►
From YouTube: Kubernetes SIG Storage 20170622
Description
Kubernetes Storage Special-Interest-Group (SIG) Meeting - 22 June 2017
Meeting Notes/Agenda: https://docs.google.com/document/d/1-8KEG8AjAgKznS9NFm3qWqkGyCHmvU6HVl0sk5hwoAE/edit#heading=h.gnduln4s4ct
Find out more about the Storage SIG here: https://github.com/kubernetes/community/tree/master/sig-storage
Moderator: Saad Ali (Google)
Chat Log:
09:06:33 From SIG Storage : https://docs.google.com/document/d/1-8KEG8AjAgKznS9NFm3qWqkGyCHmvU6HVl0sk5hwoAE/edit#
09:45:06 From SIG Storage to Tushar Thole (VMware) (Privately) : Would you be wiling to have an agenda item next meeting with concrete suggestions for how we can improve collaboration?
09:45:44 From SIG Storage to Tushar Thole (VMware) (Privately) : I added a line to the end of the agenda doc for today’s meeting, we can move that to next week. I think it’s important to get this right as we continue to grow
09:51:47 From hekumar : this data is btw available as metrics saad
10:00:30 From Luis Pabón : Could that model be documented for storage plugin developers?
10:00:42 From Luis Pabón : The model used by StorageOS and secrets
10:04:13 From Simon Croome : I can write something up, but some of the API bits were quite new to me so I’d like to ask Jordan or Saad to review
A
All
right
welcome
everyone.
This
is
the
storage
sig
meeting
today
is
June
22
2017
as
a
reminder,
if
using
public
and
recorded
I'm
going
to
take
a
look
at
the
agenda,
the
biggest
thing
we
have
on
the
agenda
is
q1
q3,
V,
1.8
planning.
We
are
just
wrapping
up
1.7
and
there
are
a
couple
of
bugs
that
still
need
to
be
closed.
On
the
storage
side
of
things,
one
is
related
to
flex
volumes,
I
believe
the
pr
for
that
is
done.
It's
just
awaiting
merge.
A
B
Basically,
he
sent
out
he
sent
out
one
PR
to
put
the
reported
capacity
to
hide
it.
Unless
you
specify
the
Alpha
feature
gate
he
had
so
that
pier
has
been
merged,
but
then
it
ended
up
breaking
the
test,
for
it.
I
think
we
need
to.
He
miss
some
places
where
it
need
to
be
changed.
The
resource
name
needed
to
be
changed,
so
he
just
sent
out
another
PR
to
fix
that
I
think
last
night.
So
we're
reviewing
it
now.
B
C
A
Okay,
cool
another
remaining
issue
for
1.7
was
finishing
up
the
docks
and
it
looks
like
everybody
got
a
docks
PR
in
with
in
time
so
looks
like
we're
OK
on
that
end,
so
moving
forward
to
1.8.
If
you
open
up
the
agenda
dock,
we
have
a
section
called
the
wish
list,
as
well
as
the
planning
spreadsheet.
The
planning
spreadsheet
has
been
pre-populated
with
all
the
items
that
we
were
working
on
in
1.7
that
we
may
not
have
completed
yet
or
they
were
alpha
and
we
want
to
move
into
beta,
etc.
In
just
a
minute.
A
Matt
DeLeo
is
going
to
kick
off
talking
about
that
and
going
through
that.
But
if
there
is
anything
in
that
spreadsheet,
not
on
that
spreadsheet
that
you
think
we
should
be
working
on
for
this
quarter.
Please
add
it.
Under
the
wishlist
section
of
the
agenda
I'm
going
to
paste
the
agenda
doc
into
the
chat
window
and
with
that
I'm
going
to
hand
it
off
to
Matt.
Matt
is
our
sig
p.m.
A
D
Great
see
so
I
don't
think
we
need
to
go
over
the
one
seven
end
of
release
status.
Unless
people
want
to
do
that,
it
sounds
like
we
pre-populated
the
1.8
list,
so
I'm
going
to
go
ahead
and
share
my
screen
with
the
1.8
spreadsheet
and
we
can
get
started.
So
what
I
was
hoping
to
do
is
assigned
a
priority
and
kind
of
a
feature
owner
for
each
one
of
these
bugs
sorry,
each
one
of
these
items
and
for
I
think
a
lot
of
them
they've
been
on
here
for
a
while.
D
So
we
kind
of
know
the
what
they
are,
but
maybe
it
would
be
good
to
just
give
it
a
quick
10-second
description
for
the
people
that
know
about
them.
So
does
the
rest
of
the
community
can
understand
so
getting
started
with
this
first
one?
We
got
this
volume
plugin
metrics.
We
don't
have
an
owner
but
yeah,
who
knows
about
that
and
who
can
help
us
and
I.
A
The
app
here
is,
instead
of
or
in
addition
to
instrumenting
at
that
layer,
let's
instrument
a
layer
back,
which
is
the
actual
rapper
calls
for
the
volume
plug-in
and
soap
and
logic
here,
is
that
by
doing
so,
we
would
be
able
to
get
instrumentation
for
all
volume
plugins
instead
of
just
specific
cloud.
Metrics
api's
does
that
make
sense
to
folks
yep.
D
E
D
G
I
have
one
a
quick
question
and
that's
in
terms
of
breaking
out
the
cloud
provider
code,
so
obviously
I
think
it's
targeted
for
1-9
to
have
the
cloud
provider
be
totally
deprecated.
Is
there?
Is
there
any
lot
going
into
how
these
sorts
of
scraping
activities
are
going
to
happen?
What's
the
cloud
providers
completely
outside
the
offices
of
the
core
yeah.
A
That's
kind
of
one
of
the
reasons
we
want
to
do
this
is
that
we
don't
have
to
rely
on
that
functionality
anymore.
If
we
instrument
at
the
volume
plug-in
layer,
then
we're
going
to
get
volume
plug-in
related
metrics,
regardless
of
whether
it's
implemented,
if
the
clouds
plug-in
or
some
other
volume
plug-in,
and
we
can
pretty
much
deprecated
those
and
once
they
move
out
of
trees,
we
don't
really
have
to
worry
about
them.
Ok,.
E
The
focus
of
the
matrix,
the
cloud
wider
matrix,
the
API
matrix
of
different
asad,
so
the
matrix
was
like
we
were
running.
We
were
overshooting
the
a
terrific,
a
quotas
and
we
had
no
idea,
which
call
was
still
making
most,
like
example,
attach
call
makes
so
many
play
calls.
It
makes
like
five.
Eight
eight
calls
a
number
of
API
calls,
but
it
makes
it
cause
described
volume
like
a
No
ten
times.
E
A
Fair
I
think
to
Jason's
point.
There
is
an
effort
underway
to
try
to
decouple
the
cloud
metric.
Sorry,
the
cloud
code
from
the
the
generic
kubernetes
code
and
his
question
was
around
how
these
metrics
would
be
surfaced.
Once
back
code
was
decoupled
as
far
as
volume
volume
buttoned.
It
would
be
this
first
line
item,
but
to
what
a
mantra
was
mentioning
having
actual
instrumentation
for
the
further
cloud.
Api
calls
I
think
that
remains
to
be
seen
once
that
implementation
happens,
for
decoupling.
H
Right
this
is
a
t-shirt
from
the
embed
I.
Have
a
question
and
I
miss
some
of
the
power
stations?
Maybe
you
guys
covered
it
already.
Excuse
me,
if
that's
the
case,
so
so
what
I'm
trying
to
figure
out
is
so
we
also
work
with
him
and
quickly
add
the
support
for
this
case
as
you're
reading
chat.
So
we
had
a
question
that
you
know
why
do
every
volume
plug
in
or
whatever
float
garden
is
to
do
this?
So
why
not
do
this
in
the
book
when
it
is
itself
and
looks
like?
H
So
if
I
understand
correctly,
I
don't
know
the
timeline
for
when
the
volume
currency
more
for
free.
So
whenever
that
happens
at
that
point,
instrumentation
will
lean
from
volume
the
world
within
the
volume
to
order
to
outside
order,
so
that
whenever
the
cumulative
cell
calls
at
LG
fetch
is
that
correct?
So.
A
With
line
item
number
two,
we
want
to
move
it
move
it
out
of
the
individual
volume
plug-in
to
the
generic
layer
where
kubernetes
calls
through
to
the
volume
plugin.
So
it's
at
the
kubernetes
Blair
once
the
volume
plugins
move
out
of
tree
and
the
way
that
I
imagine
that
happening
is
a
new
volume
plug-in
that's
CSI,
it
so
happen
at
the
same
layer.
You
wouldn't
really
need
to
do
anything
special
in
order
to
handle
that
alright.
H
H
Yeah,
so
the
question
I
have
is
then,
because
we
want
to,
we
are
doing
right
now
for
this
visa
without
border,
we
are
doing
some
changes
and
I'm
just
going
to
figure
out.
We
want
to
continue
doing
that
before,
because
there
is
value
that
we
see
Indian
for
visas
also,
eventually,
this
will
move
to
the
as
part
of
si.
Si
really
have
a
better
architecture
for
that
right.
H
J
D
So,
let's
move
on
to
the
next
line
item
unless
there's
any
more
comments
on
that
one
I
think
this
one's
just
a
bug
fix
for
the
NFS
wedging
issue.
It's
been
here
for
a
while
I
would
say
it's
still
at
two
I,
don't
think
it's
a
one
I,
don't
think
anyone's
getting
bitten
by,
but
it's
something
we
think
we
should
probably
try
to
do
so.
I'm
going
to
go
ahead
market
this
it
sounds
like
we
have
an
owner.
A
So,
as
a
lot
of
you
are
aware,
last
quarter
we
came
up
with
a
preliminary
spec
for
what
an
out
of
tree
volume,
plug-in
interface
would
look
like,
and
we
try
to
align
that
interface
across
multiple
cloud
orchestration
systems,
including
through
Nettie's,
mesas
foundry
in
docker.
We
have
a
preliminary
doc
published
for
what
that
spec
should
look
like
and
now
the
plan
is,
let's
all
go
back
and
try
to
implement
this
preliminary
draft
of
an
interface
and
see
what
stumbling
blocks
we
run
into
on
and
we
can
use
that
to
revise
the
the
interface
spec.
A
A
That's
going
to
be
a
CSI
volume
plug-in
how
the
packaging
for
these
plugins
is
going
to
look
like
how
the
deployment
for
these
plugins
is
going
to
happen,
and
basically,
I
think
the
end
goal
here
would
be
to
come
up
with
a
design
doc
at
the
very
minimum
and
hopefully
a
prototype
as
well.
The
way
that
I
see
this
playing
out
based
on
our
previous
cadence
is
that
we
would
come
up
with
a
design,
doc,
slash
prototype
in
this
quarter
and
then
for
a
subsequent
release.
We
moved
to
alpha
then
beta
and
then
to
stable.
A
J
L
J
Our
kind
of
our
blocker
for
getting
things
out
the
door-
it
really
you
know
it's
frustrating
for
I,
think
everybody
so
I
think
at
a
working,
maybe
in
release
branch
for
prototyping,
something
in
that
side.
I
said
that
we
should
consider
at
the
same
time
since
we've
done
so
much
examination
of
this
API
in
the
CSI
and
the
CNCs
space.
I'm,
also
curious
if
we
can
maybe
accelerate
the
status
of
this
work
from
like
data
just
because
a
we
knew
you
and
be
it's
already
gotten
quite
a
lot
of
scrutiny.
M
Hey
guys
I
think
go
ahead.
Just
hear
that
I
think
I
wanted
to
talk
about
is
if
we
can
accelerate
the
design
doc
around
the
packaging
and
interfacing
with
CSI
from
kubernetes
that
would
that
would
help
figure.
You
know
help
us
decide
how
far
this
this
is
going
to
take
because
I'm
not
at
all
outside
and
stuff
of
CSI
yeah.
G
M
B
J
N
L
J
I
think
the
packaging
is
really
not
quite
as
CNCs
focus
as
a
difficulty.
Nettie's
problem
really
I
think
the
CAI
API
and
being
keypad
or
signal
and
maybe
decide
we
want
to
hosting,
but
I
think
how
we
package
and
how
we
just
we
get
kubernetes
a
super
nettings
challenge
and
not
CMC
have
challenges
much
and
I'm,
hoping
that
we
keep
it
focused
into
the
Betties.
J
The
end
of
the
day,
it's
going
to
be
the
people
in
this
group
right
now
that
are
going
to
have
to
deal
with
supporting,
executing
and
really
to
sort
of
how
we
interact
with
the
kubernetes
structure.
Overall,
this
is
not
a
problem
that
is
also
specific
to
just
the
storage
side,
so
I
think
the
the
that's.
J
Basically,
it
I
think
we
were
careful
when
we
were
discussing
this
in
CN
CF
to
leave
the
packaging
question
external
to
that,
because
it's
got
any
different
for
docker
or
may
doze
and
kubernetes
and
the
needs
and
the
execution
are
probably
going
to
be
very
specific.
So
we
we
should
probably
the
CSI
spec
sort
of
left
a
lot
of
these
doors
open,
and
we
should
be
focusing
I
think
here.
M
J
H
A
B
D
A
A
A
D
F
D
Local
persistent
storage,
oh
I,
think
this
one
is
pretty
self-evident
from
the
description,
but
it
adds
support
for
persistent
storage
on
locally
attached
discs.
I
think
this
is
going
to
be
a
p1.
It
sounds
like
you
already
have
owners,
it's
probably
populated
from
last
last
time,
but
I
think
probably
the
same
people
should
be
keep
it
moving
forward.
Yeah.
B
B
D
Okay,
maybe
I'll
clean
mark
it
as
an
X
and
we'll
try
to
opinion
if
we
can
yeah.
Okay,
yeah
testing
is
always
something
that's
always
going
to
be
here.
I
can
get
you
always
a
p1
yeah
I'm,
not
sure
that
has
like
a
target
status,
but
do
we
have
people
I
know
that
you've
had
a
lot
of
people
work
on
this
in
the
past.
Do
we
want
to
have
the
same
people
or
who
wants
to
sign
up
to
do
this.
A
O
And
in
in
addition
to
that,
we
had
a
conversation
which
I
brought
up
on
rustic
call
about
leveraging.
You
know
through
the
CNCs,
to
get
an
environment
oil
compressed
all
of
the
plugins.
It
came
up
on
a
call
I
just
had
with
Jeff
an
hour
ago
that
a
lot
of
the
plugins
aren't
being
tested
through
the
CI
right
now,
and
so
there's
some
pretty
serious
repercussions
of
that.
So
we
need
to
keep
driving
the
testing
to
be
able
to
stop.
J
K
J
I
J
O
I
kind
of
already
put
that
together
my
phone
or
wonderful
in
that
we
agreed
you
know
we
would
need
to
have
vendor
representation
for
each
one
of
these
plugins.
We
can't
just
have
people,
contribute
test
cases
and
then
go
away.
We
need
to
have
some
sort
of
governance
in
terms
of
what
we
test
and
who
represents
those,
so
that
has
to
get
flushed
out
from
each
one
of
the
areas
in
which
we're
contributing
plugins.
O
J
D
J
I
You're
online
right,
yes
and
I'm,
trying
to
do
a
status
on
this,
we're
straddling
170-180
with
it
it's
pretty
much
fully
external
and
I
know
there
are
some
concerns
about
it,
not
being
integrated
with
you
know
the
core
cube
constructs
and
where
those
touch
points
are
I,
so
yeah
I
don't
think
it's
necessarily
a
priority
for
the
release.
But
it's
something
that
we
want
to
have
a
prototype
super
then
later
and
then,
maybe
as
part
of
1/8,
we
figure
out
where
the
touch
points
are
with.
J
Else,
I
really
worry
that
we're
going
to
head
a
little
fast
at
this
point
with
out
knowing
enough
use
cases
and
integration
with,
like
you
said,
stateful
sets
and
some
of
the
other
like
figs
that
are
actually
using
storage
and
understand
state
and
the
need
of
this
and
kubernetes
so
it
you
know,
I
feel
like
Brad
and
I.
We've
talked
about
this
once
just
when
we're
talking
about
something
else.
J
I
want
to
make
sure
that
we
don't
end
up.
I
know
there
are
some
other
parts
of
kubernetes
of
late
where
we
wrote
a
lot
of
code
and
we
did
a
lot
of
investment
and
then,
as
we
got
closer
to
data,
we
said
oh
you're,
nuts,
it
really
you
know.
We
have
some
neat
tech
and
we've
sort
of
made
some
of
the
vendors
happy,
but
for
the
users
it
may
not
actually
be
as
great
an
experience
as
we
expected.
J
I
think
for
me,
like
the
difference
between
p1
and
p2,
is
that
if
the,
if
we
look
at
the
the
released
in
1/7-
and
we
say
oh
look,
the
P
ones
are
in
danger.
We
say
guess
what
we
need
to
take
people
off
peaches
and
start
focusing
on
T
ones,
and
the
difference
between
peaches
would
be
3.
R
P
3s
are
sort
of
nice
to
have,
and
you
know
we
should
start
thinking
about
it
and
put
some
attention
to
it.
But
as
a
sig,
you
know
we
don't
see
it
as
a
high
priority.
I
J
You
fight
it
with
live
snapshot,
keeping
the
state
of
applications
consistent.
You
know
before
and
after
an
upgrade
you're
going
to
get
it
wrong,
and
so
then
I
presented
the
you
know
the
reasons
why
I
think
this
is
valuable
and
the
scenarios
where
it
is
and
it's
kind
of
like
a
yeah.
Alright,
good
luck
with
that
and
I
guess
I
kind
of
want
to
know
more
convincingly.
You
know
what
is
the
the
Minimum
Viable
Product
and
what
is
the
actual?
You
know
what
how
are
people
going
to
use
this,
our
customers
problem?
N
J
That's
the
feedback
I
got
from
the
informal
conversations
I've
had
with
them,
which
is
you
know
why
I'm
kind
of
bringing
this
up
in
the
first
place.
So
if
they
don't
have
those
use
cases
where
are
they
coming
from
and
I
want
to
make
sure
that
it's
not
just
hey,
we
can
do
snapshots
and
they're
cool
I
understand
why
the
infrastructure
layer,
it's
really
cool
I,
want.
I
J
A
One
thing
I'd
like
to
add
is
we
need
to
start
thinking
about
the
quote:
unquote
the
kubernetes
way
of
doing
things.
Instead
of
thinking
about
storage,
features
and
saying,
okay,
there's
a
storage
feature,
it
doesn't
exist
in
the
kubernetes
api.
Let's
expose
it.
Let's
take
a
step
back
and,
let's
think
about
how
our
kubernetes
users
going
to
use
this
for
for
things
like
deployment
for
things
like
stateful
sex,
and
how
is
it
going
to
benefit
them
if
they
are
using
it,
for
you
know,
taking
snapshots
of
their
sharded
database?
A
What
is
the
actual
user
workflow
going
to
look
like
and
instead
of
designing
from
the
bottom
up
and
hoping
that
we
converge
to
a
solution
that
that'll
make
sense
to
our
end
users,
let's
start
from
the
top,
but
consider
the
use
cases
from
our
users
perspective
and
then
design
down
from
that
and
I
realized
that
a
lot
of
people
in
the
sig,
even
I,
thought
near
the
end
of
the
last
quarter.
That
snapshots
was
in
a
in
the
design
that
we
were
converging
on.
A
We
were
happy
with,
but
then
I
think
I
also
started,
realizing
that
what
we
came
up
with
was
essentially
taking
the
existing
the
paradigms
that
storage
systems
expose
around
snapshots
and
exposing
it
directly
through
the
kubernetes
api,
and
not
really
considering.
Is
that
really
the
best
that
kubernetes
can
do
to
expose
this
kind
of
functionality?
So
the
Android.
A
I
I
actually
think
that
we
took
that
direction
intentionally
because
we
were
overwhelmed
with
some
of
the
top-level
use
cases
to
a
point
that
we
couldn't
get
the
building
blocks
right.
And
so
we
tabled
some
of
that
stuff
as
phase
to
work.
So
we
could
get
the
core
pieces
in
and
then
we
wanted
to
like
the
second
phase
and
an.
J
I
J
Think
Brad
I
think
maybe
maybe
okay
I
think
this
is.
But
what
you're
hearing
from
me
is
more
of
this
is
why
I
think
it
should
be
a
2
which
is,
if
we
have
these
other
items,
this
quarter
coming
up
that
we,
you
know
we
get
them
round
at
that
point
where
things
are
tight
and
we
need
to
know
what
we
invest
in
and
we
need
to
make
sure
what
makes
it
for
1/8.
I
Sorry
I
think
I
actually
think
we're
on
the
same
page
like
we
we're
trying
to
get
the
you
know,
we're
still
finishing
up
our
one-seven
straddling
1/8
with
the
basics
on
snapshots.
I
think
that
the
phase
2
priorities
for
1/8
are
definitely
a
2
and
we
should
start
gathering
information
and
doing
design
talks
for
it,
but
implementations
off
the
table.
We
want
to
let
it
incubate
for
a
while
make
sure
the
core
constructs
are
correct.
H
Have
some
feedback
from
VM
Verity
in
Bergen
this?
So
maybe
we
are
missing
something
but
leave.
We
are
finding
it
hard
to
move
who
pitched
three
volumes
after
support
people
I
see
that
there
are
no
meetings
happening
for
I,
guess
two
to
three
week
and
I
may
be
completely
wrong.
So
correct
me
if
I'm
wrong,
but
so
that
is
what
we
see.
H
So
what
we
are
doing
is
we
are
basically
following
the
document
and
then
we
give
feedback,
and
then
we
get
a
feedback
that
oh,
this
summoned
does
not
make
any
sense,
because
the
document
has
changed.
So
basically
we
are
basically
running
behind.
You
know
a
moving
target,
so
so
what
we
actually
so
our
team
reached
out
who
I'm
not
sure,
jiaying
or
read
on
that
you
know.
H
Is
there
some
forum
meeting
or
whatever
right
it
doesn't
have
to
give
formal
meeting,
but
where
we
can
actually
see
that
this
is
a
version
of
the
spec
and
now
give
you
a
feedback,
because
now,
if
I
understand
the
description
that
you
Brad
and
you
guys
are
having
is
that
you
know
how
to
do
this
is
AB
speak
or
whatever?
Is
the
basic
unit
is,
we
are
finding
it
hard
to
even
want
to
go
to
the
basic
volumes
have
short
support
interrelated,
so
we.
I
Had
a
spreadsheet,
so
the
communication
has
been
on
the
sig
mailing
list.
We
did
meetings
for
a
while,
we'll
the
design
was
being
nailed
down
and
it
got
to
a
point
there.
You
know
we
considered
it
complete
for
the
first
phase
and
we
had
a
spreadsheet
with
like
tests
for
each
item
and
then
also
I
know
groups
are
meeting
on
the
slack
channel
to
discuss
development,
which
was
happening
a
for
3po.
It's
still
going
on
there
and
you
know
we're
we
kind
of
paused
like
design
discussions.
Well
the
development
for
that
first
phase
closed
out.
I
H
Yep
so
I
guess
this
is
back
so
so
yeah
we
have
young
late
to
the
party
I
acknowledge
that.
So
what
I'm
saying
is
that
you
know
if
there
decisions
have
been
finalized
and
I
would
assume
that
the
dog
hasn't
changed,
will
not
change
as
much
straight,
but
it
seems
to
be
changing
so,
and
so
that
is
where.
H
A
That
that's
good
feedback,
and
we
should
we
should
take
action
items
here
and
see
if
we
could
communicate
better.
Our
community
continues
to
grow,
larger
and
larger
I
think
we
have
over
five
participants
in
the
sig
right
now
and
making
sure
everybody
is
getting
all
the
information
they
need
is
important.
A
How
about
for
every
one
of
these
features,
keeping
a
running
a
Genda
kind
of
doc,
like
we
have
for
the
the
primary
meeting
doc,
where
every
time
there's
a
meeting
summarize
what
the
what
was
discussed
and
a
link
to
the
most
updated
version
of
the
spec,
and
so,
if
anybody
who
wasn't
able
to
attend
the
meeting
would
be
able
to
look
at
that
dot
and
then
link
that
dot
from
the
agenda.
Doc.
Is
that
something
everyone
can
commit
to
with
that
help?.
H
So
that
I
have
one
one
feedback
on
the
side,
so
just
to
clarify
my
request
rate.
What
I'm
saying
is
that
I'm
not
requesting
that
you
know
that
you
know
start
the
meeting
girls,
because
you're
late
explain
everything
again:
that's
not
what
I'm
saying
it.
What
I'm
suggesting
is
that
there
are
few
set
of
people
who
are
working
on
these
features,
who
are
more
knowledgeable.
They
have
the
background
they're
doing
the
right
job
right.
So
question
about
that
right
now.
H
The
dog
that
they're
working
on
is
the
one
that
we
see
right
and
that
dog
is
changing.
Who
the
simple
thing
could
be
that
you
know
if
the
dog
is
still
changing.
Maybe
we
can
have
two
separate
things
and
you
know
so
that
we
look
at
the
stable
version
and
you
know
you
compile
the
changes
that
you
have
done
during
the
week
and
then
you
see
you
look
at
the
next
version,
because
if
the
dog
is
missing
on
a
daily
basis,
it's
just
not
sustainable
for
us.
So.
I
I
We
went
to
custom
resource
definitions
away
from
their
party
resources
and
that's
some
of
the
flux,
that's
in
the
design
document.
Now
all
right,
maybe
we
can
say
that
any
issue
owner
or
on,
if
any
feature
your
owner
on
this
sheet
I,
you
could
email
them
directly
an
answer
status
or
how
to
get
plugged
into
this
I
mean
I'm
open
for
that.
You
know
if
there's
features
that
were
responsible
for
I,
don't
mind
people
reaching
out
directly
and
asking
where
we're
at
on
them
or
how
to
get
involved.
No.
H
A
E
Yeah
so
I've
opened
a
proposal,
request,
design
proposals
and
it's
not
get
help.
Tim
has
raised
some
points
about
open
coding.
The
resize
commands
that
that
plugin
will
execute.
So
that
is
something
that
the
proposal
needs
to
take
into
account
and
some
other
details
has
to
be
flex
out.
It
can
be
to
the
own
problem,
but
I
intend
I,
think
Express.
J
Please
also
make
sure
that
this
one
gets
Tim
Hawking
view
and
if
you
need
help
getting
the
attention
I
can
help
with
that.
I
think
he's
mentioned
that
it's
sort
of
a
drag.
It
changes
some
fundamental
assumptions
and
how
we
look
at
resources
and
objects,
and
it
also
may
touch
upon
nodes
figure,
resource
management.
So
it
seems
like
a
small
thing
from
storage,
but
it
may
have
ripples
that
we
need
to
consider
right
yeah.
We.
J
I
We
were
shooting
to
get
this
done
it
early
for
1/8
and
in
either
winter,
walk
with
him
and
make
sure
these
okay
with
where
we're
going
on
this.
So
it
is
a
high
priority
to
us.
Why
did
the
time
priority
with
the?
What
is
the
use
case
in
the
business
need?
Are
we
back?
Customers
asked
to
resize,
storage
and
I
mean
that's
something
like.
E
Well,
like
like
users
who
are
running
like
a
like
a
workload
on
blustered
or
something
they
can
resize
the
it.
The
photons
until
a
cluster
doesn't
image
the
qualifier
system
resize.
So
they
want
to
be
able
to
use
a
bigger
cluster
volume
and
they
can
just
do
that
using
one
command
actually,
but
if
the
volume
is
managed
by
kubernetes
it
becomes
and
the
end
user
doesn't
have
access
to
that
that
command.
Then
it
becomes
harder.
E
E
J
J
No
I
can't
I
apologize
for
cutting
you
off
and
just
know
that
we
have
so
many
more
to
do.
There's
a
more
general
comment:
I'm
going
to
send
an
email
to
the
cig
storage
in
general.
Just
about
the
the
difference
of
I.
Don't
I've
got
a
good
understanding
of
when
a
kubernetes
storage
administrator
boundary
is
versus
the
infrastructure,
storage,
administrator
and
I
think
it
would
help
us
a
lot
in
a
lot
of
these
items.
A
The
other
feedback
that
we
had
earlier
for
snapshots
I
think
applies
to
this
as
well,
where
we
should
look
at
features
not
just
hey.
There
exists
a
future
in
storage.
Let's
expose
that
through
kubernetes.
Let's
look
at
it
from
the
top
down
and
see
how
exactly
would
a
kubernetes
user
benefit
from
this?
How
would
the
workflow
look
like
fit
into
the
existing
paradigms?
I
know
that's
difficult
to
do
it's
hard
for
me
to
do,
but
I
think
it
will
be
worthwhile
for
our
users
I.
J
Think
the
youth
case
is
kind
of
clear,
more
clear,
at
least
from
my
end,
I
think.
What
stinks
is
that
we've
got
stayed
in
two
places.
You've
got
state
inside
the
infrastructure
that
says
how
big
the
volume
is,
and
then
we
also
have
state
inside
of
the
objects
that
live
in
@cd,
and
it's
sort
of
you
know
where.
Where
is
the
centralized
control
and
and
how
are
we
going
to
administer
it?
And
that
makes
it
hard.
E
Yeah
I
think
both
this
picha
doesn't
interact
with
a
bigger
system
like
states
will
sit
and
whatnot
like
to
buy
yeah.
So
I
have
yes,
but
I
agree
with
Mike
that
yet
we
need
to
there's
a
state
like
that's
why
we
have
on
PVC
to
free
the
status
that
size
and
spec
top
size,
because
these
two
states
represent
alike.
What
is
the
expection,
but
what
what's
the
size
of
the
device
so
yeah,
there's
like
I
mean
we
have
the
controller
to
basically
sync
them
up
and
make
sure
that
they
they
are
same
other
than
saying.
E
I
D
Let's
go
excited,
okay,
I
think
the
next
one
is
somewhat
related.
It's
the
dynamic
update
of
the
TV
in
PVC,
so
I
think
this
one
is
kind
of
needed
for
the
resource
sizing
of
the
volume
support.
Is
there
anything
else
we
want
to
do
on
this
one?
Do
we
want
to
mark
it
as
a
two
then
and
then
maybe
revisit.
M
D
I
So
we
talked
about
replicating
data,
possibly
against
or
between
zones
or
across
cloud
providers.
It's
been
somewhat
of
an
exploration
into
use,
cases
and
I
know
there's
a
few
people
that
have
taken
steps
that
are
different
times.
I
last
Serg
leading
this
effort,
but
I
haven't
seen
him
on
the
storage,
so
maybe
put
me
down
as
the
leader
on
this
and
I'll
take
it
from
there,
but
I
don't
think
it's
I.
J
Think
we
can
I,
don't
see
a
lot
of
activity
going
on
with
all
the
other
things
in
1/8
and
Brad,
also,
the
more
and
more
I'm
working
with
Federation
sig,
the
let's,
unless
I
think
this
is
a
puzzler
for
attacking
even
this
year
sure
this,
for
so
we
can
just
keep
a
low
priority
on
it.
Then
right
I.
Actually,
we
could
even
almost
take
it
off
the
list
for
now.
G
J
L
A
Folks,
we
got
9
minutes,
let's
drive
through
this
as
fast
as
you
can
next
item,
TVC
capacity
usage
stats.
This
is
a
attempt
at
improving
the
user
experience
around
the
PVC
object.
It
currently
tells
us
what
the
capacity
is
that
the
user
requested.
The
PV
object,
tells
you
what
the
capacity
of
the
actual
object
is.
We
don't
really
have
any
mechanism
to
tell
the
end-user
how
much
of
that
capacity
is
under
is
being
used.
This
is
a
low
priority.
I'd
say
2
or
3
is
okay
with
me.
I
D
A
Class
today,
parameters
are
all
set
by
the
cluster
administrator
for
power.
Users
would
really
like
to
give
the
cluster
administrator
the
ability
to
expose
some
of
these
parameters
to
the
end
user
and
let
them
select,
maybe
things
like
his
own
or
some
other
parameter.
Yon
had
a
proposal
on
this
last
quarter.
We
didn't
really
get
to
it.
I
think
it's
worth
doing,
it'll
be
a
value,
add
for
our
end
users.
It
might
make
things
like
controlling
zone
easier,
sounds
right
to
me
and
I
believe
yeah,
correct
owner,
for
this
go.
A
D
I
H
G
A
Slippery
good
they're,
improved
to
route
hack,
I
would
assign
this
to
jinguk
and
I
believe
she
may
have
already
completed
the
work
share,
I'm,
not
sure
whether
there's
an
additional
work
required
or
not,
but
we
can
sync
up
with
Jing
and
figure
out
necessary
for
everyone
else.
This
is
a
hack
that
exists
today
to
allow
mount
dependencies
to
be
injected
into
the
file
system
of
the
root
filesystem
and
right
now,
it's
very
constrained.
The
idea
here
would
be
to
make
it
a
little
bit
more
extensible,
but
ideally
it
should
be
replaced
by
number
15
soon.
A
J
D
A
Punted
from
that
order,
I
want
to
rank
this
very
highly,
because
I
think
it's
a
very
important
recovery
scenario
that
we're
not
handling
very
well.
That's
it
I,
don't
like
going
back
to
Michael's
the
difference
between
a
1
and
a
2.
If
you
know
it
falls
through,
are
we
going
to
pull
people
off
of
it?
I,
don't
think
so.
I
think
it's
a
2
is
valid,
even
though
I
want
to
rank
it
higher
and
all
except
the
person
working
on
this.
G
D
A
Is
a
proposal
and
a
PR?
That's
already
out
it's
considering
extending
the
host
path,
volume
plug-in
API
to
allow
mentioning
the
type
of
host
path,
whether
that's
a
a
directory,
a
socket,
and
the
idea
here
is
right.
Now
the
failure
scenario
for
host
path
is
that,
for
example,
if
you
try
to
mount
a
directory
that
doesn't
exist,
it'll
automatically
get
created,
and
that
may
not
be
what
users
intend
nice
to
have
I
would
call
it
a
two
or
three
I
forget
who
the
auto
store
is
with.
I
A
Think
this
is
an
improvement
on
the
existing
host
path,
volume
plug-in
sure,
I.
Think
a
lot
of
the
people
that
were
using
host
paths
are
probably
going
to
move
over
to
local
storage,
but
the
PR
exists.
There's
people
working
on
it,
I
think
Tim's
already
reviewed
it.
I
have
no
objection
for
it
going
in
and
I
don't
really
feel
strongly
towards
is
not
being
going
in
either
so.
A
D
A
D
A
L
I
just
created
well
I,
just
actually
the
follow-up
of
the
discussions,
but
under
Jordan
long
term
the
kids
had
on
those
reviews
of
slow
us.
So
what
happens?
Has
country
has
a
volume
source
and
for
this
and
one
source
a
mirror
image
in
the
50
objects,
and
they
don't
have
to
be
in
that
case
and
as
their
issues
of
that.
So
we
should
look
after
this
API,
it's
an
X,
so
they
are
referral
sources.
Also.
They
are
on
this
stage
and
it's
a
monosoff
so.
A
We
have
to
be
careful
here,
so
the
context
here
is
that
the
way
that
we've
been
adding
secrets
that
are
required
by
by
their
provisioning
or
attaching
or
mounting,
has
been
kind
of
haphazard.
We
haven't
really
gotten
the
security
team
involved
and
when
the
storage
OS
plugin
was
going
in
well,
luckily,
Jordan
Liggett
took
a
look
at
it
from
the
security
point
of
view
and
he's
like
hang
on.
A
We
need
to
pump
the
brakes
on
how
secrets
are
being
passed
around
here
and
the
solution
that
we
ended
up
going
with
was
that
we
made
the
precision
chromium
source
object
different
from
the
volume
source
object
because
for
administrators
who
create
the
TV
object,
have
more
privileges
than
some
of
the
references
is
released
from
a
pas
de
pree
provision
volume.
We
are
happy
with
the
way
that
it
landed
for
storage,
OS,
I
believe
so
I
mean
added
this
to
kind
of
revisit
this
for
the
other
volume
plugins.
A
A
D
D
L
L
L
L
D
A
Yeah
so
this
feature,
this
is
a
new
volume
plug-in
that
was
hoping
to
get
in
in
1.7.
They
missed
the
deadline
for
it
and
I
don't
have
any
problems
with
it
going
into
1.8.
They
already
have
a
PR
out
that
is
under
review.
There
are
some
questions
about
what
the
design
for
the
PR
should
look
like
it's
a
little
bit
different
than
some
of
the
other
volume
plugins,
and
that's
it.
Okay,.