►
From YouTube: Kubernetes SIG Storage 20190926
Description
Kubernetes Storage Special-Interest-Group (SIG) Meeting - 26 September 2019
Meeting Notes/Agenda: https://docs.google.com/document/d/1-8KEG8AjAgKznS9NFm3qWqkGyCHmvU6HVl0sk5hwoAE/edit#heading=h.iroeuobui0yw
Find out more about the Storage SIG here: https://github.com/kubernetes/community/tree/master/sig-storage
Moderator: Saad Ali (Google)
Chat Log:
None
A
Today
is
September
26
2019.
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're
going
to
go
over
we're
going
to
go
through
1.17
planning,
so
the
1.17
release
dates
are
by
the
15th.
We
need
to
nail
down
the
features
that
we're
working
on
and
have
advancement
issues
open
for
them
and
any
caps
merged
for
those
features
by
November.
14Th
is
code
freeze
and
then,
when
member
19th
Doc's
must
be
completed
and
then
December
9th
is
the
release.
A
This
cycle
overall
is
shorter
than
normal,
because
Q
Khan
is
early,
is
late,
November
and
then
holidays
already
December.
So
just
like
you
are
in
general,
a
shorter
and
then
I
also
want
to
discuss
the
face
to
face
meeting.
Excuse
me
guys
anything
else
you
want
to
discuss,
please
feel
free
to
add
to
the
agenda
and
we'll
take
it
from
there.
A
Ok,
next,
let's
move
over
to
the
planning
spreadsheet,
we
moved
over
items
from
1.16
that
were
not
completed,
and
so,
let's
go
through
those
one
by
one.
If
there's
any
item
that
you
think
should
be
on
here
feel
free
to
go
ahead
and
add
to
the
bottom,
and
you
have
the
next
two
weeks
to
do
that,
if
you
think
of
something
within
the
next
two
weeks
feel
free
to
add
to
the
bottom,
and
we
can
review
that
next
time
or
if
you
want
to
you,
could
do
it
during
this
session
as
well.
A
B
I
think
all
three
of
us
will
be
focusing
on
this.
The
couple
of
new
items
I
think
we
want
to
pick
up
for
117-
is
support,
disabling
and
basically
tie
in
with
the
provider
less
bill
flag,
that's
also
being
introduced
separately
in
communities,
so
provided
less
well
get
rid
of
the
in
creeped
logins
at
a
compile
level,
and
this
you're
trying
to
introduce
a
stage
before
that
bill
flag
takes
effect
so
that
you
have
more
confidence
that
things
will
actually
work.
Then
the
entry
plugins
completely
get
wiped
out,
or
rather
compile
out.
Okay,.
A
B
A
A
C
Yeah
yeah
pretty
much.
We
want
to
allow
calling
node
expand
volume
on
all
the
nodes
where
they
try
it.
What
is
it
I,
so
sweet
right
for
read-only
handling?
We
need
expect
change
in
CSI
that
I'll
proposal
for
that
I'm
working
on
it
and
yeah.
That's
the
that's
the
mean
of
the
issues
actually
and
they're
trying
to
fix
for
170,
okay,.
A
A
A
Okay,
no,
we
are
still
committed
to
this
and,
as
far
as
I
understand,
Travis
will
continue
to
work
on.
It.
Next
item
is
per
node
volume
attached
limits.
We
are
trying
to
get
this
to
GA.
This
quarter.
I
am
concerned
about
how
capable
we
are
going
to
be
if
that
Fabio
has
been
the
one
working
on
this
hobby
or
Yan
RI
there
for
you
on
the
line.
C
A
I
mean
I.
What
I
remember
was
the
issues
around
how
I
think
AWS
has
some
weird
issues
with
how
it
calculates
attachment
limits,
instead
of
it
being
only
volume
based.
It
takes
into
account
all
the
resources,
networking
and
other
things
and
calculates
limits
based
off
of
that
which
I
don't
know
if
we're
capable
of
doing
today
and
then
similarly
I
think
be
sphere.
A
C
A
D
E
D
A
A
C
We
are
working
on
it
actually,
as
a
like,
like
Brad
me,
I
had
a
call
with
Jing
last
week
about
where
she
is
with
her
and
we're
trying
to
see
how
we
can
unify
both
a
single
next.
But
how
hard
is
that
was
more
just
about
like
fixing
the
CH
owns
and
chmod
stuff,
and
we
are
also
interested
in
fixing
the
CH
con
I
think
that
is
still
the
right
person
in
at
least
okay.
So.
C
E
I
I'm
not
super
familiar
with
this
issue,
but
my
sense
is
that
the
the
underlying
problem
is
that
when
you're
in
a
container
and
the
kernel
is
doing
GID,
UID
mapping.
E
For
the
for
the
file
system,
G
IDs
and
you
II's,
it
needs
to
be
configurable
separately
so
that,
if
you
run
a
container
one
day
and
root
is
UID
1
million,
and
then
you
run
the
same
container
the
next
day
and
you
ID
2
million
that
the
the
file
system
you
ID
bits
are
the
same.
I
think
the
problem
is,
is
that
they're
not
and
and
I
I
need
to
look
I
need
to
look
into
exactly
what's
going
on,
but
I
think
there's
just
no
facility.
It's
a
sort
of
deal.
B
E
Translation
on
on
a
like
the
file
file
system,
UID
tid
bits
in
when
you're
running
in
a
container,
and
that's
why
we
have
at
this
problem
I
think
so.
So
any
attempt
to
like
to
you
know,
challenge
and
stuff
is
just
gonna
lead
us
around
in
circles.
If
the
kernel
just
did
the
mapping
for
us-
and
you
could
hide
all
that
and
then
the
problem
would
go
away.
A
All
right,
thank
you.
I'm
gonna,
put
down
Brad's
name
for
all
three
of
these
since
they're
very
closely
related
I
think
at
a
minimum.
We
should
probably
hit
the
non-recursive
volume
ownership
issue.
Since
it's
a
performance
issue,
we
could
go
from
there
next
item
closely
related
is
file
permission
handling
for
Windows
deeper.
You
still
the
right
person
to
work
on
that
yeah.
F
A
C
So
the
cap
was
approved
by
Sega
apps.
It
was
approved
by
I'm
gonna,
approve
it
ILD
200,
but
right
now
it's
Jordan
is
doing
it
for
API
changes
and
he
has
some
comments
open
for
it
and
I
think
that's,
there's
a
there's,
a
open
concern.
It's
already
covered
in
the
cap
itself
that
and
and
enjoyin
saying
it
issue
resolved
before
the
Cape
is
implementable.
Okay,.
E
C
The
CSI
block
volume
has
some
issues
about
like
cleaning
up
is
right
now
because,
like
me,
Michelin
year-
and
this
could
be
the
the
thing
is
that
map
the
stage
and
publish
both
are
done
in
the
same
operation
and
one
fails.
The
other
one
is
like
it
just
left
it
as
it
is,
and
and
the
cleaning
up
handling
is
kind
of
ugly.
So
there's
a
PR
outstanding,
that's
pretty
big,
which
kind
of
detangles
and
and
and
these
two
things
so
that
they're
called
separately.
C
C
E
A
G
A
A
Go
let
me
copy
that
over.
A
And
then
also
for
anybody
else
on
the
call
I
am
just
assigning
the
the
features
out
to
folks
who've
already
committed
to
them
feel
free
to
jump
in.
If
you
are
interested
in
contributing
to
say,
hey,
I'm,
open
and
I'm
willing
to
work
on
something
we'd
like
to
increase
the
number
of
contributors
we
have
I
know
there's
a
handful
of
people,
in
this
sake,
Amethi
on
been
John
and
Michelle,
and
a
handful
of
others
who
work
on
a
lot
of
stuff,
and
they
have
a
lot
of
load
and
a
lot
of
items.
A
A
B
So
this
work
is
in
progress.
We
managed
to
get
like
a
nice
IKEA,
but
little
earth
in
father
sit
the
proxy
itself
in
the
CSI
proxy
repose
of
Michelle
reviewed
it
that's,
and
so
now
we're
working
on
the
API
groups
to
start
building
up
the
real
proxy
support
for
like
file
system
disk
and
things
like
that.
A
A
A
Next
item
is
snapshots
to
beta
I.
Think
that's
another
item.
We
need
to
prioritize
higher
because
it's
been
a
very
long
time.
There
are
a
number
of
changes
going
into
API
changes,
controller,
split
and
I.
Think
those
are
the
two
big
things:
metrics
Shing
and
Shawn
are
working
on
those
and
then
Michelle
can
help
review.
The
API
and
I
can
help
review
the
code
on
that
Shangri
Shawn.
Do
you
want
to
add
anything
here?
I.
A
A
A
A
Okay,
as
far
as
I
know,
he's
still
committed
to
coming
up
with
a
design
for
this
here
and
I
can
also
put
down
my
name
I'm,
going
to
look
into
how
we
can
surface
some
of
the
operational
metrics
in
the
CSI
drivers.
Up
next
issue
is
related
to
volumes
issue
related
to
assuming
volumes
or
mount
points.
I
haven't
looked
into
this
one,
but
seven,
two,
three,
four
seven
I
think
we
were
looking
for
people
to
help.
With
this
last
quarter
we
weren't
able
to
find
anyone.
A
G
G
What
I
would
like
to
do,
though,
is
within
the
next
couple
of
weeks
if
we
could
finalize
on
our
objective
in
terms
of
the
API
and
agree
and
what
I
mean
by
that
is
agree
that,
even
though
doing
a
generic
API
for
it
we're
not
actually
doing
it
with
the
intent
of
requiring
an
implementation
for
the
other
resources.
I
think
if
you
look
at
smarter
clayton's
comments
and
Tim's
comments
of
what
that
might
be,
causing
some
confusion.
G
A
Sense
and
we
are
planning
a
face-to-face
meeting
the
day
before
cube
Con
in
November
I
believe
November
18th,
I'm
gonna
talk
a
little
bit
about
that
at
the
end
of
this
call,
I
think
this
will
be
an
important
design
to
discuss
there.
So
if
we
have
something
kind
of
more
fleshed
out,
we
can
present
it
to
the
wider
community
there.
Okay,
it's
for
visibility
and
discussion.
A
A
That'd
be
great,
and
what
I'll
do
is
I'll
put
together
an
agenda
dock
where
people
can
start
listing
items
that
they're
interested
in
in
discussing-
and
we
can
add
this
there
as
well.
Next
item
is
volume
group
API.
This
is
something
that
we
use
kind
of
pump
the
brakes
on.
We
should
probably
pick
it
up
soon.
I
think
having
it
a
lower
priority,
p3
is
still
fine
and
designed
for
this
quarter
would
be
fine
Shing.
Are
you
still
interested
in
helping
drag
back?
Yes,.
H
A
H
A
A
A
All
right
next
item
are
CSI
drivers,
we've
got
NFS,
I,
scuzzy,
fiber,
channel
and
flex
for
NFS
and
I
scuzzy.
Both
of
them
have
standalone
repos
already
and
they
have
code
in
those
repos,
but
we
need
to
do
image.
Building
testing
see
ICD
documentation
to
actually
make
them
usable
for
fiber,
channel
and
flex
they're,
currently
in
a
common
repo,
and
we
want
to
split
that
out
and
kind
of
do
what
we
did
with
as
you
Nana
fest
for
those
Mithu
sin
was
assigned
to
work
on
the
NFS
one
Mathewson.
C
Hehe
was
a
intern
in
our
team
and
I
think
I'm
going
back
to
university.
Okay,.
A
G
G
Last
week
he
had
an
interesting
proposal
about
modifying
some
of
these
a
little
bit
so
rather
than
just
offer
the
library
to
make
the
connections
and
stuff
like
that,
the
helper
libraries
he
had
talked
about
actually
having
a
base,
for
example,
I
scuzzy
driver,
so
all
of
the
basically
all
of
the
node
operations
would
be
implemented
and
a
common
thing
that
you
would
just
deploy,
and
then
you
would
only
have
to
implement
the
controller
side
I
just
wanted
to
raise
it
here
is
something
that
I
think.
Maybe
we
should
talk
about
and
investigate.
G
A
So,
just
to
make
sure
I
understand
correctly
currently
there's
two
models
that
are
supported.
One
is
you
get
the
melt
code
for
NFS
or
ice
cozy
or
fiber
channel
as
a
library
that
you
can
import,
and
then
you
write
going
to
write
the
rest
of
your
driver,
including
the
controller
and
provisioner,
and
the
second
model?
Is
you
get
a
very
stripped-down
version
of
these
drivers?
That
only
knows
how
to
do
the
mount
doesn't
know
how
to
do
provision
and
so
I
guess.
A
G
Correct
that
that's
the
that's
the
gist
of
it.
There
were
some
details.
He
couldn't
quite
sort
out
for
me
personally,
I
wasn't
a
I'm
personally,
not
a
fan
of
the
idea.
I
think.
The
last
thing
we
need
is
yet
another
path,
but
it's
not
fair
for
me
to
to
make
that
decision
on
my
own,
so
I
wanted
to
raise
it
to
the
recipe
so.
A
A
Keyboards
are
the
most
important
thing,
so
if
anyone
on
the
call
is
interested
in
working
on
these
in
any
capacity,
please
speak
up
now
or
feel
free
to
reach
out
reach
out
to
me.
Offline
and
I
can
connect
you
to
the
right
folks.
Otherwise,
for
the
time
being,
beads
will
continue
to
be
sadly
neglected.
D
A
In
some
capacity,
yes,
the
I
think
the
problem
is
that
drivers
code
is
there,
but
they
haven't
really.
It
hasn't
really
been
tested.
We
don't
actually
build
images.
So
how
do
you
deploy
the
driver
that
kind
of
got
it
at
least
for
NFS
and
I
scuzzy,
and
then
for
fibre
channel
and
flex?
There
was
an
initial
attempt
when
we
were
first
doing
CSI
to
build
these
out
and
they
haven't
really
had
much
love
since
then.
So
the
first
step
there
is
getting
it.
A
A
A
A
C
There's
a
general
problem:
I'd
set
a
set
out
to
fix
just
for
like
mount
car
like
stage
and
unpublished,
but
it's
a
problem
in
general
in
in
in
Canaries
like
in
cubelets
I'd.
If
you,
if
you
try
to
stage
a
volume
and
then
the
stage
operation
times
out
and
user
deletes
the
part,
then
that
part
that
that
mount
is
forever
lost
or
it's
not
cleaned
up
properly.
Similar
thing
would
happen.
Unpublish
also,
similarly
like
when
you
unpublish
a
volume
and
unpublished
operation
times
out.
That
has
to
be
handled
correctly.
C
So
it's
like
I'm
similarly,
like
UNPO
on
stage,
has
to
have
an
assembler
problem
and
the
rob
lock
devices
have
similar
problem,
but
they're
good
math
is
different.
They
have
to
be
fixed
in
similar
way,
so
I
have
a
peer
open
for
that
fixes,
just
at
least
right
now.
It
just
fixes
like
mount
device
and
mount
volume
good
path,
but
I'm
working
on
public,
unpublished
or
an
on-stage
also,
but
I
think
I
need
help
fixing
rob
block
cleanup
and
it's
just
a
problem
that
has
to
be
fixed
in
cubelet
as
over
our
overall
and
the.
A
C
C
A
C
A
Sounds
good,
okay,
I
think
that
is
all
the
items
that
we
have
there's
still
a
lot
of
items
and
I'm
still
concerned
about
people
who
have
multiple
items
assigned
to
them,
especially
given
that
this
is
a
short
quarter.
So
please
take
a
look
at
your
assignments
and
if,
in
two
weeks
time,
when
we
meet
again,
if
there's
something
that
you
can't
commit
to,
you,
please
feel
free
to
to
update
us
and
we'll
see
if
we
can
either
drop
the
commitment
for
this
quarter
or
find
somebody
else
to
work
on
an
item.
A
A
A
Okay,
so
now
the
date
to
remember
going
forward
is
October
15th.
So
if
you
have
a
feature
assigned
to
you,
please
make
sure
that
it
has
a
enhancement
issue
opened
and
if
it's
a,
if
it
requires
a
cap
that
the
cap
is
merged
and
approved
by
that
date,
October
15th
Oh
two
weeks,
basically
moving
on,
looks
like
deep
has
a
PR
that
needs
attention:
support
for
Windows,
I,
scuzzy
flex,
volume.
B
Yes,
so
this
is
basically
in
the
target:
D
repo,
that's
still
under
kubernetes
incubator.
So
basically,
what
we
wanted
to
do
from
the
seg
window
side
was
since
the
CSI
support
is
still
being
developed
and
it's
an
alpha.
There
myself
released
a
couple
of
flex
volume
scripts
to
support
ice,
cozy
and
SMB,
and
this
adds
the
support
for
that
flex.
Volume
script
in
the
target,
D
provisioner
nicely.
D
B
A
A
A
Where
cube
con
is
located,
it
is
going
to
the
event
is
going
to
be
on
Monday
November
18.
This
is
the
day
before
the
conference
it's
going
to
be
9:00
a.m.
to
5:00
p.m.
so
it's
a
full-day
event
keep
an
eye
out
for
an
email.
I'll
show
you
I'll
send
that
out
shortly
detailing
how
you
register
for
this
face
to
face
meeting.
Please
note
that
it
does
require
a
registration
for
cube
con
itself.
A
A
H
A
So
I'm
hoping
I'm
going
to
see
a
whole
bunch
of
faces
that
are
on
the
call
there,
and
if
you
have
any
questions,
please
feel
free
to
reach
out.
Otherwise
I
will
send
out
that
email
shortly.