
I
think
azure
file
was
just
missing
like
there
was
one
bug
that
was
preventing
it
from
going
to
beta,
but
I
think
it
was
mostly
a
matter
of
updating
the
sidecars,
so
I
will
sync
up
with
andy
on
that.
Oh,
but
I
think
hamad
raised
a
possible
concern
with
azure
file.
I
think
there's
one
one
feature
that
is
still
missing,
so
we'll
probably
need
to
sync
up
with
andy
on
on.
B
A
We
kind
of
hear
you
now.
A
A
B
Because
the
entry
driver
applies
fs
group
as
a
mount
option
like
compared
to
like
any
other
driver.
A
Okay
sounds
good,
so
it
looks
like
that
work
is
under
review
and
and
we'll
get
a
status
update
on
that
next
time.
A
Next
item
is
gce,
csi
migration,
so
matt
carey
from
our
team
is
gonna,
help
push
this
along.
I
don't
believe
he's
had
a
chance
to
start
working
on
it.
Yet
we'll
give
a
status
update
here
as
soon
as
we
can.
G
K
A
And
same
thing
for
aws
pushing
towards
ga
anyone
have
a
status
update
on
that.
K
A
So
we'll
leave
that
as
we'll
follow
up
with
matt.
Wong
next
item
is
csi
migration
for
openstack
sender.
John.
Are
you
on
the
call
by
any
chance?
I.
L
L
That's
not
this
one.
It's
the
the
core.
A
So
maybe,
if
you
could
shoot
me
michelle
schengen
email
offline,
we
can
get.
You
started
on
that.
A
L
So
you
might
say
I'll
update
at
the
next
meeting,
got
it
or
in
the
meantime,
whatever.
A
Okay.
Thank
you
john
appreciate.
It
next
item
is
csi
migration
force,
ffs
and
ceph
rbd
we've
got
humboldt's
name
down
here,
I'm
gonna
guess:
we've
got
no
updates
on
this.
A
A
There
was
concern
about
frequency
of
mounts
causing
scalability
issues
because
they
go
fetch
api
objects
and
so
sig
scalability
helped
introduce
the
option
for
immutable
secrets
and
config
maps
to
address
these
scalability
issues.
This
is
an
opt-in
feature
not
on
by
default.
So
if
you
run
into
scalability,
issues
is
something
you
would
have
to
opt
into.
A
A
Say:
okay,
leave
that
as
a
maybe
next
set
of
items
here
are
for
sig
apps.
A
First
up
is
addressing
the
issue
where,
if
you
have
a
stateful
set
that
creates
a
bunch
of
pvcs,
you
delete
the
stateful
set.
The
pvcs
today
are
not
going
to
be
deleted,
which
is
frustrating
dave
smith,
I
believe,
was
assigned
to
work
on
this
shane.
Do
you
know
what
the
status
update
here
is.
H
Go
ahead,
yeah,
so
basically
I've
updated
the
kip.
With
after
addressing
some
comments,
there
is
one
scenario
which
we
need
more
discussion.
That
is
a
scenario
where
manually.
If
a
port
gets
deleted,
what
our
behavior
should
be
so
once
that
is
addressed,
most
of
the
major
comments
would
get
addressed
with
it.
H
Yeah,
so
the
next
one,
the
volume
expansion
one
I
have
not
it-
there's
no
update
on
this.
Yet
I
have
raised
up
kep.
There
are
few
concerns
from
haman's
discussion,
but
no
updates
further.
A
D
Yeah
so
yeah
we
are
trying
to
bring
this
java.
Let's
see
how
it
goes,
so
I'm
going
to
open
an
issue,
so
this
will
be.
Is
this
owned
by
a
signal
was
jointly
by
signal,
then
the
six
largest.
A
I
think
I'll
just
put
both
yeah
we've
got
sig
note
on
there.
We've
got
sig
apps,
so
I
think
there's
three
ones.
D
Okay,
okay,
I
maybe
I
would
just
put
all
three
there
and
see
yeah,
so
sean
chen
and
I
are
working
on
a
cap,
and
then
we
will
bring
this
up
in
the
next
sig
note
meeting
next
week,
just
to
give
them
an
update
that
we
are
working
on
this,
so
basically
will
be
so.
The
api
will
be
the
same
as
what
we
have
proposed
in
that
google
doc
and
an
implementation
is
going
to
be
studying
that
in
a
controller
first,
with
no
signal
changes
to
begin.
A
Last
item
here
is
an
edict
from
sig
architecture.
They
have
asked
us
to
move
kubernetes
utils
mounts
to
its
own
independent
repo,
and
so
I
think
this
is
mostly
complete,
except
for
adding
new
end-to-end
tests
and
updating
the
kubernetes
kubernetes
repos
to
point
to
the
new
repo
go
ahead.
I
Yeah,
actually,
thanks
to
michelle
that
the
first
pr
is
merged,
but
now
I
created
another
pr
to
move
basically
search
and
replace
of
the
uts
mount
to
mount
details.
It's
a
large
pr
because
it's
having
a
lot
of
fuss
that
change
yep
across
six.
So
hopefully,
if
that
goes
through
it's
it's
built
on
the
iran
et.
So
it
should
be
fine.
But
there
are
two
external
dependencies
too.
Like
c
advisor
uses,
details
mount
and
then
azure,
I
think
it
uses
ito's
mount.
I
A
Would
it
be
okay
for
you
to
update
those
and
give
them
kind
of
just
put
them
as
reviewers
on
the
pr.
A
A
Yep
all
right
cool.
Thank
you.
Sereni,
for
that
update
sounds
like
we're,
making
good
progress
here
all
right.
Thank
you.
Everyone
for
the
updates
today,
next
up
on
the
agenda
prs
that
need
attention,
we've
got
one
here.
M
After
csi
migration,
we
are
recreating
storage
classes
with
the
same
name,
but
with
the
no
storage
cost,
with
the
new
provisioner
that
the
out
of
three
pro
provisioner
and
this
pr
is
just
updating
the
volume
expand
controller
to
to
rely
on
the
rely
on
the
pv
spec
to
to
get
the
entry
provisioner
name
instead
of
relying
on
the
storage
class
name.
M
Basically,
this
is
this
is
a
corner
case,
and
the
api
right
now
right
now
allows
you
to
delete
the
storage
course
and
recreate
a
new,
completely
different
storage
course
and
yeah.
It's
a
small
pr-
and
I
think
the
source
of
true,
at
least
in
my
opinion,
should
be
a
little
tv,
yeah
I'll,
be
happy
to
have
some
feedback.
C
G
I
also
briefly
looked
at
it
too,
and
and
kind
of
two
things
struck
out
at
me,
like
first,
was
that
it's
possible
for
someone
to
recreate
a
storage
class
but
like
completely
change
the
contents
of
it,
and
that
potentially,
in
this
case,
can
cause
like
existing
pv
objects
that
were
created
using
a
previous
version
of
the
storage
class
to
no
longer
become
to
be
like
completely
valid
and
and
then
the
second
thing
is
that
and
because
of
that,
we
we
have
been
sort
of
stating
you
should
not
depend
on
storage
class
after
a
volume
is
created,
but
here
the
resizer
is
depending
on
it
because
it
needs
to
check
if
the
allow
volume
expansion
field
is
set,
and
that
may
be
problematic
in
this
scenario.
A
Yeah,
I
think
that
might
be
the
better
approach,
but
I
guess
the
problem
here
is
also
it's
kind
of
different
from
the
other
fields
that
we
copy
to
the
pv.
Where
it's
really
a
pv
specific
attribute,
you
could
argue
that
allowing
volume
expansion
is
a
cluster-wide
option
that
a
cluster
admin
wants
to
make
for
an
entire
set
of
volumes
rather
than
volume
by
volume.
E
G
B
B
We,
the
idea,
was
that
it's
like
certain
aws
clusters,
for
example,
the
resize
quota,
is
much
lower
than
the
other
quota,
so
an
admin
one
might
not
want
to
enable
like,
in
a
multitude
and
cluster
users,
to
just
resize
their
ppc
so
and
in
other
clusters
like
they
might
so
that
was
the
original
idea
of
having
this
this
knob
in
the
storage
class,
to
allow
that
control
in
the
hands
of
admin.
G
B
B
L
A
C
B
B
And
if
admin
tries
to
modify
it,
some
sort
of
reconciliation,
overwrite
admin
changes.
A
M
G
I
think
so
this
is
kind
of
I
think
the
issue
you're
hitting
kind
of
reveals
a
bigger
issue
like
you
could
change.
The
storage
class
like
in
this
case,
you're
changing
the
storage
class
to
a
driver
with
like
the
equivalent
driver,
but
you
could
potentially
change
the
storage
class
to
a
completely
unrelated
driver,
and
that
is
problematic.
M
Yeah
we
have,
I
completely
agree
that
and
last
week
yeah
the
that
api
allows
you
to
recreate
the
storage
class
and
you
you're
not
completely
sure
when
you
change
some
fields
which
changes
compatible
and
which
change
is
not
compatible.
I
saw
multiple
times
in
seek
storage
that
you
recommend
creating
storage
class
with
the
with
the
same
name,
but
it's
kind
of
hacky
right
now.
M
The
field
in
the
storage
class
is
immutable
and
the
recommendation
slash
best
practices
to
recreate
the
storage
class,
and
at
this
I
find
it
confusing
and
yeah,
but
our
our
change,
which
is
unrelated
change.
What
are
we?
This
week?
We
changed
our
volume
binding
modes
from
from
immediate
to
wait
for
first
consumer
and
yeah,
thanks
to
michelle
for
for
making
sure
that
this
change
is
compatible.
M
This
is
just
an
example
like
you
have
to
deep
dive
in
the
code
to
check
whether
whether
this
is
safe.
A
Yeah,
I
think,
moving
forward
the
the
guidance
that
we've
given
to
developers
is
that,
if
you
introduce
a
class
object,
ensure
that
any
object
that
is
provisioned
using
that
class
object
does
not
depend
on
the
original
object
after
provisioning,
and
so
what
that
means
is
copy
in
you
know,
whatever
information
you
you're
going
to
need
for
for
the
future
from
the
storage
class
object
to
the
the
the
pv,
for
example,
and
so
cozy
is
doing
exactly
that
in
its
design.
A
This
one
is
a
little
bit
tricky
because
it's
not
necessarily
a
per
volume
specific
option.
It
is
something
that
a
cluster
admin
would
want
to
set
at
a
cluster
level,
and
so
we
should
figure
out
where
the
right
home
for
this
is
before
moving
resize
to
ga.
D
So
if
we
make
this
a
driver
level,
what
if
a
driver
supports-
let's
say
both
block
and
firewalling
and
one
supports
one
expansion,
the
other
one
does
not,
then
you
can't
differentiate
them
anymore.
You
can
use
a.
G
G
D
It
is
not,
but
you
can
create
a
different
sturdy
class.
You
can
add
other
attributes
yourself,
you,
you
know
what
that
is
or
your
driver
would
know
what
that
is
for.
B
D
We
do
have
one
we
have
one
driver
that
supports
both.
I
think
some
drivers
even
support
different
storage
systems.
How
do
you.
A
D
D
K
Or
something
yeah
yeah,
okay,
so
yeah.
D
I
think
yeah
just
thinking
there
are
use
cases
and
also
I
I
know
there
are
some
cs
driver.
They
actually
support
multiple
different
storage
systems,
then
that
would
be
even
more
possible
different,
just
resistance,
but.
G
D
So
that
would
be
the
admin
would
have
a
way
to
to
create
different
storage
class
and
then
somehow
it
has
a
way
to
differentiate.
I
don't
know
exactly.
N
Is
it's
a
user
requested
pvc
with
access
mode,
read,
write
mini
a
storage
class
might
prove
the
controller,
might
provision
a
file
implementation
versus
if
they
request
rewrite
once
it
would
do
a
block.
Potentially,
if
I
were
to
implement
it,
that's
what
I
would
do
I
would.
I
would
more
likely
to
create
multiple
storage
classes
so
that
it's
more
obvious
to
users,
of
course,
but
the
reason
you
might
want
to
make
it
the
same
storage
class
is
quota
is
extremely
painful,
with
storage
classes.
H
So
there
are
other
parameters
which
some
of
the
storage
classes
uses,
which
eventually
determines
what
kind
of
whether
it's
a
block
or
file.
That
approach
is
also
used.
H
Like
you,
give
us
cue
name
in
the
parameter
like,
for
example,
in
azure,
you
give
us
queue
name
that
determines
what
eventually,
whether
it's
a
block
or
file.
A
A
So,
given
that
in
this
problem,
anyone
have
a
proposal
on
how
to
move
forward
on
it,.
H
So
the
the
config
map
property
feature
which
you
had
mentioned
sad,
does
that
in
any
way
help
us
here.
If
we
have
such
thing
where
pvc
can
point
to
it,
could
this
go
into
one
of
those
like
one
like
the
config
map.
A
I
think
the
config
map
approach
was
beneficial
when
we
wanted
to
have
a
user,
be
able
to
override
a
storage
class
parameter
per
kind
of
usage,
so
I
I
don't
think
it
would
help
in
this
case.
This
is
kind
of
a
different
problem
where
we
have
a
value
in
storage
class.
It
seems
like
folks
like
it
where
it
is,
but
you
have
this
edge
case
of
well
after
your
volume
is
provisioned.
You
shouldn't
be
depending
on
the
original
storage
class
for
anything,
but
in
this
case
it
does
so.
A
I
guess
one
option
is
that
we
do
what
we
recommend
for
other
such
fields,
which
is
copy
it
into
the
pv
object
and
use
the
only
only
the
pv
object
after
the
volume
is
provisioned,
but
what
that
implies
is,
after
a
volume
is
provisioned
allow
volume,
expansion
becomes
a
per
volume
setting,
and
if
a
cluster
admin
wants
to
change
it,
changing
the
storage
class
would
no
longer
be
sufficient.
They
would
need
to
go
and
find
all
the
pvs
update.
All
the
allow
volume
expansion
per
pv.
F
Do
we
need
to
think
about
some
sort
of
like
a
policy
object
that
that
does
that
is
updatable?
That
volumes
can
continue
to
refer
to
after
creation
time,
because
because
the
storage
class
is
we
see
this
over
and
over,
where
storage
class
is
like
okay,
this
is
this
is
what
I
want
at
creation
time,
but
once
the
volume
is
created,
I
forget
about
that
thing
and
it's
like
well.
What
if
I
want
to,
I
see
this
with
qos
right.
F
You
have
a
you,
have
a
bad
qos
policy
or
a
moderate
qs
policy,
and
you
want
to
change
it
later.
It's
like!
Well,
you
can't
just
go
change:
a
storage
class.
You
can't
re-type
the
volume
you
kind
of
just
have
to
tweak
that
volume.
If
there
was
an
object
that
the
volume
referred
to,
that
you
could
mutate,
some
sort
of
a
policy
object
that
lived
apart
from
the
storage
class
it
would,
it
would
have
many
uses,
I
think,
for
tweaking
volumes.
The
existing
volumes.
H
N
F
Yeah
yeah,
like
the
the
implications
of
allowing
someone
to
change
the
qos
policy.
When
you
have
quotas
on
your
storage
classes,
for
different
qs
policies
becomes
a
nightmare
because
because
now
you
can
sort
of
work
around
the
quota
system
by
by
changing
some
policy
and-
and
you
want
to
give
the
admin
veto
power
over
that.
But
there's
no
obvious
way
to
do
it
other
than
just
a
blanket.
You
can't
do
that
kind
of
a
statement.
A
Okay,
we've
got
one
minute
left.
I
think
this
discussion
is
going
to
be
take
a
longer
time.
Maybe,
as
a
next
step,
we
set
up
a
one-off
meeting
to
discuss
this.
Does
anybody
want
to
take
the
action
item
for
that.
B
A
B
A
Okay,
let's
discuss
that
at
the
at
the
resizing
meeting
tomorrow
ismail.
Can
you
join
that
meeting
tomorrow.
B
Let
me
see
so
tomorrow
we
have
12
like
same
time
as
today.
Actually
this.
M
I
I'm
not
sure
how
how
this
general
discussion
about
allowing
volume
expansion
is
related
to
this
pr,
because
right
right
now,
this
usage
that
I'm
right
now
modifying
is
for
for
yeah
for
getting
the
provisioner
yeah.
A
I
think
what
is
saying
is
it
might
be
sufficient
and
we'll
go
ahead
and
merge
that,
but
let's
decide
it
at
that
meeting.
A
Thanks
all
right!
Well,
thank
you,
everyone
for
joining
we're
at
the
hour
and
we'll
reconvene
in
two
weeks
and
for
folks
that
are
interested
in
sticking
around
I'm
gonna
end
this
meeting
and
we're
gonna
start
the
cozy
meeting
right
after
this
all
right.
Take
care,
see
you
guys
there
thanks
bye,.