►
From YouTube: Kubernetes SIG Storage 20170914
Description
Kubernetes Storage Special-Interest-Group (SIG) Meeting - 14 September 2017
Meeting Notes/Agenda: https://docs.google.com/document/d/1-8KEG8AjAgKznS9NFm3qWqkGyCHmvU6HVl0sk5hwoAE/edit#heading=h.srjrt7mxhlkd
Find out more about the Storage SIG here: https://github.com/kubernetes/community/tree/master/sig-storage
Moderator: Michelle Au (Google)
Chat Log:
09:33:11 From hekumar : https://docs.google.com/spreadsheets/d/1AFksRDgAt6BGA3OjRNIiO3IyKmA-GU7CXaxbihy48ns/edit?usp=sharing
10:02:21 From eparis : she might be magic....
B
First,
up
is
CSI
volume.
Plugin
I
have
a
design
doc
out
that
I'm
going
to
send
out
the
plan,
for
this
is
going
to
be
a
new
volume
plugin
similar
to
the
Flex.
The
interesting
bit
here
is
going
to
be
how
kubernetes
is
going
to
communicate
with
the
volume
plugin.
The
proposed
plan,
I'm
thinking.
Oh,
if
at
the
moment
is
a
sidecar
container,
along
with
a
volume
plug-in
that
acts
as
a
proxy,
that
sidecar
container
would
drop
a
UNIX
domain
socket
onto
the
host
machine
through
which
cubelet
can
communicate
with
the
volume
plug-in.
B
B
C
B
D
B
C
B
B
E
So
1/8
is
still
artha.
So,
like
the
items
we
plan
on
1/8
or
finished,
including
Kota
support
for
local
ephemeral
storage,
let
me
arrange
in
downward
API,
so
together
with
1/7,
so
we
have
a
pretty
good
isolation
mechanism
available
for
local
ephemeral
storage
and
we're
target
beta
on
1/9
and
two-way
continue
to
work.
Some
items,
including
to
use
cape
cod
o
to
show
locally
thermal
storage
like
CPU
and
memory.
Some
items
I
think
make
is
start
working
on.
B
E
So
the
documentation-
yes,
I,
think
what
we
need
to
you
mean
mm-hmm
when
it
is
like
I/o
documentation,
yeah.
F
E
A
So
for
a
1/8
we
didn't
have
any
major
covered
deliverables,
just
a
few
fixes
and
more
tests
here
and
there
most
of
1/8
was
spent
working
on
prototyping
and
design
for
both
the
scheduler
and
binding
integration,
as
well
as
support
for
block
devices.
So
we
have
some
design
proposals
out
and
we
can
go
over
them.
I
put
them
on
later
in
the
agenda,
so
we
can
go
over
them.
Then.
A
C
A
A
G
So
the
design
is
close
to
being
merged,
I
hope,
Brad
brought
up.
One
last
comment
yesterday
and
Brad
can
elaborate
on
that
I'm
kind
of
not
understanding
this
piece,
how
it's
any
different
than
how
we
currently
do
file
systems
so
I
likes
and
then
put
there
I
tagged
you
on
that
Michelle,
otherwise,
I
feel
like
that's
to
a
point.
We
should
probably
merge
it,
and
if
we
have
implementation
details,
we
should
take
those
into
the
PRS
because
not
merging
this
is
blocking
everything,
including
the
merging
of
the
API,
which
is
blocking
everything
else.
So.
G
I
want
to
get
this
merged
early
I,
guess
we're
kind
of
in
a
code
freeze
so
that
we
can
get
it
done
and
I
don't
know
we're
just
in
a
weird
spot
where
we're
gonna
have
to
be
in
rebase
hell,
I,
guess
regardless,
because
everything
depends
on
the
API
PR,
which
depends
on
the
design.
T
are
being
merged,
otherwise
we're
just
full
steam
ahead,
I,
don't
foresee
any
serious
issues
and
less
Brad.
You
want
to
talk
about
the
the
use
case
you
brought
up
yesterday.
G
B
C
B
B
C
B
C
Yours
hang
up
on
this
side,
like
you
know,
when
you've
had
some
trouble
getting
some
of
the
protein
being
said
mode
and
getting
other
attention
on
this
I
mean
quite
frank
for
dropping
some
of
the
urgency
on
our
side,
because
it's
not
product,
but
we
still
see
the
need
for
end-to-end
tests.
I
I
was
wondering,
are
hoping
it
get
someone
in
Google,
you
know
maybe.
B
J
B
G
Sorry
so
I'm
working
on
creating
some
helm
charts
to
be
able
to
do
the
testing.
That's
how
the
cross
cloud
does
their
builds
to
be
able
to
do
it
on
the
different
platforms,
and
that
would
be
for
AWS
NGC
initially,
because
those
are
ones
that
are
in
the
pipeline
to
augment
persistent
storage
and
those
is
an
initial
faith.
What
we're
doing
with
the
end
goal,
eventually
having
you
know
like
a
test
harness
that
people
can
use
to
certify
on
their.
G
G
B
G
I'm
at
the
mercy
of
the
CNC
F-
and
you
know
a
bunch
of
other
teams
and
providers
as
to
what
we
can
do-
I
mean
I
I
can
make
goals.
It's
it's
not
really
like
an
effort
that
I
fully
own
myself
and
the
CNC
F
is
still
trying
to
get
their
feet
underneath
them.
In
my
opinion,
as
far
as
having
an
instant
process
but
I'm
happy
to
provide
updates
well,.
C
G
Don't
think
the
reason
why
we
did
this
is
because
the
e2e
tests
we
were
tracking
at
a
release
level
and
we're
not
doing
that
anymore,
because
we
understand
that
it's
going
to
be
futile
because
we
can't
actually
test
the
plugins
within
the
current
environment.
Therefore,
I
mean
I'm
neither
provide
updates,
they
aren't
going
to
go
to
any
sort
of
release,
or
we
can
just
remove
this
all
together.
I
think
those
would
be
the
two
alternatives.
C
G
Mean
no
one
shows
up
at
the
meeting
and
I'm
not
fully
going
to
continue
having
my
team
solely
do
all
the
end
tests
when
there's
so
many
other
things
broken
in
the
test
environment
that
cannot
be
fixed
in
the
way
that
it's
set
up
now.
It
doesn't
make
sense
to
continue
to
invest
in
that
effort.
Okay,.
H
B
So,
regarding
the
second
item,
the
end-to-end
testing
Aaron's
team
did
a
huge,
huge
effort
to
bring
storage
dig
up
to
par
for
making
sure
that
we
have
enough
end-to-end
testing
that
we
don't
have
regressions.
It
was
filling
in
a
lot
of
technical
debt
that
we
had
at
this
point.
I
think
what
we
need
to
do
is
start
enforcing
for
all
new
features
that
n
10
tests
are
written
along
with
the
the
features
for
a
couple
of
the
features
that
went
in
for
this
milestone.
B
G
B
Makes
sense
and
for
the
CNC
of
testing
sounds
like
we're
blocked
from
there
and
so
we'll
stop
well
we'll
see
if
we
need
to
continue
to
track
this
for
the
next
quarter
or
not,
we
can
revisit
that
in
the
next
meeting.
Okay,
all
right
I
need
to
jump
off
in
five
minutes,
but
I'll
see
how
much
of
this
I
can
get
through
and
then
Michelle
will
take
over.
So
for
the
green
stuff
we
can
skip
we'll
go
to
number
ten
support
attach
detach
for
our
wo
woman.
Are
you
on
the
line.
D
B
E
B
J
Yeah,
sorry,
the
the
we
did.
A
large
chunk
of
that
work
is
went
in
1.8,
the
controller
resize,
so
1.9
do
we
want
to
target
the
path
assembly
should
bring
in
support
for
resizing
GC
ADA,
where
such
kind
of
volume
types
it's
like,
not
useful
change,
but
there's
like
part.
We
done
positive
work
in
progress.
J
A
F
B
A
L
M
M
M
M
J
A
C
B
N
A
O
A
N
J
N
N
J
So
alright,
so,
unfortunately,
I
kind
of
need
sad
for
those.
So
it's
basically
we
have.
We
have
some
protection
in
place
and
in
detail
and
attached.
It
is
controller
which,
which
stays
and
attaches
a
volume,
if
part
is
deleted
for
six
minutes
actually
and
I
have
Jing
on
his
convinced,
I'm
convinced
and
I've
gone
through
the
port
through
several
times,
but
myself
into
tiannan.
We
have
walked
through
the
protection
that
exists
serves
any
purpose
we,
our
conclusion,
is
it
doesn't
help
so
the
plan
is
I
have
opened
a
pair
to
remove
that
check.
J
J
Okay,
yeah,
so
I
just
thought
it's
kind
of
fun.
It's
like
we
don't
achieve
anything,
but
by
blocking
the
discharge
operation-
and
it
is,
it
is
basically
affecting
well
use
us
a
little
bit
like
because
attached
and
attached
together,
like
12
minutes
for
no
good
reason
like
if
detach
itself
takes
6
minutes
and
signals
plus
change,
and
then
the
next
operation
will
not
start
until
that.
Until
that
completes
is
like
an
exponentially
exponential
back
off
naturally
increases
the
time
right.
J
Q
Dot
o
alpha,
there
has
been
some
weird
output,
seen
on
new
cutter
land,
cubelet
binaries
and
when
you
run
those
so
it's
basically
a
conflict
in
perturb
of
definitions.
So
it's
a
base
to
PR,
which
removes
the
conflicted
vendor
directories
and
fixed
it
so
as
it
involves
the
storage
driver.
So
probably
it
was
tagged
with
six
storage,
so
I
saw
this
assigned
as
there
were
maybe
over,
but
even
could
just
take
a
look
and
please
review
it.
O
O
B
B
N
E
Q
Q
G
C
C
A
G
We
decided
it
would
be
a
two-step
process
regardless,
so,
if
you
you
mount
it
as
a
block
and
if
someone
else
wants
to
go
do
something
with
it.
That
is,
but
we
we
covered
that
in
great
lengths,
I
think
in
the
second-to-last
design
review
call
if
that
was
the
best
way
to
handle
it,
because
it
became
too
complicated.
That's
why
we
also
removed
that
use
case.
It
had
had
trying
to
combine
the
two
in
there
because
you
realized
it
wasn't
possible
from
a
binding
perspective.
G
C
G
A
G
G
C
N
N
G
That's
a
good
point
though,
but
we
we
have
to
get
it
merged,
so
we
can
start
getting
the
other
pieces.
Large
that's
I
mean.
N
N
A
Okay,
awesome:
if
no
one
else
has
PRS,
we
can
move
on
to
design
reviews.
So
me
and
Dhiraj
put
out
a
bunch
of
design
reviews
related
to
local
storage.
There's
three
currently
out
the
first
one
is
just
the
design
that
we
did
for
one
seven
alpha
I
had
I
had
it
all
in
a
Google
Doc
I
finally
found
some
time
to
put
it
into
a
PR
and
merge
it.
B
A
B
C
A
A
There's
kind
of
two
parts
to
that.
First,
is
the
volume
plug-in
changes
to
support
the
new
volume
mode
block
I,
think
those
changes
are
not
really
complex
shouldn't
there
shouldn't
really
be
much
in
the
design.
For
that
the
more
interesting
part
is
the
the
in
the
local
storage
provision
or
static
provisioner
to
support
block
the
presenter
does
discovery
of
block
devices,
and
it
also
does
clean
up
a
block
devices,
so
we're
proposing
some
interesting
methods
to
for
the
cleanup
of
block
devices.
So
please
take
a
look
at
the
document
and
provide
some
feedback
there.
A
R
That
you
know
can
the
process
of
contributing
fixes
to
Kuban.
It
is
be
made
more
agile
and
here
is
a
problem.
Currently,
what
we
see
is
that
existing
in
so
let's
say,
I
have
a
fix
that
I
want
to
get
into
1.6.
So
I'm
sure
you
are
all
aware,
but
I
will
break
it
down.
So
we
poster
we
post
appear
for
review
review.
That's
done
that's
most
omastar,
then
we
do
actually
pick
requests
and
then
share.
That
is
the
goal
here
right.
R
So
what
we
see
right
now
is
that
every
stage
pretty
much
takes
5
to
8
days,
and
this
is-
and
we
have
been
contributing
to
given
it
is
for
a
year
now
and
I
have
specific
examples.
So
we
have
a
least
felt
out
order.
Refactoring
the
pretty
much
rewrote
the
whole
thing.
It
was
internally
completely
reviewed.
We
started
a
request
on
July
18th
but
merged
on
August
9,
which
is
still
significant
time.
20
20
plus
days,
we
have
a
terrific
equal
that
is
ongoing,
which
we
started
on
July
28th.
R
It
is
still
going
on
and
we
have
ship
it's
from
everybody,
and
you
know
it
is
failing
because
of
some
test
infrastructure
instability,
and
my
intention
here
is
to
figure
out
what
we
can
do
here
and-
and
my
last
point
here
is
that
you
know
why
should
we
care
about
this?
First
of
all,
I
feel
it
is
significant.
It
is
not
the
six
storeys
problem,
there
is
organizational
neutral.
It
is
not
being
there
that
had
Google
photos
for
diamonte
problem.
R
It
are
fixed
customers,
because
even
if
we
ask
customers
to
you
the
stable
release
of
given
it
is
the
they
we
need
to
figure
out
the
quicker
way
to
give
them
fixes.
And
my
last
comment
is
that
I
wonder
that
you
know,
even
if
you
know
to
CSI
or
out
of
three
cloud
providers,
it
still
won't
solve
the
problem
because
we
still
have
to
so.
R
N
Two
shards,
so
you're,
not
you're
kind
of
preaching
to
the
choir
I
think
we
all
agree.
There
is
a
bunch
of
efforts
going
on
in
kubernetes,
and
none
of
them
are
happening
quite
as
fast
as
I.
Think
as
we
want
to
address
this
issue,
you
know
the
first
one
is,
of
course
you
know
you
are
gonna,
hear
it.
You
already
anticipated
it.
You
said
it
wasn't
sufficient,
which
is
CSI
or
even
flex.
N
If
you
want
to
start
using
flex
today,
I
know
that
there
are
some
customers
in
the
users
that
you
know
being
able
to
launch
fixes,
underneath
your
own
control
and
in
your
own
release,
schedule
and
doing
less
in
core
is,
you
know,
is
one
way
out
of
this
predicament.
You
know
you
can
have
your
faith
in
your
own
hands
and
you
can,
you
know,
put
in
fixes
as
soon
as
you're
ready
and
get
them
in
production.
You
know
this
is
it's
obvious.
You
know
we
can't
do
everything
inside
one
big
codebase,
it's
a
monolith.
N
R
And
that
is
actually
pretty
much
my
team
is
currently
working
on
is
that
we
want
to
have
sufficient
tests.
Overt
aggression
from
other
components
might
break
our
code,
so
we
are
working
on,
but
but
to
answer
your
question,
the
fixes
that
we
ran
into
so
we
found
a
personal
problem
and
we
found
that
regression
was
caused
by
not
by
us,
but
something
else.
Yes,.
R
N
We
and
I
think
Aaron
has
been
sort
of
leading
the
charge
here.
We
definitely-
and
you
know,
if
you
went
to
the
the
leadership
summit
that
was
a
few
months
back-
there
is
widespread
agreement
among
kubernetes.
You
know
developers
in
general
that
there
needs
to
be
larger
investment
and
more
development
in
making
the
system
more
stable
and
being
able
to
catch
bugs
both
before
they're
checked
in
or
when
they
are
checked
in
I
guess,
I
have
to
to
sort
of
questions.
N
One
of
them
is
getting
bug
fixes
and
patch
releases
doesn't
seem
to
take
that
long.
For
my
takeaway,
you
know
the
fact
that
you're
saying
it
takes
over
a
month
or
so
is
the
problem
that
you're
not
able
to
get
someone's
attention
that
it's
a
bug
or
is
that
you
can't
get
someone's
attention
to
review
it?
Yeah.
R
N
You
know
the
reviews
or
the
functionality
that's
on
their
area,
but
just
to
be
experienced
across
the
whole
area,
and
on
top
of
that
you
know
the
the
I
really
do
feel
that
once
we
get
CSI
done
and
it's
not
so
much
that
the
plugins
will
be
bug
fixes
that
are
easier.
It
reduces
the
load
on
a
lot
of
the
people
reviewing
code
overall,
because
suddenly,
the
the
the
amount
of
code
landscape
in
the
amount
of
code
that
we
are
responsible
for
in
the
storage
sig
goes
down
dramatically.
N
We
can
actually
focus
more
on
the
core
than
on
all
of
the
drivers,
but
on
top
of
that
you
know
it
would
be
interesting
to
know
why
that
book
happened
and
then
how
we
can
prevent
it
from
happening
in
the
future
with
with
testing
at
the
same
time,
by
feeling
there's
two
things
going
on
here.
One
of
them
is
how
do
we
make
sure
it
was
this
above
that
was
affecting
other
folks,
are
just
the
VMware
environment.
N
R
B
R
N
Actually,
you
know
what
maybe
we
should
take
this
offline
and
talk
about
it.
There's
a
lot
of
things
that
you
can
do,
but
also
I,
don't
know
if
CN
CF
is
the
way
to
go
as
much
it's
like
what
we're
just
doing
in
kubernetes.
Overall
there's
a
lot
of
efforts
underway
right
now
to
try
to
accelerate
everything
from
how
long
it
takes
to
get
a
bug
fix
in
and
to
you
know,
grow
more
reviewers
and
on
all
of
this
I
think
your
frustration
is
one
that
is
shared
by
everybody.
N
You
know,
even
in
you
know,
I
think
Red,
Hat
and
Google.
I
think
that
we
want
more.
You
know
velocity
in
this
area
and
I
can
tell
you
even
in
you
know
these
I
used
those
two
groups
because
they've
been
contributing
and
have
a
lot
of
resources
for
a
while,
but
even
those
two
groups
you
know
they're
been
calls
for
I've
had
people
from
Red
Hat,
you
know
very
senior
people
go.
N
We
need
reviewers
for
this
feature,
we
have
a
bug,
and
you
know
that
this
is
not
something
that
is,
you
know,
is
just
being
hit
by,
like
maybe
your
organization
or
other
organizations.
This
is
a
a
widespread
kubernetes
problem
and
we,
you
know,
if
you
want
to
talk
about
what
we're
doing
now
and
then
brainstorm
or
ideas
or
sounds
awesome.
R
N
Let
me
say:
I
don't
want
to
be
dismissive,
I,
think
everything
you're
saying
is
something
that
everyone
is
feeling
and
you
know,
I
see
Eric
or
you
want
to
line
this.
Oh.
S
N
Actually,
it
wasn't
just
do,
there's
been
others,
but
I
mean
I.
Just
you
know,
I
wanna.
Can
you
comment
on
what
I'm
saying
you
know
I'm
curious
for
your
perspective
on
this?
Am
I,
in
line
with
your
review
of
things,
cuz
I
know,
you've
been
doing
this
for
a
while
longer
than
I
have
not
much
longer.
At
this
point.
S
But
no
I
I
mean
you,
you
absolutely
hit
on
them.
I
mean
it's
definitely
a
systemic
issue.
There's
definitely
some
places
that
are
better
or
worse
than
others.
Intention
of
those
who
can
make
a
difference
on
occasion,
glede
should
always
be
somebody
who
you
should
be
able
to
reach
out
to
if
you're
really
really
stuck
that,
but
it's
definitely
a
broad
problem,
one
that
I
know
the
architectural
review
and
the
steering
committee
are
both
going
to
continue
talking
about
in
the
future.
N
Everyone
here
to
do
in
until
we
have
enough
processes
in
order,
because
I
don't
think
we
really
do
like
one
thing
that
I
ballon
is
that
we
don't
really
have
a
great
system
now
to
analyze
all
of
the
issues
that
we
see
in
github
and
say:
oh
wait:
these
are
the
top
five
that
need
to
be
addressed
right
now.
It
still
is
a
little
bit
word-of-mouth
and
unfortunately
you
know,
if
that's
how
it
is
I,
don't
mind
at
least
myself
depending
on
word-of-mouth,
so
you
know
kind
of
everyone
in
the
group
in
the
past.
N
If
you
see
something,
especially
in
a
release
that
has
been
released
already,
that
you
feel
you
know,
that's
live
to
customers.
That
is
a
bug
that
you
think
is
highly
impactful
or
even
moderately
impactful
go
to
the
list.
Reach
out,
feel
free
to
reach
out
to
you
know
me
Jesse
sod,
at
least
on
Google
and
say
you
know.
How
do
we
draw
attention
to
this
and
you
know
the
real
answer
is
it
might
be,
but
guess
what
it's
not
as
big
as
the
other
ones,
we're
reviewing
right
now
or
the
answer
may
be.
Oh
man.
B
N
N
N
G
It's
kind
of
had
less
involvement.
To
be
perfectly
honest,
Michael
I
mean
we've
had
less
community
members,
be
interested
in
that
and
it's
been
pushed
more
to
the
PR
level,
and
maybe
it's
something
we
need
to
really
address
as
far
as
a
formalized
process
of
all
these
things
beyond
merging
PR,
so
yeah
I
appreciate
you
speaking
up
and
voicing
your
concern.
I
think
that's
important
to
keep
this
SIG
healthy.
Is
that
we're
not
brooding
about
it
and
not
contributing
this.
N
Sig,
for
me,
has
been
wonderful
in
that
we're
actually
trying
to
tackle
these
things.
You
know
shar.
One
thing
that
would
be
really
helpful
is
looking
at
the
areas
right
now
that
you
feel
where
things
are
falling
through
the
cracks,
especially
in
verification
and
sending
that
to
the
list,
and
let's
start
trying
to
tackle
this
again
I.
We
had
a
whole
bunch
of
these
in
1,
3
and
1/4
I
think
we
tackled
them.
N
I
I
keep
forgetting
that
Aaron's,
not
magic
because
she's,
so
you
know
being
able
to
make
so
many
magical
things
happen,
but
it
if
we
have
other
specifics,
I
think
that's
really
the
problem
that
we
don't
always
know
where
these
specific
issues
are,
and
once
we
have
a
list
of
what
to
tackle
this
sig
has
been
really
good
jumping
on
them.
So
please,
like
you,
know,
to
give
us
the
benefits
of
what
you're,
seeing
because
you
know
maybe
in
the
Google
GCP
environment,
we
don't
see
quite
as
many
of
them
as
you
do.
Yes,.
N
And
it's
not
just
about
saying:
oh,
go
read
some
tests,
it's
about.
You
know,
raising
the
awareness
and
incorporating
them
in
the
framework
that
we've
got
yes
sweet.
Thank
you
so
much
for
bringing
this
up
and
then.
Lastly,
I
do
think
that
we
need
to
maybe
I'm
kind
of
curious
if
the
idea
of
a
kind
of
a
reviewer
boot
camp
would
be
useful,
a
way
to
sort
of
scale
out
sod
and
Brad,
and
some
of
the
people
who
you
know
in
Jing
who
really
really
I,
know
there's
other
people
like
yawns.
N
You
know
I,
don't
even
have
the
full
list
anymore,
but
maybe
having
them
come
up
with
the
areas
of
knowledge
that
they
think
are
really
important
for
people
to
understand,
and
maybe
the
next
face
to
face
getting
everyone
together
and
kind
of
giving
an
intro
to
this
and
then
maybe
seeing
if
they're
people
who
want
to
start
reviewing
more
actively
as
a
way
to
try
to
grow
more
reviewers
over
time
and
I
have
to
go.
But
it's
just
a
thought.
Maybe
we
can
talk
more
about
it
in
the
next
meeting.
Okay,.
A
All
right,
so
the
last
item
we
have
on
the
agenda
is
our
face
to
face
its
candidates
are
confirmed
for
October,
10th
and
11th
location
is
not
exact
locations
not
quite
confirmed
yet,
but
it's
most
likely
going
to
be
in
Sunnyvale,
I.
Think
Brad.
You
started
a
sign-up
sheet
with
topics,
so
if
you
could
link
that
she
that'd
be
great
and
everyone
who
wants
to
attend-
please
add
your
name
to
the
sheets
and
also
suggest
topics
for.