►
From YouTube: Kubernetes SIG Storage 20170302 [Partial Recording]
Description
Kubernetes Storage Special-Interest-Group (SIG) Meeting - 2 March 2017
Partial Recording: First 40 minutes or so of the meeting was not recorded due to logistical issues.
Meeting Notes/Agenda: https://docs.google.com/document/d/1-8KEG8AjAgKznS9NFm3qWqkGyCHmvU6HVl0sk5hwoAE/edit#heading=h.dcbscdxsqbg
Find out more about the Storage SIG here: https://github.com/kubernetes/community/tree/master/sig-storage
Moderator: Michelle Au (Google)
Chat Log:
[None]
A
Class
and
I'm
just
I
can
create
my
regular
application
so
so
I
have
my
sequel
of
a
my
sequel
application.
Here
is
pretty
much.
You
know
it's
just
regular
by
neyo
mexico's
you
have
your
service.
You
have
a
persistent
claim
participant
in
plane
from
dash
tourist
class
that
I
created
and
you
have
your
your
application
itself.
A
So
let's
just
create
that
and
right
now,
if
I
do
a
kid
pbc,
you
see
that
it
created
my
sequel
volume
and
it's
bounded
and
if
I,
if
I
like,
if
I
do
this
here,
I
can
see
that
my
my
sequel
is
running
so
so
diary.
So
I
have
an
ax.
So
just
just
summarize,
I
you
know
create
a
fruit
cluster,
the
who
classic
create
the
group.
Sorry
I
have,
I
created
a
cooperator
who
particular
group
cluster.
A
Then
storage
array
was
created
and
then
I
created
an
application
that
consuming
by
using
purchasable
in
claim
that's
demo.
So
I
go
back
to
the
presentation.
So
what's
next
for
us.
So
what
next
that
we
want
to
create
a
route
bullying
plugin-
and
this
is
kind
of
to
force
our
message
that
not
only
we
want
to
be
part
of
a
canadian
only.
A
We
want
to
be
insulated,
but
we
also
wanted,
if
integrate
deeply
with
Cooper
natives,
and
I
feel
we
feel
that
by
having
a
problem
to
anybody
by
the
spine
control
for
work
and
become
more
seamless,
integrated
companies,
we
are
looking
at
features
like
cloning,
backup
snapshots
procedures
and,
as
most
other
services
up
with
the
companies.
We
also
want
to
integrate
to
monitoring
system
like
promises.
So
that's
something
that
we're
korean
looking
at
and
just
before
I
finish
you
know
I
will
introduce
this
route
project
q
con
that's
over
amber
in
seattle,
and
you
know
we.
A
We
got
a
lot
of
feedback
from
from
the
community.
We
and
we
took
those
feedback,
and
we
know
me
and
that
we
implemented
a
lot
of
that
it
wouldn't
work
right
now.
We
also
maintain
brick
consists
of
communication
with
folks
on
within
the
community
of
folks,
like
from
Korres
from
from
google
help
you
etc.
So
we
and
we're
growing
on
right
now.
Our
github,
we
have
dozens
more
than
one
toxicity
help
star.
A
So
so
we
do
have
a
lot
of
traction
right
now
and
people
following
us
and
we
have
developer,
who
are
deploying
brook
in
the
development
trying
out
and
giving
a
creeper.
So
so
this
is
really
good,
so
yeah.
So
if
you
have
any
question
I'm
here
to
answer,
I
also
have
some
folks
that
I
work
with
here
on
the
call
who
are
smarter
than
me
and
who
can
also
help
answer
questions
so.
B
A
A
So
in
witting
group
we
deploy
always
be
a
famous
SOA.
Be
note
that
you
have
that
you
being
part
of
the
classroom
and
that
can
be
done
with
labeling
it
would,
it
would
implement
it
will
bring
about
a
fee
and
then
basically
that
any
any
devices
or
any
hard
drives
on
that
node
will
become
part
of
the
of
the
of
the
fruit
platter.
Did
that
is
that
your
question
in
the
answer
is
so.
D
I
do
this
is
travis
yummy,
take
this
charge,
yeah
yeah!
So
we're
right
now
we
you
know,
and
we
mount
remount
down
flash
dad.
So
we
can,
you
know,
have
access
to
the
volumes
that
are
there
and
today
we
just
automatically
see
what
bonds
are
available
and
if
it
doesn't
look
like
their
needs,
we
use
them,
but
were
one
of
the
next
features
we're
working
on
to
make
it
flexible?
E
Hi,
this
is
the
notes
and
portals
had
a
couple
of
questions.
The
first
is
actually,
how
do
you
explore
block
storage?
Is
there
a
kernel
module
that
is
in
bulk,
or
what
kind
of
protocol
d
is
to
actually
have
a
set
of
export
block
storage?
The
second
was
that
it's
completely
unrelated,
but
it
says
I
notice
that
there
are
some
notes
that
show
up
as
Stockmann
and
some
notes
show
up
as
chef
mdf.
How
are
these
notes
picked?
E
E
D
Yes,
otal,
the
metadata
is
stored
in
corona.
These
resources,
such
as
secrets
and
config
maps,
doesn't
reach
and
when,
when
the
pot
starts,
it
will
look,
we
get
okay,
get
those
resources
from
those
communities,
because
Cooper
daddy's,
metadata,
really
I,
see
yeah
and
it's
still
early
days.
So
there's
there's
more
of
that
type
of
work
to
do
to
make
sure
that
all
the
failover
works.
But
that's
the
plan.
F
That's
a
currently
within
the
demo
that
you
just
saw
from
speed.
That's
that's
using
the
rbd
plugin.
That's
do
the
entry
plugin
that's
already
available
there.
It
then
we
saw
a
bit
of
the
experience
around
that
we
r
Stevie
needed
to
give
it
a
certain
data
like
monitor
IP
addresses
now
I'm
gonna
need
some
unis
FX
credentials
to
be
able
to.
You
know
for
that
kernel
module
to
be
able
to
talk
to
the
South,
luster
and
surface
up
that
storage,
and
you
know,
block
the
Vice
format.
Tell
them
so
they
do.
F
F
A
C
G
Have
something
so
with
the
first
volume
aerospace
being
developed?
Is
there
hypothetically
if
you
are
designing
a
volume
plugin
starting
today?
Is
there
a
reason
to
use
the
entry
versus
ease
of
flex
volume
like
what
what
are
the
advantages
and
disadvantages
there
and
are
the
entry
volume
plugins
eventually
going
to
be
moved
to
the
flex
volume
NSA?
Is
that
a
plan
or
hostility
yeah
in.
H
The
future,
so
the
plan
is
to
keep
as
much
stuff
out
of
trees
possible
as
kind
of
unfortunate
that
any
entry
plugin
suctions
about
300
different
files
and
recorders
for
fifteen
dollars
on
lines
of
code.
So
it's
a
first
of
all,
it's
kind
of
a
nightmare
being
being
free.
You
know
getting
a
new
plug-in
reviewed
and
actually
committed
in
trees,
and
then
we
have
stabilization
issues
and
testing
issues
that
come
along
with
it.
So
you
know,
kind
of
our
long-term
strategy
is
to
keep
stuff
out
of
three
impossible.
H
You
can
flex
and
right
now
we're
I,
don't
know.
Maybe
ninety
percent
I
know
feature
complete
I'm
flex
compared
to
the
entry
stuff.
We
are
shocked
where
you've
done
a
lot
of
work
for
16
to
have
the
attached
detach
controller
working
with
a
what
phone
plug
in
and
score
was
just
a
note,
so
you
shouldn't
very
be
able
to
implement
anything
with
flex
that
you
could
do
entry
now.
Would
you
come
across
bugs
we've
had
a
couple
requests
for
attached,
detach
to
include
pod
or
namespace,
along
with
it
on
a
namespace
information.
H
H
You
have
questions
while
you're
working
out
feel
free
to
reach
out,
but
we,
you
know,
we
a
couple
of
people
here-
have
done
flex
implementation,
so
we
night
we're
pretty
close
to
the
code,
so
we're
not
really
sure
where
our
documentation
pain
points
are
where
you
know
where
some
of
the
complications
are
at.
So
it
would
be
output.
You
know
well,
you're
doing
any
plan
for
my
station.
Ask
questions
so
we
know
what
the
dr
manner
you
know
we're
like
our
whole
trip.
Sure.
G
I
am
I
I
gotten
this
entry
Ladia
that
entry
plug-in
working
that
are
our
issues
specifically,
is
that
we
are.
We
are
a
smaller
company
by
headcount
and
plus
all
user
interface.
Is
that
it's
a
simple
contract
and
if
I
move
that
the
project
or
leave
the
next
wish
that
comes
in
and
work
on,
it
is
going
to
happen
easier
time,
updating
it
or
understanding
the
API.
Where
and
that's
why
I'd
like
to
switch
over
to
using
the
selected
essays,
but
that's
basically
it
I
mean.
I
H
I
don't
know
the
nature
of
your
the
file
system
you're
trying
to
connect
to,
but
we
do
have
an
external
storage
repository
now
where
we
can
have
is
external
parishioners
or
the
Community
Interest
flex
volume
plugins.
So
if
you
have
like
an
open
source
or
a
public
or
some
some
kind
of
file
system
type,
it's
you
know
we're
available
to
the
community
and
other
people
might
find
interesting.
You
also
have
the
option
you're.
Putting
a
more
public
refer
to
well.
G
Sure
this
is
done.
This
is
going
to
be
a
bought,
it
is
a
DVD
or
the
new
version,
divot
eee
ippity
9,
and
it
managed
that
utility.
That's.
The
business
is
for
sodas,
so
QT
9
and
gives
a
marriage
which
is
going
to
be
the
nsa's
that
the
volume
point
needs
a
spot
is
near
buddy,
that's
all
open
source.
What
volume
is
not
as
of
yet
in
a
public
place
we
can
to
have
a
good
just
as
a
policy.
We
seem
to
have
a
good
implementation
out
there
that
it's
stable
before
it's.
H
K
We're
kind
of
in
this
funky,
like
we
talked
about
a
filmic
w
fig
meeting,
and
especially
the
cig
face-to-face
meetings,
we're
in
this
halfway
transition
point
and
their
pros
and
cons,
whether
your
provider
or
whether
we
as
Cooper
Nettie,
maintain
errs.
You
know
that
we
all
want
to
get
the
Flex
out
of
tree
as
soon
as
possible,
we're
not
there
and
if
you're
a
provider.
K
There
is
pros
to
have
it
yourself,
entry
just
for
the
reason
with
Assad
just
said,
but
as
the
maintain
there's,
we
really
in
trying
to
restrict
what
can
be
done
entry
because
it's
hard
to
maintain
or
exactly
what
you
just
said
is
a
provider
you're,
a
small
company
and
they're,
not
sure
about
you,
know
the
maintenance
of
burden
on
yourself.
You
know
that
that
these
are
the
sort
of
things
that
we
think
about
when
we
are
deciding
what
one
entry
to
accept.
K
You
know,
as
as
the
provider
they're
kind
of
balancing
the
I
want
to
ensure
compatibility
with
Cooper
Nettie's
as
much
as
possible.
But
then
again
there
is
the
sort
of
place
at
work
that
you'll
be
doing
now
to
get
in
that
early
access.
That,
as
you
go
to
flex
you're
going
to
undo
there's
like
double
work
that
has
to
be
done.
The
transition
period,
unfortunately,
is
probably
going
to
go
on
at
least
a
couple
more
quarters,
but
it's
where
we're
at
I
think
we
see
the
light
at
the
end
of
the
tunnel.
K
G
K
You
will
have
to
do
that.
That's
the
right
concern
rate,
that's
a
17,
all
right,
I,
a
certain
value,
call
on
your
side
and
then,
as
God
was
bringing
up
the
case
call
also
on
the
equipment.
You
think
inner
side.
For
how
many
of
these
we
want
to
keep
putting
into
the
tree,
because
you
know
what's
relationship,
we
have
with
them
too.
B
J
Are
the
interface
is
looking
very
similar?
We've
got
attached
detached
mount
unmount,
which
work
on
publish
on,
publish
and
sell
and
create
delete,
we're
unsure
of
what
the
backwards
compatibility
implications
are
going
to
be
between
the
current
iteration
of
flex,
and
that,
once
we
get
a
more
formal
implementation
will
have
a
better
idea.
J
L
J
Don't
know
yet
it
could
be
a
iteration
of
flex
where
we
take
this
new
interface
and
we
more
flex
into
to
basically
comply
with
that
interface
and
which
case
it
would
basically
be
an
iteration
of
flex
if
we
can
do
that
without
breaking
existing
drivers.
That's
what
will
be
the
past
that
we'd
want
to
go
down
on.
It's
also
possible
that
we
may
create
a
separate
I,
don't
know
if
you
want
to
call
it
flex
two
or
something
and
keep
the
existing
iteration
of
flex
and
keep
them
both
side
by
side.
J
J
K
L
J
Yeah
so
I
think
you're
you're
getting
at
whether
you
can
have
multiple
provisioners
for
the
same
plug
in
essentially
I
think
we
kind
of
had
this.
This
tension
in
the
existing
plug-in
system,
right
where
you
cannot
how
Kapoor
provisioners,
because
provisioners
or
it
could
be
external
from
the
driver,
plugging
themselves,
yep.
L
J
H
A
group
of
concept,
external
provisioner,
it
uses
the
same.
A
lot
would
strip
that
flex
with
you.
So
basically
called
a
shell
script
with
provision
than
Michelle
Turkish
belief.
Ok,
you
know,
could
be
read
as
a
stopgap,
and
so
we
figure
out
exactly
whether
you
get
caught
in
the
flex
line,
plugin
or
cool
on
you.
The
actual
formation
place.
H
J
So
I
like
if
the
plugin
you
know,
exposes
an
arbitrary
set
of
parameters
and
those
parameters
can
be
set
by
storage
classes.
Then
I
don't
see
the
need
for
more
than
one
positioner
per
volume
provider
and
you
could
still
have
the
flexibility
of
having
exposing
different
types
of
storage
through
storage
classes.
That
all
of
that
is
that
is
already
built
into
Cooper
Nettie's,
but
I
wonder
if
there
is
really
a
need
to
have
multiple
provisioners
for
a
single
volume.
Plugin.
H
G
J
N
My
understanding
was
that
the
move
was
toward
out
of
tree
and
I
didn't
really
know
how
flex
it
into
that.
But
it
sounded
like
Brad
said
that
flexes
is
the
preferred
approach,
but
at
a
high
level
the
my
notes
there
or
the
reason
I
added
something
to
the
agenda
listed.
This
is
to
get
a
consensus,
or
at
least
a
feeling
from
the
community
on
the
urgency
of
proceeding
with
an
e
2e
framework
that
works
for
flex
volumes
and
or
out
of
tree,
so
that
sword
providers
have
it's
easy
for
them
to
create
and
end
test.
N
There
needs
to
be
a
way
to
get
those
tests
run
and
hopefully
automated
in
CI,
but
but
I
thought
it
would
be
important
to
have
a
good
framework
established
so
that
it's
easy
to
plug
in
new
storage
providers
test
cases
to
ensure
that
you
know
that
their
storage
works
as
expected,
and
there's
no
regressions,
as
as
things
move
forward,
that's
sort
of
it
at
a
high
level.
We
have.
B
N
J
Cool
this
is
exactly
the
kind
of
discussions
we
should
be
having
on.
We
should
be
having
them
hand
in
hand
with
the
actual
design
of
the
external
provisioner
work.
That
way,
we
you
know,
testing
is
not
an
afterthought.
Testing
and
certification
should
be
something
that
we
think
about
throughout
the
process
on
glad
you
brought
it
up,
so
the
design
is
where
I
think
we're
going
to
look
spoken
it
up
to
the
broader
community
in
the
next
few
weeks
and
Jeff
and
Arin.
J
O
J
So
code
freeze
for
1.6
was
earlier
this
week
on
Monday
and
since
this
didn't
make
it
it's
not
going
to
be
part
of
1.6
the
next
release.
1.7
is
you're.
Welcome
to
try
to
be
part
of
that
code.
Freeze
is
going
to
be
lifted,
probably
in
about
three
weeks
or
the
main
branch
at
which
point.
If
you
want
to
try
to
get
this
in,
that's
perfectly
fine,
but
during
that
period,
even
if
this
gets
approved
it
will
not
be
merged,
know.
P
H
I
asked
me
to
push
all
right:
I'm
talking
go
for
it
I
some
questions
around
some
of
the
relief,
so
we
had
a
fairly
substantial
full
request
for
some
rework
of
the
attached
detached
controller.
Basically,
the
bug
is:
if
the
attached
detached
controller,
it
goes
down
and
then
comes
back
up.
It
doesn't
remember
its
previous
state
and
things
facially,
so
crazy,
I,
the
change
was
pretty
BB
and
we
didn't
make
it
in
by
the
feature
freeze,
unfortunately,
I
but
I'm.
H
You
know
it's
actually
enough
of
a
bug
that
I'm
wondering
if
we
should
try
to
get
this
as
a
bug
thing
and
it
is
kind
of
technically
a
bug
fix.
But
you
know
at
the
same
time,
I
we
had
some
concerns
on
our
side.
Babe.
You
know
you
guys
were
somewhat
bsme
I'm
get
to
actually
touch
controller
and
me
you
know
we
could
reach,
maybe
a
seventy
five.
Eighty
percent
confidence
honestly.
So
how
encapsulated.
K
From
pod,
so
here's
my
rule
of
thumb
that
I
so
but
I'm
going
to
tell
you
what
I
tell
the
people
on
my
team
right
now
about
by
the
way
I'm
the
transparent.
So
the
first
question
I've
got
is
how
encapsulate
it
is.
It
is
something
where,
if
we
fix
this
bug,
what
is
the
because
we've
been
soaking
this
code
for
so
long
I'll
repo?
K
What
is
the
chance
that
we
could
have
adverse
side
effects
outside
of
this
area
and
then
the
other
one
is:
how
do
we
have
confidence
that
what
we
did
in
this
drug
six
actually
doesn't
introduce
any
more
new
problems,
even
within
this
area?
How
good
is
the
testing
in
this
area
and
when?
Lastly,
what
is
the
impact
of
not
doing
it?
Okay.
H
So
the
impact
is
whenever
I
mean
we
thought
we
usually
get
about
two
or
three
bugs
for
release
that
are
related
to
this.
Someone's
nose
goes
down,
then
it
comes
back
up
and
none
of
their
volumes
or
catch
the
right
thing
and
they
active
mainly
go
and
kill
every
thing
and
like
bring
it
back
up.
It's
somewhat
of
a
gnarly
bug.
H
As
far
as
like
the
confidence
on
the
code,
we
you
know
first
off,
we
would
like
you
know
to
take
it
like
a
real
deep
look
at
it,
make
sure
everything
oops
in
checks
out,
but
secondly,
we'd
also
like
to
see
some
capacity
testing
on
it
load
testing
make
sure
they
work
well
under
a
heavy
load
and
that
we
haven't
introduced
you
deadlox
or
race
condition.
But
beyond
that
I
mean
it
does
touch
you
catch.
That's
been
gorgeous
reports,
amazing.
The
initial
excitement
without
risk,
so.
K
Here's
my
thinking
of
it
first
off,
as
you
know,
thought
is
one
of
my
favorite
people
on
the
planet.
Yeah
I
also
was
quite
happy
to
see
that
Young
has
been
really
gaining
a
lot
of
trust,
also
in
King
and
her
knowledge
of
the
attached
to
touch
controller,
and
so
I
want
to
make
sure
that
we
don't
know
either
overboard
inside
with
all
discussion
that
we
also
spread
the
knowledge
as
much
as
possible
and
ayana
mostest
passes.
He
won't
do.
Is
he
the
author
of
this
particular
brand
or
even
know
we
had
too.
H
K
It's
that
fundamental
I
guess
I
have
two
questions
one.
It
feels
to
me
like
this
is
something
that
we
could
do
you
add
more
tests,
and
especially
at
scale
and
possibly
think
of
a
candidate
for
a
patch
release.
If
you
know,
basically
it
is
my
thinking.
I
have
more
testing
so
that
we're
more
confident
with
it,
because
I
don't
think
we
going
to
get
that
confident
in
the
next.
K
You
know
two
weeks
because
you
know
after
two
weeks
from
now,
because
the
whole
thing
is
supposed
to
basically
freeze,
you
know
really
really
feeling
like
martin
is
0.
Second
right,
so
you
know
a
week
before
hands.
No,
no,
no
the
right
mind
will
let
us
do
this.
I'm
already
frozen
well,
it's
frozen,
but
you
know
for
bug
fixes
right.
There
sounds
like
this
is
one
where
we
won't
have
a
confidence
ourselves
to
ask
about
it
in
at
least
a
week
and
then
probably
a
week
after
that,
you
know:
you'd
have
time
to
review
it.
K
Everything
else
so
I
feel
like
this.
One
won't
make
160
I,
do
wonder
if
it's
something
that
and
given
enough
confidence
in
research
and
effort.
Consider
for
a
patch
release,
because
we
have
done
things
and
patch
releases
that
are
based
on
stability,
even
if
they're
something
big
like
this
in
the
past.
You
know,
especially
during
the
13
days,
the
one
not
for
and
I
think
we
justify
that
by
showing
why
we
were
confident
that
it
was
a
good
cause,
many
problems
and
to
me
this
one's
a
good
good
candidate,
for
something
like
that.
K
H
H
K
Right,
well,
it's
not.
Functionality
is
unfinished
right
right
and
that's
exactly
what
the
patch
releases
are
for,
and
it's
kind
of
cool
of
reiterating
this
point
over
and
over
again
to
people.
This
is
what
I
love
is
that
coo
pernetti's
is
really
getting
to
get
to
this
point
where
you
know
we
have
trains
that
leave
the
station
and
push
patches.
We
have
trains
that
leave.
We
really
need
to
get
some
work
so
that
this
to
me
feels
like
a
good
candidate
for
it.
Actually,
you
know,
that's
it.
Okay,.
H
Cool
I'll
start
working
on
that
strategy.
Then
I
had
one
more
question.
We
had
a
feature
that
was
a
shot
I
remember
was
feature
was
it
was
either
the
blue
storage,
west
GA
or
the
mount
options,
and
we
opened
up
on
the
feature
repository.
You
know
with
all
the
necessary
checklist,
but
something
came
along
and
closely.
We
weren't
sure
why
you
got
clothes
or
to
know
who
it
was
a
picture
for
mount
options.
So
you
know
we,
you
know,
there's
a
long
process,
okay,
featuring
and
yeah,
let's
check
with
since
we.
J
H
K
K
And
scream
if
it
was
really
supposed
to
be
closed,
so
Brad
I
really
really
really
want
to
get
mount
options
in
16.
Okay,
and
my
first
question
is:
is
it
merged
into
the
16
treat
I
just
assumed
that,
yes,
let's
partying,
so
it's
in,
so
this
is
more
about
paperwork
on
the
future
repo.
Exactly
it's
just
your
paperwork
on!
It
got
weird
all
right.
So
please,
if
you
don't
make
progress
with
this
and
you
need
any
help.
Feel
free
to
drag
me
into
this.
K
H
K
H
K
Q
Along
here
we
we
need
a
little
bit
of
leeway
to
secure
the
facility,
but
we
have
that
corporate
briefing
center
in
santa
clara,
so
we've
offered
to
host
it
there
during
the
first
week
of
april.
I
think
we've
been
what
about
that
specific
Korean
off
and
on
oh
I'm
nominating
the
wednesday
and
thursday,
but
it
probably
smell
during
the
couple
days
during
that
week,
I
could
get.
Could.
P
J
Q
Q
Q
H
K
Q
J
That'll
be
cute
con
tu
week,
so
that's
not
going
to
work.
Let's,
let's
just
say
first
week,
April
I'll
see
if
I
can
dial
in
otherwise,
let's
not
block
on
that
and
we
can
take
it
offline
Steve.
Do
you
want
to
send
an
email
to
Detroit
big
as
first
because
it
suppose
doing?
Okay,
okay,
I'll?
Do
that
cool
and
feel
free
to
get
an
agenda
started
as
well
similar
to
last
time?
Okay,
thank
you
for
volunteering,
a
safe
again,
very!
It's
awesome!
Thank
you!
That's.